This PDF file contains the front matter associated with SPIE Proceedings Volume 12978, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

Semantic feature extraction has become one of the hot research directions in remote sensing images, and the semantic segmentation technology of remote sensing images based on deep learning has achieved many results. However, with deepening of the network, semantic segmentation models based on deep learning have problems such as loss of detailed information and excessively increase of parameters, resulting in inaccurate extraction. To solve this problem, this paper devices a new type of Multi-Path Cyclic Residual Network. A Multi-Path Residual GCN is designed to enhances detailed features and reduces using of parameters. In addition, a Cyclic Residual Cross module is referenced to make the surrounding pixels aware of each other in the form of crossing. The model has achieved significant results on the published Postdam and Jiage datasets, and has also achieved significant improvements compared to other advanced models.

The manual interpretation of traditional side-scan sonar images has problems such as solid subjectivity, low efficiency, high labor cost, and large time consumption, and the machine learning method requires manual feature selection, which lacks adaptability and robustness. In this paper, we introduce a convolutional neural network method, which can automatically learn features from side-scan sonar submarine aircraft wreckage images and complete classification recognition. Using 48 side-scan sonar submarine aircraft images in the SeabedObjects (ship and aircraft) dataset after preprocessing 1686 images, the convolutional neural network model is trained and tested. The results show that the trained CNN model can accurately identify and classify the side-scan sonar submarine aircraft wreck images with an accuracy of 98.85%, which is highly efficient, accurate and robust, and can effectively improve the recognition and classification level of side-scan sonar submarine aircraft wreckage images.

In view of the disadvantages of the Douglas-Peucker algorithm applied in map generalization, an improvement algorithm based on the skeleton line has been put forward to replace the simplification index of vertical distance, which extracts the skeleton line from a convex vertex to baseline as the simplification index through the triangulated network. Based on the improved Douglas-Peucker algorithm, a new method of coastline generalization is provided. The main process is: on the basis of curve monotone section bending identification, the bending skeleton line is extracted through the construction of the triangle network of bending segments. By this new method experiment of coastline generalization is carried out according to the principle of“expanding land and shrinking sea”, the validity and feasibility of the method in maintaining the linear features of the coastline are verified.

With the development of GIS, more and more application systems have turned from 2D to 3D, and the construction of 3D virtual scene has gradually become a research hotspot. This paper proposes a technical route and method to construct the 3D virtual ocean scene based on Skyline, and discusses the potential publication way of the scene. Multi-source image data integration method is introduced to solve the problem of low efficiency of organization and management of massive image data. Meanwhile, virtual basic terrain model, virtual island and coastal zone model, virtual seabed surface model and virtual sea surface model are built and integrated to display the complete panoramic ocean sphere. The constructed 3D virtual ocean scene can be regarded as a carrier for marine information visualization and has been applied in many marine application systems.

To guarantee both efficiency and robustness of coordinate transformation, a 3D coordinate transformation model based on adaptive M-estimator is proposed. With the influence of the variance factor on the M-estimator in the coordinate transformation scene analyzed, the RW algorithm is improved to determine global optimal solution of variance factor, and the outliers are eliminated through the adaptive adjustment of the factor. Through the simulation experiment of a control network of mining area, it is indicated that the outliers are located by the adaptive model even under the high rate of outlier. In terms of accuracy, M-estimator is improved to reach or even exceed RANSAC algorithm by this method. As for efficiency, the performance of M-estimator is basically preserved, which is far better than RANSAC algorithm.

Lots of speckle noise exists in synthetic aperture radar (SAR) images inherently. The speckle noise causes various difficulties in the subsequent SAR image processing. In this paper, a novel Lee filtering algorithm based on an adaptive size sliding window is proposed. The adaptive sliding window makes the local statistics acquired from the image more accurate, which improves the performance of proposed algorithm on the edge protection and noise filtering. In this paper, a series of comparative experiments for speckle suppression is conducted on both optical simulation images and SAR images. The experimental results show that the performance of the proposed improved Lee filter to suppress speckle noise is better than the traditional Lee filtering algorithm while more details are preserved.

Curve features formed by irregular edges are common in real building images, and effective extraction and matching of curves in images is important for the recovery of real architectural scenes. However, existing curve extraction and matching algorithms are highly specific and difficult to be directly applied to 3D reconstruction of building images with viewpoint variations. Therefore, a full-flow algorithm for curve extraction and matching is proposed in this paper for building images with viewpoint differences. Firstly, the Canny edge detection optimisation algorithm is used to obtain the target contours, and the closed contours are collated to obtain reliable curves with definite endpoints. Then the curve pairs are constructed using the geometric relationships of the curve endpoints, multiple sets of simulated images are set to estimate the best affine transformation parameters, and the image pairs are pre-transformed to obtain corresponding intersections. Finally, the corresponding curves are determined according to the intersection matching results. Experiments using image pairs with different imaging conditions show that the proposed algorithm can overcome the image viewpoint variations and achieve extraction and matching of curves in building images, with good robustness.

Deep learning has achieved good results in recent times in the field of hyperspectral images (HSIs) classification. For traditional convolutional neural networks, the excessive increase in network depth will lead to overfitting and gradient disappearance. In addition, most of the methods for HSIs classification usually don’t consider the strongly complementarity and relevance between different levels or directions. In order to solve the two problems above, this paper proposes a spatial-spectral joint framework based on Gabor filtering and feature fusion network (GFFN) for HSIs classification. Firstly, Gabor filter can effectively extract spatial information including edge and texture in different directions. Moreover, Gabor filtering can effectively alleviate the overfitting problem, especially when the training sample set is small. Secondly, this paper extracts spectral features by combining convolutional neural network (CNN) and residual network (ResNet). This effectively solves the degradation problem of CNN. Thirdly, the proposed method integrates the feature outputs in different directions to further improve the classification accuracy. The experimental results obtained with three public HSI datasets show that the GFFN is superior to other competitive ones.

The severe soil erosion on the Loess Plateau is the root of the Yellow River problem, and the mathematical model of soil erosion is the core of intelligent soil and water conservation. How to extract soil and water conservation measures accurately, quickly, and quantitatively is of great significance for developing the parameterization method of soil and water conservation measures in the soil erosion model and quantitatively revealing the influence of soil and water conservation measures allocation on runoff and sediment yield process. It is the direct demand to support the research and development of distributed soil erosion models, soil erosion prediction, and digital twin watershed construction. Based on analyzing the current high-resolution remote sensing image data and deeply learning the development frontier of artificial intelligence, this paper studies and explores the technology and method of intelligent extraction of terrace measures underpinned by an artificial neural network algorithm model under deep learning. The main research results are as follows: (1) The current deep learning theory development and model structure are deeply studied. (2) The intelligent extraction algorithm of high-resolution remote sensing images based on depth learning is developed. (3) Accuracy analysis of the model results are carried out. The research results of the project will provide basic data for regional soil erosion prevention and control, and at the same time, provide technical support for high-precision parameter acquisition and calibration of distributed soil erosion model, which has important theoretical practice and application value.

Abyssal topographic information is critical for abyssal scientific research. The ship-borne multi-beam system efficiently measures deep-water seafloor but has low precision and resolution. By contrast, a muti-beam system mounted on AUV can obtain precise depth data and effectively supplement ship-borne multi-beam data. When combining datasets from these two platforms, improved spatiotemporal coverage can characterize the abyssal seafloor change. However, the AUV platform fixes its position using acoustic methods, which introduces a significant bias in its positioning reference. Consequently, the depth variation can be misinterpreted when combining depth data from ship-borne and AUV platforms. In this paper, the robust terrain matching method is used to register the data of different platforms. The method was tested on the Juan de Fuca Ridge datasets. Computational results demonstrate that the proposed method eliminates the positioning reference bias effectively and thus improves the accuracy of the interpretation of the abyssal seafloor change.

Given the difference between the external antenna and built-in antenna, taking Zhonghaida GNSS geodesic receiver, Honor 60 and Huawei MatePad Pro as examples, 10 hours of continuous observation data were collected on campus to verify the availability of smart tablets and smart phones in Global Navigation Satellite System Interferometric Reflection (GNSS-IR) surface height monitoring. Experiments were conducted with the elevation angle intervals of 5°-20°, 5°-25° and 5°-30°, respectively, to determine the best elevation angle interval in surface height monitoring. The experimental results showed that the optimum altitude angle range of the GNSS receiver, Honor 60 and Huawei MatePad Pro is 5°-30°. Considering the results of RMSE, this study believes that the monitoring accuracy of different SNR types is G-S1C>R-S1C>C-S2I. Moreover, the surface height monitoring effect of the Zhonghaida geodesic GNSS receiver is relatively good, and the RMSE value in the experiment of 5°-30° elevation angle range is kept within 10 cm, while that of Honor 60 and Huawei MatePad Pro is within 20 cm. It can be seen from their median values that the deviation between the retrieved surface height and the measured value of the three devices is kept between-5 cm-5 cm, thus verifying the availability of smart tablets and smart phones in GNSS-IR surface height monitoring. In a word, this study gives an example to demonstrate the application of low-cost smart equipment in GNSS-IR technology. Furthermore, it lays a foundation for its application in snow depth retrieval, sea level height monitoring and soil moisture detection in the future.

The temporal and spatial variation of precipitable water (PWV) has an important effect on the forecast of short-time severe weather such as rainstorm and hail. In this paper, cloud mask is used to remove cloudy data, and then inverse distance weighting (D-IDW) algorithm of DEM with elevation information is used to process study area 1(Visalia, California, USA), Study area 2(Los Angeles, California, USA) and study Area 3(Bishkek, Kyrgyzstan). The problem that MODIS cannot be used in cloud areas is solved. Finally, the improved fusion model method is used to fuse MODIS data with ERA5 data to obtain PWV with high spatial resolution and high precision. The fusion of PWV image and ERA5-PWV image has a good similarity in the spatial distribution changes of study area 1 and study area 3, which have a large elevation fluctuation (elevation change is greater than 2 km). The Global Navigation Satellite System (GNSS)-PWV was used to verify the accuracy of the fusion data. The minimum deviation was found in December in the study area 1, with an average deviation of 1.79 mm. Study area 2 was the smallest in December, with a mean deviation of 1.91mm. Study area 3 had the smallest deviation in December, with a mean deviation of 1.52mm.

Road information is an important part of remote sensing information. The accurate and efficient extraction of road information from high-resolution remote sensing images is closely related to human life. Deep residual Unet has certain achievements in road extraction, but there is still room for improvement in the integrity of road extraction results. In order to improve the segmentation completeness, robustness, and anti-occlusion ability of the model, this paper proposes an improved dual decoder structure based on the deep residual Unet (ResUnet) to solve the problem of incomplete road extraction in the deep residual Unet. High-resolution remote sensing image road information extraction method. In the original model, the bottom up-sampling structure was first added, which retained features to facilitate segmentation and expand the receptive field. Then add spatial channel compression and activation modules to correct the input features and enhance meaningful features. Experiments on the Massachusetts road data set show that the improved method has improved evaluation indicators compared with the original network. Precision, Recall, F1-score and Dice coefficients have reached 91.84% and 87.27%, respectively. 89.55%, 88.89%, and has certain advantages in anti-interference and extraction integrity.

