Improved YOLOv5 Algorithm for Remote Sensing Image Detection

To address the issue of low object detection accuracy in remote sensing images, this paper proposes an improved algorithm based on YOLOv5. The improved algorithm enhances the focus on important information and suppresses background noise interference by introducing a multi-scale channel attention feature extraction module in the backbone network. In the feature fusion network, an attention feature fusion module is introduced to adaptively fuse the detailed information of shallow features and the semantic information of deep features, thereby improving the detection capability for multi-scale objects. Finally, a polarized self-attention module is incorporated into the detection output layer to capture the dependency relationships of distant pixels, thus obtaining more accurate pixel-wise semantic information and improving the algorithm's target localization capability. Experimental results show that the proposed algorithm achieves higher detection accuracy on remote sensing images than the original YOLOv5 and other algorithms.