Electric Bicycle Detection Based on Improved YOLOv5
Caifeng Zhang, Aimin Xiong, Xusong Luo, Chaowei Zhou, Jingfeng Liang
Abstract
Electric bicycle has become one of the main travel tools for residents because of its convenience, economy . However, on the occasion of the prosperity and development of the electric bicycle market, electric bicycles are frequently parked and placed disorderly, entering elevators and charging indoors in violation of regulations, and the number of casualties caused by electric bicycle fires has increased significantly. In order to improve the management efficiency of electric bicycle parking and charging, this paper presents an intelligent electric bicycle detection algorithm based on improved YOLOv5. We use GhostNet to improve the backbone network of YOLOv5s to make the model lighter. In order to make the prediction box more consistent with the real box, the loss function is changed to CIOU loss. And use QFocal Loss to learn better classification score and positioning quality. The experimental results show that the improved YOLOv5s algorithm proposed in this paper reduces the number of parameters by 23.6%, and the detection map of electric bicycle can reach 93.3%. Therefore, the algorithm can not only ensure the accuracy, but also save memory and computing resources, make the network model lighter and more suitable for the deployment of mobile terminals.