@article{YANG2025, 
author = {Jun YANG and Jinying ZHANG},
title = {U-shaped semantic segmentation network of high-resolution remote sensing images embedded with self-attention mechanism},
year = {2025},
journal = {Journal of Beijing University of Aeronautics and Astronautics},
volume = {51},
number = {5},
pages = {1514-1527},
keywords = {high-resolution remote sensing images, semantic segmentation, deep learning, self-attention mechanism, feature fusion},
url = {https://www.sciopen.com/article/10.13700/j.bh.1001-5965.2023.0269},
doi = {10.13700/j.bh.1001-5965.2023.0269},
abstract = {To reduce the difficulty of extracting small object features from high-resolution remote sensing images, a dual-encoder feature fusion network model based on convolution structure and self-attention mechanism was proposed, which was suitable for semantic segmentation of high-resolution remote sensing images. Firstly, a dual-encoder structure was designed to extract global and local detail information of remote sensing images and improve the segmentation accuracy of small object features. Secondly, a feature aggregation module was used to aggregate feature information at different stages, so as to embed more global contextual information. Finally, an edge thinning loss module was used to improve the recognition ability of the model for object edge information. The  mF1 average value of the segmentation results on the ISPRS Vaihingen and Potsdam datasets achieved 91.28% and 93.16%, respectively. Compared with the current mainstream algorithms, the segmentation accuracy of small objects like cars and the overall segmentation accuracy were improved. The proposed model solves the problem of inaccurate segmentation of small objects and edge information in the semantic segmentation of high-resolution remote sensing images to a certain extent.}
}