U-net Remote Sensing Image Segmentation Algorithm Based on Attention Mechanism Optimization
Shuwan Feng, Ruixiang Song, Sihan Yang, Desheng Shi
Abstract
In recent years, U-Net network based on deep learning has achieved remarkable results in the field of image segmentation, but its performance still needs to be improved in remote sensing image segmentation tasks. In this paper, by analyzing the shortcomings of U-Net network in remote sensing image segmentation, we propose a method of introducing SimAM and SE attention mechanism in U-Net network to improve its performance. Through experimental validation on remote sensing image datasets, the results show that adding the attention mechanisms SimAM and SE modules improves MIoU by 17.41% and 13.23%, and Mpa and Accuracy by 16.88% and 13.045, 13.98% and 10.67%, respectively, and the U-Net that integrates the SimAM and SE attention mechanisms network achieves higher accuracy and better visual effect in remote sensing image segmentation task with strong generalization ability and robustness. The research in this paper provides new ideas and methods for the development of remote sensing image segmentation technology, and has important reference value for the selection and improvement of future remote sensing image segmentation algorithms.