Small Object Detection Algorithm Combining Coordinate Attention Mechanism and P2-BiFPN Structure

Aiming at the difficulty of small object detection, a small object detection model combining coordinated attention mechanism and P2-BiFPN (P2 Bidirectional Feature Pyramid Network) structure is constructed based on YOLOv5. Firstly, we introduce the coordinated attention mechanism into the residual units of the backbone network to achieve more accurate localization of small objects. Secondly, to reduce the number of model parameters, we decompose the square convolution in the residual unit into parallel asymmetric convolutions. Then, the P2-BiFPN feature fusion network was constructed to enrich the information of small objects, so as to improve the small objects detection accuracy. Finally, we train and test the model on the WiderPerson dataset. The experimental results shows that compared with YOLOv5, our small object detection model has a 1.7% improvement in mAP and a 5.66 m reduction in the amount of parameters, with better detection performance for small-object pedestrians.