Aiming at the problem of inaccurate classification of textures at different scales in traditional texture classification methods, this paper proposed a deep convolutional neural network (CNN) based on improved residual blocks to increase the accuracy of texture classification. First, the two convolution layers in the original residual block were replaced with two dilated convolution layers, and a spatial attention module after the second dilated convolution layers was inserted. Thereafter, the residual connections were used for feature fusion to obtain a greater receptive field and attention-enhanced features. Second, based on the improved residual blocks, a multi-scale texture classification CNN was stacked in a way of increasing the number of block channels. The experiment was performed on a 64-class texture dataset. Experiments show that, compared with the state-of-the-art methods, the proposed method achieved a higher classification accuracy of 99.17%.
Research on object detection algorithms with higher accuracy and faster detection speed is currently the main concern. In order to improve detection performance, an improved object detection algorithm using YOLOv3-tiny based on pyramid pooling is proposed. First, an improved pyramid pooling module using adaptive average pooling is designed to efficiently extract global feature information, and then combine the module with YOLOv3-tiny to explore the impact of different combinations on the detection results. The experiment used PASCAL VOC2007 trainval and all PASCAL VOC2012 for training and validation, and used PASCAL VOC2007 test for testing. Experimental results show that the proposed network improves mAP by 1.8% compared to YOLOv3-tiny while the detection speed is almost the same, which better achieves the balance of detection speed and accuracy.
In order to more accurately locate and segment the varistor image to achieve the varistor image data set necessary for automatic construction of deep learning. This paper proposes a method for locating and stitching the body and stitch of varistor based on Hough transform and mathematical morphology. In order to obtain an image that eliminates surface reflection, the method first acquires a varistor image through a coaxial light source. Secondly, performing preprocessing on the image based on denoising, graying, and binarization; then, using the Hough transform based on circle detection to locate the body of the resistor; further separating the body and the stitches, firstly performing edge searching on the positioned body portion, and then performing background filling on the inside of the body, and finally using a mathematical morphology etching operation to eliminate the edge marks of the body to obtain the positioning of the stitches. The experiment aimed to locate and segment 91 varistor samples, and use the effective and correct data indicators to evaluate the segmentation results. The experimental results show that the actual results of the proposed method are ideal and have a good target segmentation effect, which is beneficial to provide reliable varistor image data sets necessary for deep learning.
Surface defect recognition is one of the key technologies for varistor quality inspection, which can greatly improve detection efficiency and performance. In order to more accurately identify the surface defects of a varistor body and the pins, a method for identifying the surface defects based on deep convolutional neural networks (CNN) is proposed. The proposed method mainly includes four stages: image acquisition and data set construction, convolutional neural network modeling, CNN training and testing. Firstly, varistor images are acquired, and the body and pins of the varistor are segmented by image segmentation method. The number of samples is increased by data augmentation to make a data set of 5 classes. Secondly, according to the appearance characteristics of varistor, a CNN model is designed for varistor surface defect recognition. Third, using the created data set, the training data set with category labels are input to the proposed CNN for training. Finally, 1200 test samples were tested on the trained model in the test phase and the performance of the proposed algorithm was evaluated using mean average precision. The experimental results show that our method can identify the surface defects of the main body and pins of varistor efficiently and accurately.
Segmentation of pigmented lesions is often affected by factors such as hair around the skin lesions, artificial markings, etc., and the complexity of the lesion itself, such as lesions and skin boundaries is not clear, the internal color of lesions is variable, etc., resulting in segmentation difficulties. Aiming at the problem that the segmentation method of pigmented skin lesions using only random forests is not accurate, a segmentation method for pigmented skin lesion using a combination of random forest and fully convolutional neural networks (FCN) is proposed. This method firstly classifies and recognizes skin lesion images based on random forests to obtain a probability distribution of the lesions and the background. Then, the other probability distribution is obtained using FCN based on an improved loss function. Finally, the classification results of random forest and FCN are fused into the final image segmentation results. The experimental results show that the combination of random forest and FCN yields better performances than using random forest alone, in particular, can increase the sensitivity by about 20%.
The incidence of cutaneous malignant melanoma, a disease of worldwide distribution and is the deadliest form of skin cancer, has been rapidly increasing over the last few decades. Because advanced cutaneous melanoma is still incurable, early detection is an important step toward a reduction in mortality. Dermoscopy photographs are commonly used in melanoma diagnosis and can capture detailed features of a lesion. A great variability exists in the visual appearance of pigmented skin lesions. Therefore, in order to minimize the diagnostic errors that result from the difficulty and subjectivity of visual interpretation, an automatic detection approach is required. The objectives of this paper were to propose a hybrid method using random forest and Gabor wavelet transformation to accurately differentiate which part belong to lesion area and the other is not in a dermoscopy photographs and analyze segmentation accuracy. A random forest classifier consisting of a set of decision trees was used for classification. Gabor wavelets transformation are the mathematical model of visual cortical cells of mammalian brain and an image can be decomposed into multiple scales and multiple orientations by using it. The Gabor function has been recognized as a very useful tool in texture analysis, due to its optimal localization properties in both spatial and frequency domain. Texture features based on Gabor wavelets transformation are found by the Gabor filtered image. Experiment results indicate the following: (1) the proposed algorithm based on random forest outperformed the-state-of-the-art in pigmented skin lesions detection (2) and the inclusion of Gabor wavelet transformation based texture features improved segmentation accuracy significantly.
