基于机器学习的电动汽车无线充电异物目标检测方法  

ML⁃based foreign object detection method for electric vehicle wireless charging

作　　者：钱强 陈海[1] 郑义[1] 闫丽华 梁文熙 吴开荣 QIAN Qiang;CHEN Hai;ZHENG Yi;YAN Lihua;LIANG Wenxi;WU Kairong(Beijing Normal University at Zhuhai,Zhuhai 519087,China)

机构地区：[1]北京师范大学珠海分校,广东珠海519087

出　　处：《现代电子技术》2023年第13期43-48,共6页Modern Electronics Technique

基　　金：广东省教育厅科研项目(2020KTSCX176,2021KTSCX181)。

摘　　要：生物体和金属异物存在于发射线圈附近时,影响电动汽车无线充电系统的传输功率、传输效率,造成安全事故,线圈电路检测、超声波/雷达检测、图像特征检测等目前常见检测方法存在不足。文中研究并设计了一套基于机器学习的异物目标检测算法,并应用在电动汽车无线充电安全运行中,可及时对事故做出预警和处理。该系统由图像采集模块、无线传输模块、云平台和服务器四大部分组成,通过摄像头对充电过程中可能混入的猫、狗生物体,以及易拉罐、螺丝钉、硬币金属异物进行图像采集,无线传输到云平台服务器上,利用深度学习的YOLOv5训练模型检测区域内是否存在异物,检测结果发送给充电控制器和用户。实验结果表明:YOLOv5在测试集上经过1 000次迭代训练后,检测精度达到0.855 9,召回率达到0.998 1,速度达到62 FPS;在实际复杂的充电环境下,对不同光照条件、不同停车位地面、不同尺寸的5个类别异物进行推理测试,具有较高的检测精度和适应性,满足目标检测的高效性要求,为实现车辆充电安全提供了重要保障。When living organisms and metal foreign objects exist near the transmitting coil,the transmission power and efficiency of the wireless charging system of electric vehicles will be affected,resulting in safety accident.However,the commonly⁃used detection methods such as coil circuit detection,ultrasonic/radar detection,image feature detection and other common methods are insufficient.A foreign object detection algorithm based on machine learning(ML)is studied and designed,and is applied to the safe operation of wireless charging of electric vehicle to give early warning and deal with accidents in time.The system consists of four parts:image acquisition module,wireless transmission module,cloud platform and server.The images of cat and dog organisms,as well as metal foreign bodies of cans,screws and coins that may be mixed in the charging process are collected by the camera,and transmitted wirelessly to the cloud platform server.The YOLOv5 training model of deep learning is used to detect whether there are foreign bodies in the area,and the detected results are sent to the charging controller and the user.The experimental results show that the detection precision reaches 0.8559,the recall rate reaches 0.9981,and the speed reaches 62 FPS after training of 1000 iterations of YOLOv5 on the test set.In the actual complex charging environment,reasoning tests were carried out on five categories of foreign bodies with different sizes in different lighting conditions and,different parking spaces on the ground.The result shows that the method has high detection accuracy and adaptability,meets the high efficiency requirements of target detection,and provides an important guarantee for the realization of vehicle charging safety.

关 键 词：电动汽车 无线充电 充电安全 异物检测 目标检测 机器学习 YOLOv5模型 

分 类 号：TN02-34[电子电信—物理电子学]