Classification and identification of the materials the earth’s surface have long been a fundamental but challenging research topic in geoscience and remote sensing (RS). Although deep learning technology has achieved certain results in remote sensing image classification, it still has certain challenges for multi-modality remote sensing data classification. In this work, we propose a deep learning multi-modality remote sensing image classification network based on pixel-level cross fusion, called DMLCF-Net. In other words, DMLCF-Net provides a unified deep learning multi-modality remote sensing image classification framework, and introduces the strategy of cross fusion to classify multi-modality remote sensing images. The cross-fusion module can obtain compact feature representation from multi-modality remote sensing data, and different modality can exchange information with each other effectively. In addition, to validate the proposed scheme, extensive experiments conducted on multi-modality remote sensing dataset, demonstrate the effectiveness and superiority of the proposed DMLCF-Net in comparison with several state-of-the-art multi-modality remote sensing data classification methods.

As urbanization and the rapid development of remote sensing technology, the interpretation of urban roads and intersections using high-resolution remote sensing images has become the focus of studying urban planning and road network structure. However, the current research is only limited to extracting road intersections, and few studies have constructed their topological relationships and analyzed the spatial distribution of road networks. Therefore, in this study, the intersection targets are extracted based on YOLOv5s model, and then the topological relationship of intersections is constructed by Delaunay Triangulation. The accuracy of the optimal model is 0.9415 for mAP, 0.8985 for F1, 0.8611 for Recall, and 0.9394 for Precision. By analyzing their characteristics such as area, side length, PESL, AESL, ESLR and interior angle, it can be found that the density of road intersections in the study area is high, the spacing of adjacent intersections of most roads is relatively similar, and most of the roads intersect vertically. This study first extracts road intersections from remote sensing images and then analyzes their topological relationships to provide research ideas for large-scale urban road network analysis and multi-source road data matching.

When using radius filters for point cloud filtering of multibeam point clouds, problems such as poor adaptability, difficulty in determining the threshold value, and easy filtering of edge terrain points occur. To address these problems, this paper proposes a multibeam point cloud denoising method that takes into account the edge effect of multibeam bathymetry data . The radius filter is combined with density clustering, and the radius filter is used to filter out the edge noise, while the density clustering is used to recall the mistakenly deleted terrain points. The experimental results show that the proposed method is able to filter edge noise while preserving the edge terrain well. The work in this paper provides a new solution for the automated processing of multibeam point cloud data, which is relevant to the development of multibeam data processing.

Simulation of multispectral broadband data from hyperspectral profiles is applicable for many purposes, mainly including sensor assessment and cross-calibration. For the simulation process, several aggregation methods have been proposed. In this paper, three aggregation methods were implemented, to investigate the impacts associated with method on the simulation of multispectral broadband. Comparative analyses were conducted in terms of channel reflectance and several derived vegetation indices, while mainly taking the wide field of view (WFV) sensor onboard Gaofen-6 for case study. Furthermore, impacts on the between-sensor comparability with the Landsat 8 Operational Land Imager (OLI) were discussed. Differences among aggregation methods in reflectance of the Gaofen-6 WFV varied along with channels, and obvious differences were observed in channels over near-infrared (770–890 nm) and red-edge (690–730 nm). Between-sensor difference relative to the Landsat 8 OLI was observed mainly over near-infrared, while the variations among aggregation methods were about 0.4%-2.0%.

Recently, transformer self-attention mechanisms have had significant advantages in the deep learning field, and it has been extensively used for natural language processing and video tracking. Furthermore, self-attention mechanisms have also been applied in hyperspectral unmixing. Although self-attention mechanisms are usually efficient and flexible tools, the original transformer might break the inner structure of data during learning, causing negative effects to unmixing. In this work, we employ transformer self-attention mechanisms to achieve a deep self-embedded transformer network(DSET-Net) for hyperspectral unmixing. The proposed DSET-Net adopts an autoencoder framework and achieves local and overall feature parameter sharing in the encoder through a 'Transformer in Transformer (TNT)' strategy. The DSET-Net preserves the spatial details of hyperspectral images and involves only one convolution operation in the encoder, substantially improving the learning performance. The effectiveness of the proposed method is evaluated by using real hyperspectral datasets. Our experimental results indicate that the newly proposed DSET-Net is very competitive compared with other state-of-the-art approaches.

Ground electromagnetic method is an effective geophysical method to explore the water abundance of coal mine aquifer. The multi-solution of geophysical prospecting method greatly increases the difficulty of data interpretation, which makes it difficult to effectively detect the hidden danger of coal mine water disaster by single method. In order to realize the fine exploration of water abundance in coal mine aquifer, the one-dimensional joint inversion objective function of transient electromagnetic method (TEM) and controlled source audio magnetotelluric method (CSAMT) is established, and the joint inversion of the two methods is studied by using Occam algorithm. On the basis of TEM, CSAMT and joint inversion of the data processing by different methods, combined with known geological data, the inversion results of each method are compared and analyzed. It is verified by underground drilling that the stable water inflow of underground coal seam roof drainage borehole is about 80m³/h, and it is found that the joint inversion results are most consistent with the actual situation. The research shows that the joint inversion method can effectively reduce the multi-solution interference of electromagnetic inversion, greatly improve the accuracy of ground electromagnetic method to detect the water-richness of rock strata, and provide safe geological guarantee for coal mine production.

As a controllable, lowcost and repeatable test method, reverberation simulation plays an important role in developing sonar device. With the development of sonar systems such as multi-beam bathymetry systems, the requirements for reverberation simulation are getting higher and higher. In this paper, the unit scattering model is used for modeling, and parameters such as the propagation loss, sound absorption coefficient, bottom reflection loss and seabed roughness are considered.The seafloor reverberation simulation model is established by using the three-parameter model and the semiempirical formula of sound absorption coefficient. The output of the simulation is in concodence with the statistical properties of a large number of seabed reverberation secondary beam data. And in comparasion with real reverberation signal, the effectivemess of this model is verified.

This article introduces three-dimensional laser scanning technology for monitoring in the application field of mountain tunnels, and proposes a full cycle deformation point cloud processing method ICP algorithm for mountain tunnels. Firstly, the tunnel point cloud is cut into a single circular point cloud. After denoising the point cloud, the ICP algorithm is used to accurately register the two phase point clouds. After registration, the corresponding original point cloud and target point cloud with the closest and farthest Euclidean distance are calculated, which are the maximum deformation points. At the same time, after registration, the deformation settlement at the top of the tunnel can be obtained by calculating the highest point deformation. Leica Scanstation C10 is used to obtain the data of Qinghe project mountain tunnel, which proves that the method proposed in this paper has practical significance in tunnel section Deformation monitoring.

The positioning error of the navigational underwater acoustic positioning method is mainly limited by the GNSS measurement error, transducer position deviation, time delay difference and sound velocity error. In view of these problems, this paper uses the method of inter-epochal difference to weaken or eliminate the systematic error, and compares and analyzes the positioning accuracy of the underwater non-differential model and the underwater differential model through the navigation underwater positioning experiment. The experimental results show that the accuracy of the underwater differential positioning algorithm is higher than that of the non-differential underwater positioning algorithm in the horizontal and vertical directions, which can reduce the proportion of systematic error in positioning error to a certain extent.

In this paper, we study the sum rate maximization problem involved in an unmanned aerial vehicle (UAV)-assisted maritime wireless network. Explicitly, the quadratic transform is first applied to transform the sum rate maximization problem into a tractable one by introducing an auxiliary variable. However, the recast objective function is nondifferentiable at some points, making it hard to solve especially when using gradient based methods. In order to tackle this problem, we propose a projected steepest descent algorithm, where smoothing and normalization methods are also used. Finally, simulation results are provided for characterizing that the proposed algorithm is capable of providing an improved sum rate performance for the UAV-assisted maritime wireless network.

The accurate acquisition of high-precision fixed-point positioning information in complex terrain scenarios such as mountain slopes is of paramount importance in the field of surveying and geographic information, particularly for landslide detection and terrain monitoring. The existing methods are plagued such as high cost of data acquisition and complex data processing. To address this problem, this study proposes an approach that leverages a UAV (unmanned aerial vehicles) landing platform to facilitate the precision landing of a UAV at observation points, with the ultimate goal of obtaining high-precision positioning information at that points. The proposed method offers significant advantages such as low-cost, simplified, and robust. We developed a low-cost RTK (real-time kinematic) positioning module that can be easily mounted on a UAV platform, and designed a landing platform to assist in the precision landing of the UAV. The positioning information was obtained using the carrier phase differential algorithm. The results demonstrate that the proposed method can achieve planar positioning accuracy at the centimeter level. The proposed approach offers a simpler data processing method, and holds significant potential for a wide range of applications in the field of surveying and geographic information.

One-bit compressed sensing (CS) synthetic aperture radar (SAR) imaging greatly reduces the hardware cost of SAR systems and has great technical advantages, but one-bit CS SAR imaging algorithms face the problem of poor performance at a low signal-to-noise ratio (SNR). This paper applies the deep unfolding idea to the binary iterative hard thresholding (BIHT) algorithm and introduces the threshold parameter to propose the DeepBIHT-net network. Experiments are conducted to verify that the proposed method can effectively improve the reconstruction performance at low SNR.

Precise precipitation data is essential for effective hydrologic modeling, water resource management, and climate research. IMERG and GSMaP, which are high-resolution and wide-coverage satellite products, have great potential in monitoring precipitation. However, further research is required to understand the error characteristics of precipitation. This study focuses on the Yellow River Basin, results demonstrate that: (1) GPM IMERG and GSMaP display strong spatial patterns, with more precise detection in the east than the west. Hit bias dominates in the total bias, and each product overestimates the error. The GSMaP_Gauge product performs better in reducing hit bias and improving detection accuracy. (2) The Yellow River Basin is generally biased towards low-intensity precipitation events, with the GSMaP_Gauge product performing the best, followed by the IMERG_Final product with the smallest missed precipitation. (3) Altitude has a significant impact on the total bias and error components, mainly showing a negative correlation. As altitude increases, the total bias and error components gradually decrease. This study can facilitate the creation of quantitative bias models and the implementation of techniques to reduce bias in GPM IMERG and GSMaP by examining the structure and attributes of error components.

Feature extraction is one of the key technologies for aerial target attitude estimation. Traditional methods rely on manual feature extraction, which only uses shallow structural features, thus limiting the accuracy of target attitude estimation. In this paper, for the first time, a deep learning method is adopted to conduct research on aerial target attitude estimation based on radar cross section (RCS). Considering the small dimension of RCS data and the complex mapping relationship between RCS and attitude, one-dimensional Cut-ResNet50 with 6 Bottleneck is proposed by pruning ResNet50, which simplifies the model while increasing the generalization performance of the model. The loss function is modified by combining cross-entropy and mean square error to enhance the learning ability of the model. Finally, attitude recognition testing results on simulated RCS dataset and comparison with various models validate the effectiveness of the proposed method.

