Remote Sensing Target Detection Method Based on Deformable Attention Mechanism and Adaptive Feature Fusion

To address the problem of information aliasing that may occur in remote sensing images with complex backgrounds and multi-scale features fusion, a remote sensing target detection method integrating the deformation attention mechanism and adaptive feature fusion is proposed. Firstly, the deformation attention module is introduced, and attention maps are generated from feature maps with multi-scale deformable receptive fields, which make the method adapt to different remote sensing targets better and reduce the interference of complex backgrounds on remote sensing object detection. Secondly, the adaptive feature fusion module that adaptively learns spatial weights at each scale is adopted. In this way, feature maps are fused to combine the relationships between different feature layers better, highlight the unique information of each scale, and reduce feature confusion during feature fusion. The experimental results on the Optical Remote Sensing Image (DIOR) and Remote Sensing Object Detection (RSOD) data sets show that the algorithm in this paper has higher detection accuracy and speed compared to other algorithms.