Remote Sensing Image Object Detection Method Integrating Spatial Coordinate Information

In response to the shortcomings of existing remote sensing image object detection methods in dealing with complex spatial layouts and multi-scale targets, this paper proposes a remote sensing image object detection method that integrates spatial coordinate information. This method ingeniously incorporates spatial coordinate information onto the foundation of the YOLOv7 detection network. In the feature extraction section, a module for ELEN-Co has been designed, which employs a spatial coordinate attention mechanism (Coordinate Attention) to dynamically emphasize key areas in remote sensing images. In the multi-scale feature fusion stage and detection head section, spatial coordinate convolution (CoordConv) is further introduced to enhance the feature extraction network's ability to express and model target position and scale information in complex remote sensing scenes. Finally, the model is optimized using the WIOU loss function. Experiments conducted on the RSOD and SSDD datasets demonstrate that the proposed method yields an improvement of 4.7% and 1.4% in mAP compared to the baseline, respectively. The proposed method not only achieves significant accuracy improvement in complex remote sensing image object detection tasks, but also demonstrates the effectiveness of integrating spatial coordinate information into detection networks.