Using Gaofen-5 hyperspectral data and vegetation index analysis method, the leaf area index inversion model of the study area was established, the results show that: the correlation coefficient between enhanced vegetation index and leaf area index reached 0.7334, the linear regression inversion model of growing season in the study area was established, and the root mean square error was 0.1807; spatial analysis results show, In the southwest part of the study area, the leaf area index was larger and the vegetation growth was flourishing, while the leaf area index was smaller in the southeast and central part of the study area, and the vegetation growth was normal. This conclusion was consistent with the vegetation coverage.

In recent years, the laying of urban underground cables has become more and more extensive, and the external breaking accidents of underground cables caused by construction have occurred frequently, seriously endangering the safety of power grids and people's property. How to apply the advantages of satellite and change detection technology to the transmission corridor patrol inspection with high efficiency, simplicity and practical needs needs further research. In this paper, the realization of remote sensing technology to quickly identify the signs of broken transmission lines in the transmission corridor is studied. The idea of combining pixel level and object level methods is adopted, and the difference principal component analysis method is used to compress and enhance the multi band image information, and the main change information along the cable is compressed into the third principal component; On the third principal component of the difference value, the OTSU adaptive method is used to complete the initial recognition of the external broken signs; Finally, combined with the object-oriented constraints to eliminate salt and pepper phenomenon, the final recognition results are obtained. Through comparative experiments and accuracy verification, the overall accuracy of the quick identification result of external force damage signs along the cable is 0.96, and the identified boundary range is consistent with the actual external damage sign area.

China has developed two miniaturized synthetic aperture radar (SAR) satellites called HiSea-1 and Chaohu-1. These satellites, which were launched in December 2020 and February 2022, respectively, can obtain high-resolution sea surface wind fields in day and night conditions. In a recent study, the Cross-Calibrated Multi-Platform (CCMP) wind vector, SAR look direction, and SAR incidence angle were used as inputs into CMOD5.N to output an expected σ₀. The wind field was then retrieved from a matrix determined by SAR-measured σ₀ and the expected σ₀. This method was applied to 90 HiSea-1 and Chaohu-1 SAR images. The retrieved wind speeds were compared with the ECMWF Reanalysis v5 (ERA5) reanalysis data. The root-mean-square error (RMSE) and mean bias error (MBE) were found to be 0.98 m/s and -0.02 m/s, respectively.

At present, the monitoring of debris flow in China is mainly oriented to the middle and low altitude areas, and the geological disaster monitoring and warning network mainly covers the risk points of geological disasters in the middle and low altitude areas. For the high cold and high altitude areas, the monitoring is blind at present, but once the eruption of debris flow in the high cold mountain area, it will have a huge impact on the downstream debris flow. Based on 4G+LoRa multi-network fusion communication technology, this paper studies how to realize long-term stable monitoring and early warning of debris flow gully in the alpine mountains by means of the cooperative working mode of rain gauge and camera and by means of self-adaptation to the weather.

Lots of studies on large scales maize estimation was used remote sensing data at home and abroad, nevertheless, the yields estimated accuracy were poor in mountainous region due to complex surface-information such as cultivated land topography and land fragmentation. In order to improved the accuracy in mountainous region, taking spring maize in Binzhou City Weibei dryland area, Shaanxi Province as a case, we used geographic conditions data and remote sensing data, explored a set of maize yields estimation method. In this study, the classification scale of cultivated land plots was optimized by using land use data. Based on the MOD09GQ products from 2001 to 2014, MODIS NDVI ten-day synthetic cloudless time series products were reconstructed. The ten-day synthetic data of time series NDVI, ten-day temperature, ten-day cumulative precipitation and yields statistical data were analyzed by canonical method, multiple linear regression model construction and leave-one-out cross validation. The results showed that the estimated yields of the model were close to the measured yields, and the error was less, indicating that the plot data weakened the interference of same spectrum foreign matter in the process of yields-growing area identification and extraction. The yields remote sensing estimation model optimized by the plot had a certain reliability and could be used for actual yields estimation. The yields of spring maize in Binzhou were closely related to NDVI in late July, temperature and precipitation in late June, and the influence of temperature and precipitation on the yields of spring maize had an obvious lag effect. This method of maize remote sensing yields estimation model could be used for reference in the study of remote sensing yields estimation of wheat, soybean and other yields in a mountainous region in the future.

Mapping, monitoring and parameter inversion in the mountain areas covered with dense vegetation as well as cloud and mist is always a complicated subject. Synthetic aperture radar (SAR) is an effective sensor of great potential to solve this problem. In this paper, a high resolution P-band airborne SAR system is developed, which is the first quantitative airborne P-band SAR system for business operation in China. The inherent signal ambiguity phenomenon for wide beam radar system is discussed, and the solution is suggested and validated with this radar system. Then a flight campaign is launched with this P-band SAR system in the Southwest China. The defocusing introduced by the dramatic terrain undulations is analysed detailedly. By co-registering the external digital elevation model (DEM) to the SAR image and motion compensation, the refocused quality of the polarimetric SAR images satisfies the user’s requirement. After polarimetric and radiometric calibration, the polarized composite and geocoded image products are obtained for further studies. With this P-band airborne SAR system, the repeat-pass interferometric data are expected to be collected in the future for the vegetated mountainous area, and it will play more important role in the fields of topographic mapping and deformation monitoring.

In order to monitor the surface deformation of landslides on the Loess Plateau, the loess landslides were taken as the research area in Linfen section of Fenhe River Basin. 25 SAR images covering the experimental area were collected, and the time-series deformation information of the landslide from 2020 to 2021 was extracted by SBAS-InSAR technology. The comparison with GNSS monitoring data shows that the time-series InSAR technology can effectively extract deformation information of loess landslide, which is in good agreement with the GNSS measured results.

The stability and safety of a concrete dam are important for hydropower projects. Before analyzing the stability of the dam, it is crucial to collect monitoring data like dam stress, uplifting pressure and dam displacement. In Taipingwan hydropower dam, the vacuum laser collimation system has been applied to monitor the vertical and horizontal displacement on the dam top respectively. According to the displacement monitoring results collected from last 10 years, several conclusions are drawn: there have been no abnormal deformations on the dam body. The long series of measured results show that the vertical and horizontal displacement of the dam changes regularly and periodically. In general, the vertical displacement change of the overflow dam section was basically synchronized with the air temperature. The right bank of the 0# dam section is connected to the mountain, and the displacement change is very small with the annual variation for several years less than 0.7mm. The suggestion for the fifth regular safety assessment is that detailed numerical simulation might be needed to find how the high uplifting pressure influencing the dam stability.

Global Navigation Satellite System (GNSS) and low-cost Inertial Navigation System (INS) is commonly used in fields such as vehicle navigation. However, when the carrier is in a GNSS rejection environment, the navigation accuracy rapidly decreases and cannot meet the positioning accuracy requirements. Therefore, this paper uses the Autoregressive Integrated Moving Average model (ARIMA) to learn the positioning information of GNSS during normal operation, output predictive values in the case of GNSS unavailable, and fuse them with the MEMS-INS mechanical arrangement results through Kalman filtering. Vehicle experiments show that the proposed algorithm significantly improves the navigation performance compared to traditional GNSS/INS integrated navigation, reducing the maximum position error in the east and north directions by 76.4% and 69.2%, respectively, and increasing the root mean square error of position accuracy in the east, north, and up directions by an order of magnitude.

The probability of the extreme true errors on each survey point excess the bound of the true errors and the variation of the root-mean-square errors with the sample size of the detection, are analyzed by the comparison of the detected value and the approximate true value. The reliability of the evaluation of the detection precision with the index of the inner conformance precision in the positioning of the open-cut cable ducts by the inertial positioner of 3D trajectory is studied. It is found that increasing the numbers of data set participating in calculation is beneficial to enhancing the detection precision, when the data satisfy the requirement of the inner conformance precision index. However, the enhancement is limited and not much, for the open-cut cable ducts. The means of 1-set detected coordinates can be used as the detection results for the common open-cut cable ducts.

The Zagros fold-thrust belt (ZFTB) in Iran is the result of the closure of the Neo-Tethys Ocean and the convergence of the Arabian and Eurasian continents. The influence of earthquakes above Mw 6 triggered by fault activity in this area on the surrounding salt diapirs has always been one of the hot issues in the field of geosciences. The slip factor of coseismic slip distribution of Sayeh Khosh Mw 6.1 on July 1,2022 in southern Zagros, Iran, is re-estimated and applied to the linear inversion of fault slip distribution of these two seismic events. Finally, the CFF stress transfer caused by two sources is calculated, and the results show that the MFF fault is thrust and right-lateral strike-slip. The subduction power source comes from the continental collision between the Arab plate and the Eurasian plate, and the salt rock detachment mechanism is formed.

The wildfires that have raged across Australia for over three months have left an indelible impact on every state. In response, we have developed a multi-objective planning model and a fire monitoring model that are suitable for SSA (sub-Saharan Africa) unmanned aerial vehicles. These models provide assurance for fire prevention and timely understanding of fire front issues while also ensuring cost control.

The reliability of SMAP salinity data in the Yangtze River estuary and the adjacent sea was verified using in-situ observations in spring, summer, and autumn from 2019 to 2021.SMAP data observation of low-salinity water mass detachment process. Then the SCHISM model was used to reconstruct the diluted water detachment process of the Yangtze River. The results showed that the reliability of the corrected SMAP data had been improved. The detachment process of the low-salted Yangtze River diluted water mass simulated by SCHISM is the same as that was observed by the SMAP satellite, which shows the reliability of SMAP data at the Yangtze River Estuary and adjacent seas.

In view of the problems existing in the fire prevention and rescue process, such as low efficiency and accuracy of fire inspection, difficult rescue operation, long operation time of traditional equipment, increasingly complex fire environment and high difficulty in fire area investigation, the research status of key technologies of fire fighting UAV at home and abroad is summarized from three aspects: flight control system, obstacle avoidance technology and fire monitoring and analysis technology. The development trend of firefighting UAVs is predicted as follows: (1) The stable flight control optimization of uAVs in complex fire environment: the size, shape, topography, topology and other basic factors affecting the stability of uAVs are optimized, and the matching of flight control system with the moment of inertia, dynamic redundancy and power system vector control technology of multi-rotor uAVs is improved; (2) Continuous improvement of endurance: through high energy laser and intelligent signal transmission and processing technology, realize the automation of remote energy supply, combined with intelligent visual tracking algorithm, real-time guarantee of UAV signal and laser undertaking; (3) Intelligent obstacle avoidance: multiple methods are combined to solve the obstacle avoidance problem of UAV under different circumstances. In addition, the scene model is established based on digital twin to obtain the best obstacle avoidance route, and the algorithm in the airborne computer is used to obtain the optimal path. (4) Cluster compatibility: The single firefighting UAV cannot adapt to the increasingly complex fire environment. The development of UAV swarm intelligence technology is helpful for the UAV cluster to carry the distributed load to complete the multi-point measurement task that the single drone cannot complete, and provide more accurate measurement data.

