Research on Semantic Segmentation of Remote Sensing Images Based on Improved U-net

This study explores the important role of attention mechanism in enhancing the performance of U-Net model for remote sensing image segmentation. By introducing ECA (Efficient Channel Attention) and GACM (Global Attention Context Module), the model performance is significantly improved, as evidenced by the significant increase in MIoU, accuracy and F1 scores.The ECA module improves the MIoU by 6.51 percentage points, the accuracy by 11.84 percentage points and the F1 scores by 14.03 percentage points, respectively, while the use of the GACM module further improved MIoU by 9.11 percentage points, accuracy by 16.03 percentage points and F1 score by 17.73 percentage points. The strategy of combining ECA with GACM results in the improvement of MIoU, accuracy and F1 value by 22.21, 22.84 and 25.71 percentage points, respectively, which proves the complementary nature of the two mechanisms and the potential of the present method in improving the accuracy and robustness of the model, which is of significant practical application value and provides an effective performance enhancement scheme for remote sensing image segmentation.