Semantic Segmentation Algorithm for Remote Sensing Images Based on Attention Mechanism

Abstract

This paper propose to combine the attention mechanism with the U-Net model to improve the performance and accuracy of semantic segmentation tasks. The attention mechanism can better focus on task-relevant regions and features, thereby enhancing the model's perceptual and expressive capabilities. By introducing attention modules into the U-Net semantic segmentation model, fine adjustments and attention to image features are achieved, thereby improving the effectiveness of semantic segmentation. Experimental results also demonstrate that the algorithm proposed in this paper greatly improves the accuracy of semantic segmentation for remote sensing images.