SPD-CBAM: Small Target Detection in UAV Images

To address the issues of low pixel ratio of extremely small targets and the susceptibility to missed detections and false detections in traditional detection algorithms under complex environmental conditions during drone aerial surveillance, this paper proposes a new detection scheme based on YOLOv8s.This study first addresses the issue of small target feature loss caused by stride convolution during downsampling by introducing a spatial-to-depth (SPD) transformation mechanism to reconstruct the backbone network. By changing the feature mapping method, key detail information is preserved. In the feature fusion stage, a convolutional block attention module (CBAM) is specifically embedded to enhance the target response weights from both channel and spatial dimensions, thereby reducing the interference of complex background noise. Experimental results show that the improved model achieves an mAP@50 of 46.4%, a significant improvement of 15.9% compared to the original algorithm, and a significant increase in recall of 29.9%.