基于YOLO V5的改装车辆识别技术研究

Research on Modified Vehicle Identification Technology Based on YOLO V5

随着我国汽车保有量持续上升,部分用户追求个性化对车辆进行改装,部分货车司机为增加运输量对车辆进行改装等,给道路交通安全带来巨大影响。为解决非法改装车辆识别问题,文章对改装车辆识别技术进行了研究。本研究以YOLO V5为模型基础,将主干特征提取网络替换为高度灵活且易于实施的MobileNet V2,并将强化特征提取网络中常规卷积操作全部替换为深度可分离卷积,最终得到一个检测效率高、计算需求小、检测速度快的非法改装车辆检测模型。

With the continuous rise of car ownership in China,some users pursue personalization to modify their vehicles,and some truck drivers modify their vehicles to increase the transportation volume,etc.,which brings a great impact on road traffic safety,and in order to solve the problem of illegal modified vehicle identification,the article researches the modified vehicle identification technology.This study takes YOLO V5 as the model base,replaces the backbone feature extraction network with highly flexible and easy-to-implement MobileNet V2,and replaces all the conventional convolution operations in the reinforcement feature extraction network with depth separable convolution.Finally,a detection model for illegally modified vehicles with high detection efficiency,small computational requirements and fast detection speed is obtained.