基于改进YOLO-V3的汽车刹车衬芯缺陷检测  

Defect Detection of Automobile Brake Liner Based on Improved YOLO-V3

作　　者：吴皓 沙玲 Wu Hao;Sha Ling(College of Mechanical and Automotive Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)

机构地区：[1]上海工程技术大学机械与汽车工程学院,上海市201620

出　　处：《农业装备与车辆工程》2022年第5期120-124,共5页Agricultural Equipment & Vehicle Engineering

摘　　要：目前汽车刹车衬芯的法兰面缺陷主要通过人工检测,针对效率低、准确率不高的问题,提出一种基于YOLO-V3改进的检测算法。首先根据检测目标较小的场景对YOLO-V3结构进行改进,即为FPN结构增加一层特征图;其次,以深度可分离卷积代替普通卷积,以加快预测速度;最后,对数据集进行预处理并进行扩充。对改进后的YOLO-V3与一些现存的主流检测算法进行对比实验。结果表明,改进后的YOLO-V3平均准确率达到了90%,检测速度达到42 ms/张,对小目标的检测效果有所提升,满足了工业现场对衬芯检测的要求。The flange defects of automobile brake lining core are mainly detected by manual inspection,which cause the problems of low efficiency and low accuracy,and an improved detection algorithm based on YOLO-V3 is proposed.Firstly,according to the scene with small detection target,the structure of YOLO-V3 is improved,that is,a layer of feature graph is added to the FPN structure.Secondly,the depth separable convolution is used to replace the ordinary convolution to speed up the prediction.Finally,the data set is preprocessed and expanded.The improved YOLO-V3 is compared with some existing mainstream detection algorithms.The results show that the average accuracy of the improved YOLO-V3 reaches 90%,the detection speed reaches 42 ms/sheet,and the detection effect of small target is improved,which meets the requirements of industrial field for core detection.

关 键 词：刹车衬芯 缺陷检测 改进YOLO-V3 

分 类 号：U463.55[机械工程—车辆工程]