An Improved Multi-target Detection Algorithm Based on YOLOv5 Network

Real-time detection of traffic participants such as vehicles and pedestrians is an important part of information interaction in today's society, but the low accuracy of multi-target detection in complex weather conditions and dense intersections is still a challenging problem. Therefore, this paper introduces a multi-object identification algorithm utilizing deep neural networks. Firstly, the SPPF component in the YOLOv5 network is substituted with SimSPPF, resulting in an expanded receptive field. Then, the SimAM attention mechanism module is attached to the backbone network of the YOLOv5 to enhance the extraction ability of the model and reduce the false detection rate. Finally, replace the loss function of YOLOv5 with the Varifocal Loss function, which can improve the model's detection ability for targets of different scales and sizes, making it more robust.