基于改进YOLOv5的车辆多属性识别

Vehicle Multi-attribute Recognition Based on Improved YOLOv5

车辆的多属性识别可以用于交通监控和智能卡口,对交通管理意义重大,为了提高在交通监控场景下车辆检测定位的准确性和可靠性,提出一种在YOLOv5算法的基础上加入注意力机制模块的方法,以增强网络提取特征的能力,将车辆的品牌、颜色和车型三种属性进行融合训练,充分考虑到属性之间的相关性。数据集使用网络公开的Cars Dataset,并对数据集进行扩充和预处理操作。实验结果表明,基于改进的YOLOv5模型,识别精度达到95.71%,比原始的YOLOv5模型检测精度提高2.61%,在保证检测速度的同时,有更高的识别准确率,可以满足视频实时性的要求。

Multi-attribute recognition of vehicles can be used for traffic monitoring and smart bayonet,which is of great signif-icance to traffic management.In order to improve the accuracy and reliability of vehicle detection and positioning in traffic monitor-ing scenarios,an attention mechanism is proposed based on the YOLOv5 algorithm.The modular approach enhances the ability of the network to extract features,and trains the three attributes of the vehicle's brand,color and car model,taking full account of the correlation between the attributes.The dataset uses the Cars Dataset publicly available on the Internet,and the dataset is expanded and preprocessed.Experimental results show that based on the improved YOLOv5 model,the recognition accuracy reaches 95.71%,which is 2.61%higher than the detection accuracy of the original YOLOv5 model.While ensuring the detection speed,it has higher recognition accuracy and can meet the requirements of video real-time.