In this study, the canopy hyperspectral reflectance of Southern Corn Rust (SCR) was collected by ground spectrometer, and the canopy spectral characteristics of SCR, the correlation between canopy spectral reflectance or vegetation index and disease index (DI) were analyzed, then a hyperspectral remote sensing monitoring model of SCR was constructed. The results showed that: (1) There was a reflection peak at 550 nm after maize was infected by SCR, and an absorption valley near 970 nm. (2) The DI of SCR was significantly negatively correlated with canopy spectral reflectance at 350~720 nm (P<0.05), and the maximum negative correlation was near 680 nm (P<0.05), | r | was 0.47~0.76. (3) The hyperspectral monitoring model of SCR was established based on different vegetation indexes. The results showed that the model constructed by NDVI705 had better monitoring accuracy and verification accuracy. The R² of the monitoring model was 0.651, RMSE was 0.276, and the relative error between the predicted value and the measured value was 12.13%, R² was 0.679, RMSE was 0.288.

In recent years, remote sensing data has been increasingly applied to land use change with its advantages. In this paper, the dynamic land use of Xichang city over ten years are analysed based on Landsat 8 data. After the pre-processing, there are five land use types are given using the supervision classification. The results show that from 2013 to 2021, the urbanization of Xichang City develop significantly with the construction land 376.73685km² increasing to 480.152025 km². At the same time, the forest land area also increased by 29.44575 km², and the area of urban green space, bare land and water body decreased significantly.

In order to solve the problems of traditional shipping statistics, this paper puts forward the method of shipping statistics based on AIS big data, and gives complete technical process and technical scheme including big data platform construction, data access, data cleaning, data warehouse construction, navigation event analysis, voyage number mining and generation of statistical index and systematically applies the above processes and schemes. Firstly, based on the distributed storage and computing framework, the four-layer system architecture of data acquisition and processing layer, data storage layer, data analysis layer and service encapsulation layer is used to put forward the construction technology method of AIS big data platform, which meets the requirements of real-time AIS stream data acquisition, historical AIS file extraction, data cleaning, data storage management, data mining analysis, statistical analysis, visual display and application service; Secondly, AIS big data mining analysis model including electronic fence analysis, berth supplement, berth commodity identification, navigation event analysis, voyage number mining, voyage cargo capacity calculation and other model algorithms is constructed and integrated into the big data platform to meet the needs of AIS big data mining analysis and statistical analysis; Finally, taking the application of AIS data of global bulk cargo ships from March to May of 2019 as an example, it shows that this method can provide statistical indicators of shipping big data in three aspects including port, sea passage and bulk cargo, and provides a new idea for realizing real-time, accurate and refined statistical analysis and display of shipping.

Remote sensing building change detection is a key technology for understanding urban land cover and land use. In order to improve the accuracy of building change detection, this paper firstly summarizes a set of optimized building change detection methods for high resolution remote sensing images, and then introduces the steps of the method and the optimization part in detail. Finally, the feasibility of the optimized change detection method was verified in the experimental area. The results of this method are verified by an empirical experiment on the urban sub-center of Tongzhou District. The results show that this method can complete the remote sensing building change detection task under the condition of setting fewer parameters and large change area.

Aiming at the dynamic perception problem of abnormal marine targets, this paper studies the abnormal perception model of marine targets based on big data. Firstly, the classification status of abnormal behaviors of marine targets is analyzed, and a three-way classification method of "identity-status-behavior" is proposed for marine targets. Secondly, the characteristics of abnormal identity, abnormal status and abnormal behavior of marine targets are studied. Based on the knowledge-driven method, the abnormal discrimination rules of marine targets are defined, and the rule-based anomaly detection algorithm of marine targets is designed. Finally, the anomaly perception model of marine targets based on the Flink is constructed to realize the identification of marine targets with abnormal identity, status and behavior. The model test was carried out using the real-time data of radar and AIS. The results show that, under the premise of stable and continuous data, the constructed model can quickly identify abnormal targets conforming to the predefined rules, support the real-time calculation of big data and parameter configuration, and have good real-time performance and flexibility.

Studying the spatial and temporal changes of NEP in the Qinba area can provide scientific and technological data support for the realization of regional carbon neutrality goals. We use site parameters, geographical conditions monitoring data, MCD43A3 and a series of daily meteorological grid data to simulate daily NEP dataset at a spatial resolution of 500 m×500 m during 2000-2020. The results show that the spatial distribution of NEP in the ecosystem of the Qinba area is significantly different. Except for 2005, the vegetation in Qinba area is dominated by carbon absorption from 2000 to 2020. From the perspective of spatial change, the area of NEP significantly increased is 2657.37 km², mainly distributed in the northern Qinling Mountain area above 2000 meters of elevation and the southern part of Bashan area. From the perspective of seasonal changes, the proportion of carbon sink vegetation area in spring, summer and autumn is 90.3%, 99.1% and 91% respectively. Under the RCP4.5 scenario model, the NEP in the Qinba area is predicted to reach 33.41 gCm^-2a^-1 in the next 80 years. The meteorological factors that play a positive role in the increase of NEP are CO₂ concentration, temperature and precipitation.

Building segmentation from high spatial resolution remote sensing images is still a great challenge in the field of remote sensing image processing. This paper proposes an end-to-end deep convolutional neural network method for building semantic segmentation using high resolution remote sensing images. In this method, a multi-scale edge enhancement module(MEEM) with the Laplacian operator as the core is embedded to extract the building edge, which is used as an auxiliary method of semantic information to improve the segmentation accuracy. Based on the experimental results of the WHU Aerial imagery dataset, it is shown that the proposed method can not only achieve the same performance as some of the best building segmentation methods but also solve the problem that small buildings are difficult to be correctly detected.

Fractured reservoirs have become the main research object of oil and gas exploration and development. It has become a practical research object to establish a rock physical model for anisotropy caused by fractures. Combined with anisotropic SCA model and Eshelby-Cheng model, a rock physical model is established to predict the velocity of the longitudinal and transverse waves. The physical property parameters of minerals are obtained by logging data, and the elastic modulus of rock matrix is calculated based on V-R-H model. Then the elastic modulus of dry rock skeleton is calculated by the anisotropic SCA model. Finally, Eshelby-Cheng model is used to obtain the elastic modulus of saturated rock and predict the velocity of rock wave. The comparison and analysis of the obtained P and S wave velocities with the actual data shows that The longitudinal and shear wave velocities of saturated rock calculated by the rock physical model are in good agreement with the actual data, which conforms to the actual geological conditions, and proves that the method is reliable and effective.

Based on the analysis of the formation principle and characteristics of the flooded layer, the geophysical response characteristics of the flooded layer are combined with the seismic inversion to predict the distribution range of different flooded layers using the curve reconstruction seismic identification method based on the MCMC algorithm, which achieves the purpose of seismic identification of the flooded layer. The results of the combined post-test well analysis show that the prediction meets the accuracy of 82.4%. The curve reconstruction seismic identification method based on MCMC algorithm can quickly and accurately locate the distribution range of different water flood layers, which provides a new idea for oilfield reservoir prediction scheme.

The direction relationship between spatial entities is an important content of spatial relations, and is the basis of more knowledge mining based on spatial data. In this paper, a vector spatial data direction relationship expression model based on azimuth is proposed. When the reference object and the source object are point spatial entities, line spatial entities and area spatial entities respectively, the direction relationship expression model is established. According to the need of natural cognitive language, the quantitative expression model and qualitative expression are transformed. We designed typical sample data, calculated each case in the model, and the results prove the feasibility of the model finally.

Aiming at remote sensing image domain adaptation segmentation model challenges, including different feature distributions, different scales, and unequal distribution of categories, a remote sensing image domain adaptation segmentation model combining scale discriminator and attention was proposed. Firstly, for the different remote sensing image resolutions in different domains problem, in an adversarial-based discrimination network, in addition to the common feature discriminator, a scale discriminator was added to reduce the scale difference between the two domains. Secondly, the visual attention network (VAN) that considers spatial and channel attention was used as the feature extraction backbone network to improve the feature extraction capability. Finally, aiming at category samples' uneven distribution problem in remote sensing images, the FocalLoss loss function was introduced to improve the segmentation accuracy of categories with fewer samples by increasing attention to categories with fewer samples. Experimental results show that the proposed model can effectively alleviate the problems of remote sensing images with different features, scales, and unequal distribution categories. The proposed model significantly improved classification accuracy compared to the adversarial domain adaptation segmentation network based on a feature discriminator.

The optical image has high resolution, but it is vulnerable to the adverse environment, resulting in the loss of spectral details. SAR image has strong penetrating power to vegetation, cloud and snow, but it will be interfered by speckle noise. The complementarity of the two images can effectively overcome the limitations of a single image in a complex environment. However, optical image and SAR image have different imaging mechanisms and different gray information, which may lead to the failure of the performance of the two images registration. In order to solve the above problem, in this paper, we propose a method of optical image and SAR image registration based on position constraint. First, the traditional SIFT algorithm is used to register the image coarsely, and then the position of the feature descriptor is locally optimized through the spatial geometric structure characteristics between similar feature points. Experimental results have shown the effectiveness of the proposed method.

Sea fog is a disastrous weather phenomenon, and the analysis of its radiation characteristics is helpful for satellite monitoring of sea fog, so as to reduce the loss caused by sea fog. Considering the different sensor parameters of different satellite loads, the brightness temperature performance of sea fog in the infrared channels of different satellites is also different to some extent. Based on the multi-channel scanning imaging radiometer observation of Chinese new generation geostationary meteorological satellite FY4A, this paper analyzes the thermal infrared brightness temperature characteristics of sea fog from other observed objects in the middle infrared band of 3.75μm and the thermal infrared band of 10.7μm, as well as their brightness temperature differences between those two channels and the texture roughness in the thermal infrared band. It shows that the brightness temperature of sea fog is close to that of the surrounding clear sky water, lower than that of water clouds but higher than that of ice clouds and mixed clouds in the middle infrared and thermal infrared bands. The brightness temperature difference between the middle infrared and thermal infrared band is between [-5K, -2K] for FY4A. And the texture of sea fog is smooth, its texture feature index TFI10.7 mainly ranges from 0 to 0.7, which is significantly different from water clouds, ice clouds, and mixed clouds. These conclusions are of great significance for the sea fog monitoring application of the FY4A satellite.

Synthetic aperture radar(SAR) ship detection plays an important role in ship dispatching, battlefield dynamic tracking and other applications, which all require real-time inference. High spatial resolution of SAR images means a large amount of data and a large computation cost, which makes it difficult to realize real-time inference in hardware with limited resources. Therefore, we propose a new model compression scheme to learn a slim ship detector named Lite-YOLOv4. By modifying network structure to be lightweight and improving polarization-based channel pruning, we generate a compact model. In order to further compress the model, a progressive training-based mixed-precision quantization is proposed to simplify the model bit-representation, reduce its storage requirements and reduce the amount of computation. Finally, the model performance is restored by a self-designed target-guided module with attention mechanism. A lot of experiments were carried out on SSDD, which verify that our method is advanced compared with other CNN-based algorithms. The proposed detector with 0.64MB parameters, achieves 96.6% AP with the calculation cost of 36.11 bitGOPs, in which bitGOPs is 70.9% lower than SOTA, parameter storage is 46.7% lower and AP is 2% higher accuracy.

