计算机科学
萃取(化学)
比例(比率)
遥感
人工智能
环境科学
计算机视觉
地质学
化学
地图学
地理
色谱法
作者
Yiyang Wang,Shuhao Yang,Lina Yang,Xuran Pan
出处
期刊:
[SPIE]
日期:2025-06-03
卷期号:: 3-3
摘要
Building extraction from high-resolution remote sensing images is crucial for urban planning and environmental monitoring. Buildings in remote sensing images often exhibit complex geometric shapes and appearances, surrounded by intricate landscapes, which pose challenges for building extraction tasks. In this paper, we propose the Enhanced U-Net with scale-sensitive attention mechanisms (SSAU-Net) to enhance the accuracy of building extraction. The SSAU-Net is constructed based on a classical encoder-decoder architecture, integrating spatial and channel attention mechanisms to improve the model's feature extraction and selection capabilities. Particularly, the attention mechanisms are designed in a scale-sensitive manner: high-level features are prioritized through channel attention mechanisms, while low-level features are emphasized through spatial attention mechanisms. This approach aims to effectively enhance beneficial information and suppress interfering information, thereby enhancing the overall performance of the model. Experimental results on the Inria dataset demonstrate that our method achieves superior accuracy in building extraction.
科研通智能强力驱动
Strongly Powered by AbleSci AI