DMS-YOLOv5: A Decoupled Multi-Scale YOLOv5 Method for Small Object Detection

Small objects detection is a challenging task in computer vision due to the limited semantic information that can be extracted and susceptibility background interference. In this paper, we propose decoupled multi-scale small object algorithm named DMS-YOLOv5. The incorporates receptive field module into feature extraction network for better focus on low-resolution objects. coordinate attention mechanism, which combines spatial channel information, introduced reduce interference from enhance network’s information. A layer tailored small-sized added compensate loss of multiple downsampling operations, greatly improving capability Next, head branch processing classification bounding box regression tasks. Finally, function improved alleviate missed problems caused by concentration mutual occlusion between method achieved mean average precision improvement 12.1% VisDrone2019-DET dataset compared original method. comparison experiments with similar methods, our proposed also demonstrated good performance, validating its effectiveness.