Bathymetric inversion using multispectral imagery is an effective way to obtain shallow bathymetric information in water but is with relatively low accuracy. This study focuses on the solid disturbance of shallow seafloor substrate variation and proposes an image-segmentation-based method to improve the shallow bathymetry retrieval accuracy. The image is partitioned into different subregions with homogenous substrate properties, and the bathymetric inversion model is constructed separately in each subregion, thus improving the retrieval accuracy. Experimental results of the Ganquan Island region show that the accuracy of the bathymetric inversion was enhanced by 58.1% after image segmentation using muti statistic features.

Recently, ship detection in satellite remote sensing images has gradually become one of the most popular research in the field of remote sensing. In maritime traffic, maritime rescue, and military operations, ship detection based on satellite remote sensing images has important application value. Optical satellite remote sensing has the characteristics of high resolution, strong detection capability, and easy for human interpretation. It has important application significance in both military and civilian fields. In order to respond to the needs of practical applications, we constructed a ship object detection dataset with the Gaofen-2 satellite images. We optimize and improve the existing deep learning network according to the characteristics of optical satellite remote sensing images, and proposes a high-resolution optical satellite remote sensing image ship object detection and classification algorithm based on YOLOv5. In addition, a channel attention module is used to improve the network's ability to represent features. We conduct experiments on the proposed dataset, and the results show that the algorithm has the ability to detect large maritime ships and to classify multicategory military ships.

After decades of rapid development, multiple Global Navigation Satellite Systems (GNSS) have been built and put into use, including the U.S. Global Positioning System (GPS), Russia's GLONASS, China's BeiDou Navigation Satellite System (BDS), and Europe's Galileo system. As a result, global satellite navigation has developed into multiple systems operating simultaneously. To fully utilize the advantages of each system and obtain more stable and reliable results for future research and experimental fusion data, it is necessary to compare and analyse the stability and accuracy of data from each satellite system in detail. The main focus of this paper is: (1) to collect GNSS system data using GNSSLOGGER on different mobile devices; (2) to analyse the collected data using GNSSANALYSISAPP to examine the received GNSS system, satellite power, and pseudo-range variation trends; (3) to plot the collected offline data using MATLAB to distinguish the path data of different GNSS systems; and (4) to demonstrate that the stability, rigidity, and accuracy of the method using a combination of non-differential multi-GNSS systems are much better than those of a single GNSS system. Ultimately, this paper concludes that the use of a combined GNSS system can greatly improve the accuracy and continuity of GNSS positioning when improving the channel quality or hardware equipment is not feasible.

The leaf area index (LAI) of winter wheat is an important parameter for evaluating its growth and yield prediction, and UAV remote sensing technology enables rapid monitoring of LAI. In order to investigate the feasibility of adding other sensitive bands to the traditional vegetation index to improve the accuracy of LAI inversion, this paper improves the traditional vegetation index based on the UAV multispectral images of winter wheat at the time of jointing as the data source, using the more sensitive bands to LAI during the jointing stage. The improved results were also validated by constructing LAI inversion models through ridge regression and random forest regression algorithms. The results show that there is a higher correlation between the improved vegetation index and LAI than before, and that the inversion accuracy of the model based on the improved vegetation index has improved compared to that before the improved vegetation index. The correlation coefficients between each vegetation index and LAI were improved by 0-0.03 during the jointing stage. In terms of model inversion results, the inversion accuracy of the Ridge Regression and Random Forest models based on improved vegetation index modelling improved compared to the pre-improvement period, with the R² of the Random Forest model improving from 0.85 to 0.86 and the R² of the Ridge Regression model improving from 0.76 to 0.77. The study demonstrates the feasibility of improving the accuracy of LAI inversion of winter wheat based on UAV multispectral images of the jointing stage of winter wheat and introducing other sensitive wavebands to improve the traditional vegetation indices, and provides a methodological reference for improving the effectiveness of LAI inversion of winter wheat.

Building information modeling (BIM), with the accurate representation of building geometry structure and component composition, has played a significant role in digital twin development and city information modeling (CIM). Owing to the huge volume of data and complex reference relationships between BIM components, there are few applications that can efficiently transfer, store, and smoothly visualize BIM models. In this study, we propose an efficient BIM data compression and visualization method based on instancing storage and scene hierarchy instanced culling (SHIC) instancing rendering. We identify instanced BIM models based on the IFC semantic information and spatial geometry calculations and store them as instances. To avoid CPU-Bound, we partition the instanced BIM models space into clusters and calculate the visibility set with it as a culling unit. We implemented our method on Unreal Engine 5(UE5), used China Resources Building as experimental data, and a series of experiments were designed to illustrate the advantages of our method in terms of data compression and rendering efficiency. The results show that our method is effective in data compression and keeps a higher frame rate for visualization.

As a major maritime and trade province, the strategic position of marine transportation in Guangdong Province is particularly important. This study uses AIS, waterway, port, sea ownership, and other data, combined with GIS and Python technology to realize track drawing and regional density statistical analysis, extract the dense route of cargo ship routes along the coast of Guangdong and divide the 14 coastal sea areas into dense, relatively dense and non-dense sea areas according to the density of cargo ship routes. Explore the number and density characteristics of cargo routes in different sea areas, and provide a reference for government departments to optimize port resources and plan trade routes. The results show that there are significant differences in the density of cargo routes in the coastal waters of Guangdong Province. The dense routes are mainly distributed in the Pearl River Delta, and there are a small number of dense routes in the remaining sea areas. The density of cargo ships in the sea area of coastal cities shows a decreasing trend with the Pearl River Delta as the center to the two sides of the sea area.

Hyperspectral image (HSI) classification is a vital remote sensing technique that finds numerous real-world applications ranging from environmental monitoring to medical diagnosis, food safety, and security. Recently, deep convolutional neural networks have shown impressive results due to their automatic local feature extraction capabilities. However, current deep networks suffer from severe overfitting on HSI datasets with limited training samples, leading to unsatisfactory classification accuracy. Additionally, most of the attention mechanisms adopted in these networks are twodimensional, neglecting the relationship between spatial and spectral dimensions. To address these problems, we propose a dual-branch, triple-attention mechanism network (DBTANet) for HSI classification. Specifically, DBTANet adopts two parallel branches that independently extract spatial and spectral features, effectively enhancing the divergence and independence of the features. Furthermore, we have developed a unique module with a triple-attention mechanism to automatically focus on the spatial and spectral features that are most helpful for classification, while allowing for information interaction across spatial positions and spectral channels. Additionally, we introduce dynamic convolution to allow for adaptive adjustment of the convolution kernel parameters based on different inputs, further enhancing the feature expression ability of DBTANet. Experiments on three real HSI datasets demonstrate that compared to other recent methods, DBTANet achieves significantly better classification performance, particularly with limited training samples.

The advancement of deep learning has given significant improvement in fine-grained ship detection. However, a major challenge in deep learning based fine-grained ship detection in remote sensing images is the long-tail distribution, not only in the aspect of the number of objects but also in the average instance area of ships. Thus, models tend to be biased towards the categories with more samples or a larger average instance area, i.e., majority classes, and results in low accuracy in recognizing the categories with fewer samples or smaller average instance area, i.e., minority classes. To address this challenge, a novel data augmentation method is proposed in this paper, namely SegRotatePaste. Firstly, a weighted combination-based strategy is proposed by employing the number ratio and the average instance area ratio of ships as evaluation metrics, to identify the minority class and determine the required number of augmentations. Then sea-land segmentation and ship annotations are introduced as auxiliary semantic information to guide the embedding location of ship clips. In addition, various image embedding techniques are leveraged to achieve data augmentation for the minority class. The proposed data augmentation method is evaluated on the HRSC2016 and FGSD2021 datasets. Our method significantly improves the 𝒎𝑨𝑷 of the minority class by 9.3% and 11.5% over the baseline on the HRSC2016 and FGSD2021 datasets, respectively.

This article proposes a rapid monitoring and visualization method for surface changes based on multi-temporal DEM and WebGis. Firstly, high-resolution surface images are obtained using drones and processed into a digital elevation model (DEM). Then, mathematical modeling is performed on the surface change units using cubic voxels, and the voxels are rendered and animated in 3D. Finally, Cesium.js is synchronized with THREE.js camera, and the cubic voxels are integrated into the WebGis space. Application in the No. 5 mine pit in Muli Town, Qinghai Province shows that the proposed method has high efficiency in data acquisition and rich visualization effects in analysis results. It can achieve quantitative simulation of surface change processes in a web browser environment. The proposed method has a low cost and a certain ability to generalize, and can be applied to similar fields for change monitoring.

Visualization of geospatial data has traditionally played a critical role in analysis and decision-making, but recent developments have also brought a revolution in widespread online access to remote sensing data and integration tools, particularly for map-based interfaces. Building the effective tools to support user-centered geographic visualization faces a significant challenge. In this paper, we described the design, development and use of a library package to address the demand for an easy-to-use mapping library for use in online geospatial tool development, especially remote sensing applications. Although original developed for HUB platform, the library is independent and can be used in any standalone applications hosted in a cloud environment. It offers a new way for non-expert programmers such as researchers and students to quickly map-enable their scientific tools which was not an easy task previously, and hence accelerate the time for tool development and knowledge dissemination.

With the continuous development of modern science and technology, aerospace, aviation, and image sensor technologies are being constantly improved, and various new sensors and remote sensing platforms are emerging. Currently, the ability to obtain remote sensing data is improving continuously, and remote sensing data products are characterized by multiple spatial resolutions and multiple loads. Applications based on multi-source data fusion will replace single-source data applications in the future. The prerequisite for fusing data from different sources is that the spatial references for these data must be consistent. Therefore, it is imperative to conduct research on the methods for high-precision automatic registration of images from different sources in order to solve the problem of difficult registration of such images caused by different imaging mechanisms and effects. Taking the registration of optical and SAR images from different sources as an example, this paper presents a method of registering optical and SAR images based on bidirectional style transfer and hybrid feature descriptor. Firstly, a bidirectional style transfer network is used to convert the original optical and SAR images to pseudooptical and SAR images, thus achieving the homogenization of images from different sources. Subsequently, feature sets and hybrid feature descriptors are extracted from optical and SAR images, and the relationship between images from different sources in terms of feature matching is established to achieve high-precision automatic registration of optical and SAR images. The experimental results show that the method proposed in this paper is superior to traditional image registration methods in terms of both subjective perception and objective indicators such as RMSE and has achieved better registration results.

A variety of vegetation indexes were constructed by using Sentinel-2 satellite remote sensing images, and the correlation analysis was carried out with rice processing quality, appearance quality and cooking and eating quality. It was found that the vegetation index based on remote sensing images could invert the brown rice rate and milled rice rate of rice grains, but the prediction of other grain quality indexes was not ideal.

