Incremental supervision-driven network for optimized contour segmentation of buildings in remote sensing

Mingzhu Li; Hanchao Zhang; Yizhi Hu; Chang Dong; Minghui Hao

doi:10.1117/12.3035539

10 July 2024 Incremental supervision-driven network for optimized contour segmentation of buildings in remote sensing

Mingzhu Li, Hanchao Zhang, Yizhi Hu, Chang Dong, Minghui Hao

Proceedings Volume 13223, Fifth International Conference on Geology, Mapping, and Remote Sensing (ICGMRS 2024); 1322327 (2024) https://doi.org/10.1117/12.3035539
Event: 2024 5th International Conference on Geology, Mapping and Remote Sensing (ICGMRS 2024), 2024, Wuhan, China

Abstract

The high cost and complexity of obtaining pixel-level labels have hindered the development of the deep learning semantic segmentation field. To tackle the label refinement problem in remote sensing, we propose an Incremental Supervision- Driven Network (ISD-Net). In the ISD-Net, we adopt an incremental supervision strategy to continuously guide the model in learning more accurate semantic information to enhance model performance and robustness. A non-parametric background attention-guided module (N-BAG) is designed, which is combined with a conditional random field (CRF) module to better distinguishes buildings from the background. The level set loss function is used to enhance the perception of building features in coarse label regions, and effectively separating noise in pixel-level box-label. We conduct experiments on the WHU dataset to simulate the coarse label scenario. The experimental results demonstrate that the proposed algorithm achieves a maximum improvement of 17.15% and 25.68% in mIoU and F1-score, respectively, compared to the original coarse labels, respectively, proving the effectiveness of this approach in separating noise and identifying building edges. This research has a broad application prospect in the direction of building contour refinement in remote sensing images.

(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Mingzhu Li, Hanchao Zhang, Yizhi Hu, Chang Dong, and Minghui Hao "Incremental supervision-driven network for optimized contour segmentation of buildings in remote sensing", Proc. SPIE 13223, Fifth International Conference on Geology, Mapping, and Remote Sensing (ICGMRS 2024), 1322327 (10 July 2024); https://doi.org/10.1117/12.3035539

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

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.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available

Members: $17.00

Non-members: $21.00 ADD TO CART

PROCEEDINGS
8 PAGES

DOWNLOAD PAPER SAVE TO MY LIBRARY

GET CITATION

RIGHTS & PERMISSIONS

Get copyright permission Get copyright permission on Copyright Marketplace

KEYWORDS

Buildings

Remote sensing

Image segmentation

Feature extraction

RGB color model

Ablation

Visualization

Keywords/Phrases

Search In:

Publication Years