UAV image detection based on multi-scale spatial attention mechanism with hybrid dilated convolution

Unmanned aerial vehicles (UAVs) have been widely used in many fields. Recently, the abuse of UAVs has become a serious threat to public privacy and security. Therefore, directly and correctly detecting UAVs from various scene images is necessary and important. However, UAV detection is a challenging task because of the significant scale differences and complex image backgrounds. In this paper, we propose a UAV detection method based on a multi-scale spatial attention mechanism with hybrid dilated convolution to relieve these problems. Firstly, we introduce a spatial attention mechanism, which extracts the context feature to help the feature extraction network pay more attention to the target region from a complex background. Secondly, a multi-scale hybrid convolution is involved in the proposed attention mechanism to recover the respective field and improve feature capture ability for different scale UAVs. Thirdly, we insert the proposed attention mechanism in the YOLOv8 model and use a layer connection structure to predict the detection result based on multi-level feature fusion. Experimental results show that the proposed method is effective in improving the UAV detection preference under complex image backgrounds compared with the existing detection methods, and further increases the detection precision compared to the YOLOv8 model.