Intelligent video surveillance is to analysis video or image sequences captured by a fixed or mobile surveillance camera, including moving object detection, segmentation and recognition. By using it, we can be notified immediately in an abnormal situation. Pedestrian detection plays an important role in an intelligent video surveillance system, and it is also a key technology in the field of intelligent vehicle. So pedestrian detection has very vital significance in traffic management optimization, security early warn and abnormal behavior detection. Generally, pedestrian detection can be summarized as: first to estimate moving areas; then to extract features of region of interest; finally to classify using a classifier. Redundant wavelet transform (RWT) overcomes the deficiency of shift variant of discrete wavelet transform, and it has better performance in motion estimation when compared to discrete wavelet transform. Addressing the problem of the detection of multi-pedestrian with different speed, we present an algorithm of pedestrian detection based on motion estimation using RWT, combining histogram of oriented gradients (HOG) and support vector machine (SVM). Firstly, three intensities of movement (IoM) are estimated using RWT and the corresponding areas are segmented. According to the different IoM, a region proposal (RP) is generated. Then, the features of a RP is extracted using HOG. Finally, the features are fed into a SVM trained by pedestrian databases and the final detection results are gained. Experiments show that the proposed algorithm can detect pedestrians accurately and efficiently.
With the broad attention of countries in the areas of sea transportation and trade safety, the requirements of efficiency and accuracy of moving ship tracking are becoming higher. Therefore, a systematic design of moving ship tracking onboard based on FPGA is proposed, which uses the Adaptive Inter Frame Difference (AIFD) method to track a ship with different speed. For the Frame Difference method (FD) is simple but the amount of computation is very large, it is suitable for the use of FPGA to implement in parallel. But Frame Intervals (FIs) of the traditional FD method are fixed, and in remote sensing images, a ship looks very small (depicted by only dozens of pixels) and moves slowly. By applying invariant FIs, the accuracy of FD for moving ship tracking is not satisfactory and the calculation is highly redundant. So we use the adaptation of FD based on adaptive extraction of key frames for moving ship tracking. A FPGA development board of Xilinx Kintex-7 series is used for simulation. The experiments show that compared with the traditional FD method, the proposed one can achieve higher accuracy of moving ship tracking, and can meet the requirement of real-time tracking in high image resolution.
Mean shift is a traditional moving target tracking algorithm, which has some deficiencies such as: A tracking window of a target needs to be initialed manually in the first frame; the window size cannot be adaptively changed according to a moving object in the process of tracking; if a target is obscured, it might be lost in the tracking window. In order to solve these problems, a method combining Kalman filter and Scale and Orientation Adaptive Mean Shift Tracking (SOAMST) is proposed. Firstly we use Kalman filter to locate a moving target at the beginning. Then the ratio of the first order moment to the zero order moment is used to estimate its center, and the second order center moment is used to estimate its size and orientation. Meanwhile, whether the target is obscured is determined by the Bhattacharyya coefficient based on the target model and a candidate one. A candidate model is more similar to the target and the estimation result of the target is more reliable when the Bhattacharyya coefficient is closer to 1. On the contrary, if the Bhattacharyya coefficient decreases to 0, the target will be lost for being totally obscured. If the target is partially obscured or not obscured, SOAMST is used directly to track the target; if totally obscured, Kalman filter is imposed to estimate the location of the target in the next frame before SOAMST. The experiments show that the proposed algorithm can track a moving target automatically at the initial frame without prior knowledge. It can also track a completely obscured target accurately by Kalman filtering based location estimation.
Environmental protection is one of the themes of today's world. The forest is a recycler of carbon dioxide and natural
oxygen bar. Protection of forests, monitoring of forest growth is long-term task of environmental protection. It is very
important to automatically statistic the forest coverage rate using optical remote sensing images and the computer, by
which we can timely understand the status of the forest of an area, and can be freed from tedious manual statistics.
Towards the problem of computational complexity of the global optimization using convexification, this paper proposes
a level set segmentation method based on Markov chain Monte Carlo (MCMC) sampling and applies it to forest
segmentation in remote sensing images. The presented method needs not to do any convexity transformation for the
energy functional of the goal, and uses MCMC sampling method with global optimization capability instead. The
possible local minima occurring by using gradient descent method is also avoided. There are three major contributions in
the paper. Firstly, by using MCMC sampling, the convexity of the energy functional is no longer necessary and global
optimization can still be achieved. Secondly, taking advantage of the data (texture) and knowledge (a priori color) to
guide the construction of Markov chain, the convergence rate of Markov chains is improved significantly. Finally, the
level set segmentation method by integrating a priori color and texture for forest is proposed. The experiments show that
our method can efficiently and accurately segment forest in remote sensing images.
Segmenting greenbelts quickly and accurately in remote sensing images is an economic and effective method for the statistics of green coverage rate (GCR). Towards the problem of over-reliance on priori knowledge of the traditional level set segmentation model based on max-flow/min-cut Graph Cut principle and weighted Total Variation (GCTV), this paper proposes a level set segmentation method of combining regional texture features and priori knowledge of color and applies it to greenbelt segmentation in urban remote sensing images. For the color of greenbelts is not reliable for segmentation, Gabor wavelet transform is used to extract image texture features. Then we integrate the extracted features into the GCTV model which contains only priori knowledge of color, and use both the prior knowledge and the targets’ texture to constrain the evolving of the level set which can solve the problem of over-reliance on priori knowledge. Meanwhile, the convexity of the corresponding energy functional is ensured by using relaxation and threshold method, and primal-dual algorithm with global relabeling is used to accelerate the evolution of the level set. The experiments show that our method can effectively reduce the dependence on priori knowledge of GCTV, and yields more accurate greenbelt segmentation results.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.