Fast vehicle detection method based on improved YOLOv5s

In order to quickly and accurately identify vehicles and pedestrians in driving scenarios, this paper proposes a fast vehicle detection algorithm SMA-YOLOv5s based on the YOLOv5s (You Only Look Once) algorithm, which introduces the lightweight backbone network ShuffleNetv2 and multi-attention mechanism. Firstly, the original backbone network of YOLOv5s is replaced with a lightweight backbone network ShuffleNetv2 to reduce the computational complexity of the model and improve the real-time performance of the model. Experimental results show that compared with the original algorithm, SMA-YOLOv5s, a lightweight model with multi-attention mechanism, reduces the floating-point arithmetic by 55.70%, loses only 1% in accuracy, only 2.6% in mAP, and improves FPS by 26.60%. With less loss of accuracy, there is less computational complexity, faster detection, and better real-time performance, making it more suitable for deployment on mobile or embedded devices with limited performance.