Identifying exposed sewage outlets through visual interpretation of UAV images is a time-consuming and inefficient process. This article proposes a method for enhancing the efficacy of investigations by detecting changes in remote sensing images. The aim is to aid in the preliminary identification of potential sewage outlets. Firstly, the surf algorithm is employed to register the UAV image with the remote sensing image that was utilized in the previous investigation. Then, we employed gamma correction and Histogram matching techniques to adjust the brightness and color histograms of the two images. Subsequently, the two images are classified separately using a supervised classification technique to distinguish segments of man-made objects from the natural background. Finally, the changes in classification between the two images were identified. The area where the natural background transitions into an artificial structure was combined with the previously identified locations of sewage outlets, as the key areas for further visual interpretations. The practical results in the Shichuan River Basin demonstrate the potential for enhancing the UAV-based survey technology for ecological inspections, and improving the efficiency of river drainage outlet investigation through the analysis of UAV images.

Owing to the global energy conservation and emission reduction, the wind power industry is developing rapidly in China because of abundant wind energy resources. Wind power has become the third largest power source and remain growing rapidly in China. Efficient and accurate measurement of the location of wind turbine is important for assessing local wind power development. Remote sensing and deep learning technologies offer technical possibilities for wind turbine identification. However, until now there is few publicly available land remote sensing wind power turbine object detection dataset. Aiming at this question, this paper designs a land remote sensing wind turbine dataset, which includes 4459 individual wind turbines in various environmental backgrounds, including sandy land, woodland, grassland, snowy land, and wasteland. In addition, the dataset is composed of YOLO and VOC markup formats to meet the experimental needs of researchers. This paper also evaluates the dataset under the various mainstream object detection frameworks and reach 92% in mAP. The dataset can be obtained from Zenodo website https://doi.org/10.5281/zenodo.7808269.

Highway landslides are a natural disaster that can have devastating impacts on transportation, the economy, society, and people's lives. To mitigate the damage caused by these landslides, it is crucial to assess their susceptibility. In this study, we propose a novel method for assessing the risk of highway landslides using an XGBoost model that utilizes a variety of data sources, including remote sensing data, terrain, and meteorology. The results of our research are highly promising. Our model achieved an AUC score greater than 0.93/0.83 on the test/validation set, indicating high accuracy in predicting the susceptibility of landslides. By identifying areas that are at high risk for landslides, we can take proactive steps to prevent or mitigate the damage caused by these natural disasters. The insights gained from our study have important implications for land use planning, infrastructure development, and emergency management decision-making.

Forest canopy closure y is an important parameter in forest resource surveys, and plays a crucial role in evaluating and monitoring the stability of forest ecosystems. With the continuous development of remote sensing technology, the estimation of forest canopy closure using multi-source remote sensing data has become a hot research topic. In this study, a regression model was constructed using machine learning algorithms based on laser point cloud and multi-source optical remote sensing data, for estimating the canopy density of large forest areas. Firstly, the true values of forest canopy closure were calculated from airborne laser scanning (ALS) point cloud data as the dependent variable of the regression model. Secondly, 18 features such as vegetation indices and textures were extracted from Sentinel-2 MSI, Landsat-8 OLI, and Sentinel-1 SAR image data as the independent variables of the regression model. Then, 14 forest sample plots in Guangxi region were used as examples to investigate the impact of different variable combinations and machine learning methods on the estimation of forest canopy closure using Random Forest Regression (RFR) and Support Vector Machine Regression (SVR) models. Finally, the best variable combination and machine learning method were selected to map the forest canopy closure of Laibin city, Guangxi. Experimental results showed that RFR had better inversion performance than SVR, and the combination of S2+S1 had the highest accuracy, with a correlation coefficient R2 of 0.703, root mean square error RMSE of 0.19, and mean absolute error MAE of 0.13. Additionally, polarization features significantly improved the accuracy of forest canopy closure estimation.

In terms of the shortcomings of traditional vertical photogrammetry in urban modeling, such as tilt deformation and occlusion of near ground objects, a typical area in Qiaocheng District of Bozhou City, China is selected as the study area. DOM, TDOM, vertical traditional 3D model, vertical true 3D model as well as realistic 3D model based on low-altitude vertical photogrammetry and tilt photogrammetry data of the study area are constructed, comparing the modeling accuracy of different models, and the accuracy of DLG production and drawing details of different 3D models as well as the accuracy of extraction of building area are studied. The spatial data of the check points in the study area is measured to test the feasibility of constructing large scale (1:500) of digital line graphs via different 3D models. The results show that: ①the mean square error in plane of DOM, TDOM, vertical traditional 3D model, vertical true 3D model and realistic 3D model is respectively ±0.368cm, ±0.78cm, ±4.67cm, ±2.63cm, ±0.80cm, the mean square error in elevation of which is respectively±8.16cm, ±0.89cm, ±8.17cm, ±0.84cm, ±0.93cm. ②refer to the "Code for Aerial Triangulation of Digital Aerial Photogrammetry", according to the realistic 3D model, the minimum accuracy in plane is 4.79cm, the maximum accuracy in plane is 0.15cm, and the mean square error in plane is ±0.80cm; the minimum accuracy in elevation is -5.72cm, and the maximum accuracy in elevation is 0.26cm, the mean square error in elevation is ±0.93cm, it can be seen that the realistic 3D model meets the standard of large scale (1:500) topographic map drawing. ③measure 150 evenly distributed checkpoints in the study area through GPS RTK, and compare them with the elevation check points extracted from the DLGs produced by the vertical traditional 3D model, the vertical true 3D model as well as the realistic 3D model, the maximum error in elevation of the comparison result is 54.39cm, 17.41cm as well as 14.89cm respectively, and the mean square error in elevation is ±2.27cm, ±0.35cm as well as 0.33cm respectively, referring to "1:500 1:1000 1:2000 Topographic Map Aerial Photogrammetry Industry Specifications", it can be seen that the digital line graphs constructed by vertical true 3D model and realistic 3D model meet the accuracy of drawing large scale (1:500) digital line graph, specially, the digital line graph constructed by the realistic 3D model is more complete and more accurate.

The GNSS-R approach improves altimetry, which is essential for maritime remote sensing data. High-precision altimetry requires carrier phase continuity, which is limited by several variables. This paper offered a grey model solution. The proposed method works for GNSS-R sea surface height measuring with centimeter-level accuracy. Field experiments process BeiDou satellite reflected signals in the B1 and B3 frequency bands, and GEO and IGSO inversion findings are analyzed. This method improves the GEO inversion effect, the satellite with a large altitude angle, and sea surface altimetry when comparing B1 and B3.

Speed and real-time are important indexes for marine environmental monitoring. This paper proposed a method to design the multi-sensor marine environmental monitoring system for the hybrid-power fast unmanned surface vehicle (USV) to meet the requirements of the marine environmental monitoring system. This USV comprises a ground base station, data transmission system, data acquisition system, and power system. It has the functions of remote control, status monitoring, autonomous obstacle avoidance and marine environmental data monitoring, data transmission, and processing. The critical marine environment acquisition module consists of five parts: the multi-beam underwater three-dimensional bathymetry survey, shallow profile survey instrument, ADCP current survey, and CTD sound velocity profile survey. At the same time, the USV passed the prototype testing, which verified that the autonomous navigation and obstacle avoidance system performed well and fully met the navigation needs of the USV. The combination of a diesel surface propeller propulsion device and an electric propulsion device could meet the rapid arrival of monitoring sites and meet the low noise and low vibration working environment during environmental monitoring. The multi-sensor monitoring system could continuously and stably measure the data of the monitoring sea area and transmit the real-time monitoring data. The final results showed that the feasibility and stability of the USV for marine environmental monitoring provided a new idea for developing unmanned automation of deep-sea marine environmental monitoring. Meanwhile, it could achieve high-precision marine surveying and mapping, marine hydrology, and environment monitoring.

The extraction of building information is of great significance for environmental change detection and urban development, and is also conducive to the country's macro-control and scientific management. Based on the four Landsat8 OLI image data of Huangdao District, Qingdao City in 2015, 2017, 2019 and 2021, this study proposes a building extraction method combing random forest feature selection with SVM algorithm. In this method, first, 22 feature variables are extracted from the images. Second, the random forest is used to screen out the important features. Finally, the selected feature variables are combined with the SVM classifier. In this study, ten features out of 22 features are screened out by RF, and their feature importance reach 94.37%. Moreover, the building extraction effect after feature selection is significantly better than that before selection. The results have shown that the overall accuracy and Kappa of the image classification results based on SVM in each time phase are higher than those based on other machine learning algorithms, and the overall classification accuracy and Kappa are both 96% and 0.95 or more.

The aging of bridges, traffic congestion, and vehicle overloading have inevitably aggravated the damage of urban bridges in recent years, and it is extremely important to predict and analyze the settlement of urban bridges to monitor their safety conditions. The PS-InSAR (Permanent Scatterer Synthetic Aperture Radar) technique compensates for the drawbacks of the traditional InSAR technique such as atmospheric phase delay, and thus can obtain more accurate settlement information, which has unique advantages in long-term slow accumulation monitoring of surface deformation, and is especially suitable for this experiment to monitor and analyze the uneven settlement of urban bridges. In this paper, we obtain the uneven settlement of urban bridges based on PS-InSAR technique, and combine genetic algorithm and neural network algorithm (GA-BP) to achieve the prediction and analysis of uneven settlement of urban bridges. The method is also applied to the simulation experiment and analysis of uneven settlement prediction of Suzhou Bridge and the surrounding ground surface in Beijing. The experimental results show that the root mean squared error (RMSE, Root Mean Squared Error) of applying the GA-BP algorithm to predict the uneven settlement of urban bridges and the surrounding ground surface is 0.57500 mm, which has a good accuracy and proves that the method can effectively predict the uneven settlement of urban bridges and the surrounding ground surface, and with the increase of the number of experimental data. The best iteration the fewer the number of iterations, proving that its prediction performance will be better with the growth of the number of acquired data.

The detection accuracy is low in the existing deep learning detection to extract solar PV panels sticky, incomplete and under the interference of factors such as roofs and substations in four different scenarios: water, saline, low-density grass and fields. In this study, U-Net neural network is improved by adding attention mechanism and residual network. A deep learning network model for solar PV panels detection that incorporates attention mechanism and residual structure: UNet Pro, is established. This model, based on U-Net, is able to pay more attention to the detection of solar PV panels by incorporating up-sampling and down-sampling in the process of attention mechanism. It is able to pay more attention to the effective detailed information and exclude the interference factors. Meanwhile, the residual module is embedded after the Relu function in each layer of the U-Net neural network. It effectively avoids the problems of gradient explosion, feature loss and network degradation. For solar PV datasets in different scenarios, the effects of adding the attention mechanism and the residual structure have been improved to different degrees. Compared with U-Net, mF₁ value can be improved up to 94.67% and at most 4.46%. mIoU can be improved up to 93.75% and at most 4.74%.

