A land parcel segmentation method for remote-sensing images based on an improved U-Net network

Abstract Land parcel segmentation in remote-sensing images aims to achieve pixel-level content analysis of the images, extracting and classifying categories of interest within the remote-sensing data. It holds significant practical value in areas such as urban and rural planning, flood control, and disaster relief, attracting widespread attention in the industry. With the continuous development of remote-sensing technology, deep learning has demonstrated substantial potential in the analysis of remote-sensing images. In this study, leveraging deep learning techniques to address the challenge of land parcel segmentation in remote-sensing images, we employed the U-Net architecture and further enhanced it by incorporating channel and spatial attention mechanisms. This improvement aims to enhance the model’s ability to identify complex land parcel boundaries and structures, ultimately achieving more accurate land parcel segmentation.