Res50-SimAM-ASPP-Unet: A Semantic Segmentation Model for High-Resolution Remote Sensing Images

High-resolution remote sensing images contain intricate details and complex backgrounds, presenting challenges for traditional segmentation methods, which often struggle with accurate classification and contextual understanding. To address these issues, this study introduces the Res50-SimAM-ASPP-Unet model, a semantic segmentation approach for high-resolution remote sensing image processing tasks. The model integrates ResNet50 as the encoding layer of Unet for robust feature extraction, adds the SimAM attention mechanism to selectively enhance relevant details, and incorporates the ASPP module in the decoding layer to capture multi-scale contextual information. The methodology part analyzes the common ResNet model, the attention mechanism module, and the multi-scale feature extraction module, respectively, and then designs experiments to show the necessity and optimal position of adding Res50, SimAM, and ASPP. Comparative experiments on the LandCover.ai dataset demonstrate that the proposed model significantly outperforms common semantic segmentation networks, achieving a MIoU of 81.1%, MPA of 88.2%, Accuracy of 95.1%, Precision of 92.65%, and an F1 score of 90.45%. These results highlight the model's effectiveness in delivering high accuracy and adaptability across diverse remote sensing environments, establishing it as a valuable tool for applications requiring precise and scalable image segmentation.