The Surface Albedo reflects the ability of a surface to reflect solar radiation, which is a major factor in the surface energy budget and atmospheric motion in images. Low-resolution Surface Albedo cannot meet daily needs, and high-resolution Surface Albedo has become a hot research topic. GF-1 Wide Field of View (WFV) data can achieve 16 m wide range and high precision ground observation data, making it suitable for the production of high quality Surface Albedo products. In this paper, the 6S Radiative Transfer Model is used for atmospheric correction of GF-1 multispectral WFV data, and the Surface Albedo of Zhengzhou Airport is estimated based on the Bidirectional Reflectance Distribution Function (BRDF) model using its observation angle information. Finally, the inversion results of GF-1 are compared with existing MODIS Surface Albedo products. The results show that the Root Mean Square Error between GF-1 Albedo and MODIS Albedo is 0.0218 based on BRDF products as a priori knowledge. The inversion results of GF-1 data are in good agreement with MODIS Albedo products, which has certain reference significance for GF-1 satellite Albedo product generation and application.

Seabed sediment classification is meaningful for seafloor habitat mapping and marine resources exploration. Multibeam Echo-sounding System can acquire backscatter and topographic information, which has become the mainstream for detecting seabed sediment. Because topographic features are closely correlated with sediment distribution, and sediment types tend to be distributed in a continuous pattern, we propose a seabed sediment classification based on topographybased image partitioning. Multi-resolution segmentation and K-means clustering are utilized for image partition based on topographic features, then the classification maps of different subregions are integrated as one result map. The experimental results show that the partitioned methods considerably outperform the global methods in terms of accuracy. This method provides ideas for achieving high-resolution seafloor mapping in the future.

Oblique photogrammetry models are organized in an index tree which incorporates the concept of Hierarchical Level of Detail (HLOD) for optimal rendering of spatial data in the widely used 3D Tiles specification. The inefficient data structure of the index tree, such as gird, may cause problems such as rendering delay and memory overflow on the client-side. This paper proposes an adaptive quadtree index reconstruction algorithm to construct an upper quadtree filled with virtual nodes for tiled tiles of a single oblique photogrammetry dataset using the named coordinates and the actual geographic coordinates of the tiles. The improved structure enables fast elimination of invisible tiles by exploiting the visibility relationship between parent and child nodes, thus improving the traversal rate of tiles. Experiments prove that the algorithm is effective in improving the loading efficiency of local tiles in a single tileset as well as tiles at the junction of multiple tilesets, which is applicable to most browsing scenarios. In addition, the algorithm minimizes the negative impact of the adaptive quadtree structure when observing panoramic tiles by adequately balancing the number of virtual nodes, the depth of the quadtree and the efficiency of tile culling, and experiments demonstrate that the time consumption is within an acceptable range, and the negative impact is insufficient to offset the positive effect of the performance enhancement.

Transient electromagnetic method for undulating terrain is a time-domain electromagnetic exploration method considering topographic effect. By studying the influence of topographic factors on the interpretation of electromagnetic data, qualitative concepts such as the form, scale and electrical parameters of ore bodies can be given as far as possible, which is the key to efficient exploration of mineral resources in terrain areas. In this paper, based on the three-dimensional forward modeling theory of time-domain electromagnetic method, the multi-physical field simulation software COMSOL is used to establish a profile measurement and exploration model. The transient electromagnetic response characteristics of different fluctuation degrees in flat and undulating terrain under different emission current intensities are studied respectively. The results show that the amplitude of electromagnetic response increases with the increase of emission current intensity under flat terrain conditions. In undulating terrain, the induced electromotive force amplitude at the barrier increases with the height of the barrier, but decreases with the depth of the graben.

With the improvement of imaging accuracy required by oil and gas exploration, the anisotropy of underground media has attracted more and more attention. In order to improve the accuracy and stability of the forward modelling of seismic wavefield in VTI (Transversely isotropic media with a vertical symmetry axis) media. In this paper, a solution based on multiple-angle formula expansion in VTI medium is proposed. Combined with the multiple-angle formula and Taylor's expansion, the harmonic function operator including the pseudo-differential operator is approximated by spectral method, and then a VTI medium based on the multiple-angle formula is solved. Numerical tests based on two-dimensional uniform and complex models show that the proposed method can maintain the stability of the extrapolated wavefield at large time steps. On this basis, the wavefield separation is realized, and the simulation test is carried out by the layered model. The results show that the proposed method is effective for up-down going separation.

Satellite remote sensing has been widely used for urban expansion analysis and land cover change monitoring due to its advantages of large coverage and short revisit cycle. With the rapid development of urban construction, urban heat island effect becomes more obvious. It is of great significance to use remote sensing technology to monitor urban temperature and analyze the internal causes of heat island effect. Zhengzhou was selected as a case study, and uses remote sensing images from March and September 2020-2022 to invert the temperature of this area. After separating the temperature changes caused by normal climate, the influence of land cover type changes on heat island effect is analyzed by interpreting and comparing different land cover types. The results show that land cover type changes have an important impact on urban heat island effect. The distribution of vegetation and water bodies can alleviate the rise of urban surface temperature to some extent, while high-density built-up areas have more significant surface temperature increase. However, through comparative studies of remote sensing images, it is found that light-colored building materials with high reflectivity have obvious improvement on urban heat island effect.

Earth Single Base Inverse Synthetic Aperture Radar Imaging (ISAR) can detect and observe the moon all day, all weather, and is not affected by weather conditions. At present, there are few studies on the simulation of Earth's single base ISAR imaging methods at home and abroad, and when conducting simulations, the lunar orbit in the Earth Moon motion model is generally idealized and assumed to be circular. In this paper, a ground-based ISAR observation model for lunar imaging is derived in detail, and two methods, range Doppler and focused range Doppler, are used to simulate and analyze the impact of the ideal model. The impact of the lunar orbit on ISAR imaging simulation is analyzed and verified by establishing different ground lunar motion models. The experiment found that when conducting Earth based single ISAR imaging simulation for the moon within 400s of t, the lunar orbit parameters can be ignored and do not affect the imaging results. However, in the case of continuous observation with t being two months, the positioning accuracy of the lunar orbit parameters differs greatly from the actual parameters, and the impact cannot be ignored. The research in this paper provides important reference and guidance for the impact of the lunar orbit approximation on subsequent lunar imaging, and provides an experimental and theoretical basis for accurate lunar imaging.

Traditional large-scale topographic map measurement has the problems of long work cycle and high cost. UAV tilt photogrammetry technology is used for large-scale topographic map measurement, and low altitude drones are used to quickly obtain multi angle image data, achieving automatic 3D modeling and 3D measurement technology for real scenes, and achieving high-precision large-scale topographic mapping. Through the specific experimental process, the process of large-scale terrain operation for low altitude drones was elaborated in detail, and the results showed that its accuracy met the requirements of large-scale terrain mapping. Low altitude drone oblique photogrammetry provides a high-precision, high-efficiency, and low-cost measurement method for large-scale topographic mapping.

With the continuous launch of spaceborne SAR, InSAR three-dimensional measurement has been further developed on the basis of one-dimensional and two-dimensional methods based on interferometry. Based on the SM-VCE method, this paper performs high-precision three-dimensional surface deformation reconstruction in Maduo County, China. Through different data sources involved in the reconstruction of three-dimensional deformation, the difference between SM-VCE and traditional algorithm (WLS) is analyzed. Finally, the difference between them is quantitatively analyzed by GNSS stations.

Building change detection is a crucial task for urban area renewal and earthquake disaster emergency management. However, the conventional assessment method mainly relies on field investigation and visual interpretation of images, which entails drawbacks such as low efficiency,and high cost, and information abstraction. These shortcomings make it challenging to meet the requirements of urban spatial change detection and disaster response and recovery. Therefore, this paper proposes a building change detection method based on convolutional neural network (CNN) to detect changes in buildings in multi-temporal remote sensing images, which was verified using two remote sensing images before and after the Türkiye earthquake on February 6, 2023. In this method, we improved DenseNet 121 to serve as a key component of the deep learning model, utilizing dense connections between layers to improve feature propagation efficiency and enhance the capture of graphic details. Simultaneously, we designed a novel loss function based on the Tanimoto coefficient to measure the difference between the predicted segmentation map and the live segmentation map. Experimental results show that the proposed building change detection method is effective in detecting building collapse in earthquake disasters, with a quantitative accuracy of F1 score exceeding 70%. Additionally, the model design approach for building change detection presented in this paper can provide valuable insights for other element detection tasks based on remote sensing images, such as road detection.

Buildings are the most important elements in cities. Building urban building models is of great significance for the establishment of digital cities. The level of its modeling technology restricts the development of urban 3D modeling technology. In order to improve the accuracy of 3D geometric modeling of urban buildings and truly reflect the spatial layout of urban buildings, the 3D geometric modeling of urban buildings based on airborne LiDAR point cloud and image is studied. The airborne LiDAR and tilt photogrammetry technology are analyzed, and the airborne LiDAR point cloud data is filtered by using the two-level grid to eliminate the gross error, and the airborne LiDAR point cloud data is filtered by using the morphological filtering algorithm. Through the principle of spatial location, the spatial geographic location information of the oblique image is restored. Extract the step line feature of the urban building point cloud and the roof patch of the urban building, accurately register and locate the urban building model based on the inclined image, and combine the urban building texture map to realize the three-dimensional geometric modeling of the urban building. The experimental results show that the proposed method has good 3D geometric modeling effect of urban buildings, can truly reflect the spatial layout of urban buildings, and effectively improve the accuracy and efficiency of 3D geometric modeling of urban buildings.

The pose information of aircraft is an important index to study flight status and aircraft performance[1]. This article mainly focuses on the research of aircraft attitude estimation based on contour matching, intending to achieve pose estimation of non-contact long-distance moving objects under the rigorous formula system of photogrammetry. The rationality of the algorithm proposed in this article has been proven through the analysis of experimental results.

Rapeseed is a valuable crop used for its oil-rich seeds and bright yellow flowers. Accurately mapping rapeseed planting areas can not only help optimize crop management and improve yields, but is also essential for various studies in agriculture, including sustainable agriculture and global food security. Remote sensing technology and image analysis methods have been widely used for rapeseed mapping. However, most of these remote-sensing-based methods producing rapeseed maps with medium-resolution multi-spectral images which cannot support high-precision rapeseed mapping of smallholders. Thus, with the help of the unmanned aerial vehicles (UAVs) and deep learning (DL), a high-precision approach was developed to mapping rapeseed at Smallholder parcel-level. The overall accuracy reaches 97. 12%, and the intersection over union reaches 81.37% on study area.

Heiquan Reservoir, Xining's seventh source of drinking water, plays a significant role in ensuring water for industry, agriculture, and ecology in Xining and the neighboring areas. Therefore, monitoring and evaluating the water quality of Heiquan Reservoir is required. Chlorophyll content in water bodies is an important indication of eutrophication because it can represent environmental changes as well as the impact of human life and productivity on water bodies. This study mainly uses Landsat-8 OLI images from 2015–2019 by constructing an inversion model of chlorophyll sample point concentrations extracted from MODIS-based inversion models. The investigation concluded that the chlorophyll concentration in Xining City's Heiquan Reservoir falls in winter each year, gradually increases in spring, and reaches its highest in summer. Moreover, the parameters influencing the distribution of chlorophyll concentration from water temperature were investigated using inverted Landsat-8 OLI images of Heiquan Reservoir's water surface temperature. It was found that the chlorophyll concentration in Heiquan Reservoir showed a positive correlation with water surface temperature.

With the development of remote sensing technology, feature extraction methods are gradually diversified, mainly through optical and radar remote sensing. Optical remote sensing images contain a wealth of spectral information, whereas radar remote sensing images are all-day and all-weather. But they have some drawbacks. As a result, different features of Sentinel-1A and Landsat-8 images are combined in this paper to exploit their advantages for feature extraction experiments fully. Data preprocessing for Sentinel-1A and Landsat-8 is performed in this paper, followed by polarization feature extraction using H-α-A decomposition for Sentinel-1A and texture feature extraction using GLCM for Sentinel-1A and Landsat-8, followed by feature combination and classification using SVM classifier. Finally, the accuracy of classification results is evaluated. The results of this paper are as follows: the worst accuracy result is based solely on Landsat-8 spectral features combination, with an overall accuracy of only 80.61% and a Kappa coefficient of 0.6702; the accuracy of features combinations based on Landsat-8 spectral plus texture and Sentinel-1A polarization plus texture is improved. The best accuracy is 90.78%, and the Kappa coefficient is 0.8473. The experimental results show that multi-source remote sensing-based feature extraction with multiple feature combinations is more advantageous.

In rock physical modeling, WIGED mechanism is selected to guide and calibrate pore fluid modeling in reservoirs with low porosity and low permeability, based on the occurrence characteristics of micro-nano pore fluid in rocks in the target area. The viscoelastic model with fewer parameters was selected to represent the petrophysical model with more parameters, which can avoid the disadvantages of too many parameters and unavailable of the petrophysical model, so as to achieve the purpose of scale coarsing of the model. Model test and field data application demonstrate the feasibility and accuracy of this strategy.

As a widespread mesoscale phenomenon in the ocean, the study of mesoscale eddies has received much attention and focus. Scientists have been using data sources such as real measurements, satellite remote sensing data, and reanalysis/numerical prediction products to automatically identify and synthesize the 3D structure of mesoscale eddies locally, but at present, the 3D structure of mesoscale eddies is mostly portrayed in the form of 2D flat expressions such as static images, which has certain limitations for the expression of mesoscale eddies with multi-dimensional and dynamic characteristics, and fails to show the complete 3D structure of mesoscale eddies. The three-dimensional structure of the mesoscale vortex is not shown. In this paper, based on the flow field vector geometry feature vortex identification algorithm, with the help of Unreal Engine 4 (UE4) 3D engine and particle system technology, we developed a 3D structure information extraction and visualization algorithm for mesoscale eddies, and constructed a 3D dynamic display and analysis model, which breaks through the traditional 2D display form of mesoscale eddies and realizes the 3D dynamic visual representation and analysis of ocean mesoscale eddies. It provides a dynamic, intuitive and convenient means to study the 3D structural characteristics of mesoscale eddies, and also provides an effective support to investigate the dynamic changes of mesoscale eddies and their interactions with other ocean processes.

To solve the incomplete point cloud model obtained by a single instrument, cross-source point clouds are widely used. When data are taken from various surveying instruments, the problems, such as density variation and viewpoint variation, are caused. A registration method based on Gaussian Mixture Model is proposed in this study. With the threedimensional information of the point clouds described by the model, the registration of two kinds of point clouds is transformed into the probability density estimation. A water tower is chosen as the example, and the difference between both the algorithm based on the Gaussian Mixture Model and the Iterative Closest Point algorithm are compared. The results show that the former method behaves better than the latter in both runtime and accuracy.

The mining of open pit quarries has caused great damage to the local ecological and geological environment, and the problems such as geological disasters brought about by mining have seriously affected the ecological environment of the local and surrounding areas. Carrying out the corresponding ecological restoration project is the main solution at present. With the completion of the ecological restoration project, the evaluation of the ecological restoration level lacks tracking monitoring and data support. This paper proposes the use of remote sensing imagery to assist in the assessment and detection of the subsequent effect of mountain mine treatment. The satellite image data of Chabuga Mountain in 2019 and 2022 are obtained by remote sensing technology, and the post-ecological restoration images in 2022 are obtained with the field inspection, and the NDVI index, WET index, NDBSI index and LST index are used as reference data, and the principal component analysis method is selected to use the remote sensing ecological index RSEI model as the evaluation standard, so as to compare the results of mountain mine restoration project. Research results show that after the mine repair work, the annual average value of remote sensing ecological index of the mines in the study area was 0.56509 in the middle of the ecological restoration project, which was restored to 0.69072, and the remote sensing ecological index was significantly improved, and the ecological restoration work of mountain mine management was effective.

At present, in the related research on geographical entity division, the multi-scale expression method is mostly used to describe the detail degree of geographical features, and there is a lack of a method to describe objects from the perspective of granularity, so it is hard to meet the application requirements of objects in different scenarios. This paper introduces quotient space theory and proposes a representation method of geographic entities at different granularity levels. Firstly, the conditions that can be divided into geographical entities within the same granularity are determined. Then, the granularity partition model of geographic entities is constructed from three aspects: text similarity, literal similarity and semantic distance similarity. Finally, each quotient space is organized by the theory of quotient space particle size synthesis, and the whole geographical entity is divided into multiple granularities, and the experimental verification is carried out. The results show that the method can well meet the multi-level and multi-granularity requirements of geographic entities, and realize the hierarchical optimization of geographic entity scenarios.

At present, feature matching and ontology matching are often used in the matching and fusion of multi-source geographic data, but the temporal characteristics of multi-source geographic data are less considered in feature matching. Therefore, according to the spatial, temporal and semantic correlation of multi-source geographic data, this paper proposes a multi feature semantic correlation calculation model of multi-source geographic data. Firstly, six kinds of similarity are extracted: spatial similarity, temporal similarity, literal similarity, speech similarity, word bag similarity and category similarity. Secondly, the six kinds of similarity are combined by multi-layer perceptron. Finally, the accurate matching judgment of multi-source geographic data is realized based on weighted multi feature fusion matching algorithm. The results show that the model realizes the accurate matching of multi-source geographic data on the basis of considering the time characteristics.

Marine forecasting plays a crucial role in promoting the development of the marine economy and the prevention and mitigation of marine disasters scientifically. In China, the development of marine forecast information system for the public is relatively lagging, and there is few applications software that can acquire the elements of marine dynamics and environment in the China seas, which cannot meet the increasing and urgent demand of users for marine forecasting information. This paper develops a public information service system for ocean forecast, takes location-based ocean forecasting as the entry point for public users, designs specific application scenarios, and gives the key methods for technical implementation. The system provides the public with fast query and personalized service of marine forecast and advisory information by mobile application and website.

As a key geographic information product to accelerate the development of basic surveying and mapping and improve the service capability of surveying and mapping geographic information, geographic entity data can realize a new pattern of three-dimensional basic geographic information collection, data materialization, efficient processing, integrated management and diversified services through unique coding, objectification processing and entity association and integration. By using multi-source heterogeneous massive basic data, this paper studies the fusion and conversion methods of multi-source spatial data to geographic entity data, solves the intelligent processing of geographic geometric features and topological relationships in the process of geographic entity data fusion, greatly improves the efficiency of geographic entity data fusion, and accelerates the pace of construction of a new basic surveying and mapping system in our province.

Geographic scene modeling is crucial for achieving digital twins, and has seen significant advancements in recent years with the development of geographic scene data collection and three-dimensional modeling technologies. However, most existing modeling methods are based on the Euclidean space and rely on layer-based data organization mode, making it challenging to address the non-uniform spatiotemporal reference of multi-source heterogeneous data during geographic scene modeling. This results in complex models that do not meet specific application requirements. To address these issues, this article proposes incorporating geometric algebra (GA) theory into geographic scene representation and exploring representation methods based on GA. By leveraging the coordinate independence and dimension independence features of GA, this approach aims to resolve the problem of non-uniform spatiotemporal reference and reduce the complexity of geographic scene models while improving their availability.

The traditional coastline detection method is affected by the surrounding environment, leading to some error of the acoustic wave, which affects the detection effect of the coastline. Therefore, a method for remote sensing image coastline detection based on local information fuzzy clustering is designed. The differential edge features of coastline remote sensing images were extracted, and the superpixels of remote sensing images were segmented and clustered for the whole remote sensing image pixels. The coastline detection model of the remote sensing image is established to realize the accurate detection of the coastline in the remote sensing image. Simulation experiments verify that the proposed detection method has high detection accuracy and can be applied to practical applications.

In order to meet the needs of information and intelligent service of power transmission and transformation project, the power industry proposed the grid information model (GIM), which is a digital information model of the whole life cycle of power grid components and is of great significance for realizing information sharing in power industry. On this basis, State Grid has formulated a unified GIM model specification, but the specification only gives the parameters of the fundamental geometry units, and does not involve the specific implementation. The existing platform cannot build the model directly according to the given parameters. And for that, this paper proposes a 3D parametric modeling method for GIM, establishes a hierarchical tree structure based on the fundamental geometry units in the specification, and proposes a GIM fundamental geometry units modeling algorithm applicable to this structure, according to which the automatic construction of GIM fundamental geometry units can be realized. Finally, the feasibility and effectiveness of the method proposed in this paper are verified by experiments.

The unsupervised algorithm extracts hyperspectral image features, focusing on one aspect and often ignoring some other information. To guarantee the effective extraction of information before image analysis, an unsupervised feature extraction technique with hyperspectral imaging data based on multidimensional feature fusion is proposed. First, we use principal component analysis (PCA) to map the high-dimensional data to a simpler space and extract the spectral features based on the elimination of redundant relationships. Then, the multi-directional spatial feature extraction algorithm of Gabor texture and morphology is utilized to extract each primary component's spatial properties. Finally, the spectral features, morphological features, and Gabor texture features are fused together by the vector stacking fusion. In this paper, the HSI information is extracted unsupervised using the previous technique, and HSI classification experiments using support vector machines are carried out to validate the efficacy of the information. The experiments demonstrate that the proposed method improves the Kappa coefficient by at least 14% in the MUUFL dataset and by 30% in the Trento dataset compared with the traditional spectral feature extraction, which significantly improves the classification accuracy.

In the application of reservoir geological description using well logging data, some well logging curves are often distorted or missing. For this reason, well logging curve restoration has always been a hotspot and difficulty in related research fields. The current methods, which use the traditional signal restoration method and the restoration method based on machine learning such as neural network, do not sufficiently express and utilize the correlation information between different logging curves in the same well, and with poor cross-well adaptability. For these problems, a logging curve restoration method based on graph representation for spatio-temporal correlation information mining is proposed. The proposed method firstly describes the graph structure between different logging curves based on the commonly used graph representation learning in the field of signal processing, so that the logging curves can establish a nonlinear mapping relationship. And then, the hierarchical structure of the deep forest is introduced to realize the representation of the longitudinal information of the logging curve. Through the experimental verification of the proposed method, the prediction results of multiple well logging curves from different wellheads are obtained.