摘要
针对CCD轮对图像背景噪声干扰较多,难以识别踏面损伤区域的难题,提出了一种基于Canny-YOLOv3的踏面损伤检测方法。采用Canny边缘检测算法对轮对踏面进行边缘检测,精准识别和分割出轮对踏面区域,并统一踏面图像的尺寸大小;又采用深度学习目标检测算法--YOLOv3检测出踏面图像中的损伤区域,从而完成轮对踏面的损伤检测。仿真结果表明,在相似图片干扰较大的背景下,该算法能够较准确地检测出损伤的位置和区域,且IoU值设定为0.5时,AP值可达83.19%。
A tread damage detection algorithm on Canny-YOLOv3 is presented for the problems such as background noise interference,difficulty of tread damage area identification in CCD wheelset image detection.Firstly,Canny edge detection algorithm is used to detect the edge of wheelset tread,accurately identify and segment the wheelset tread area,and unify the size of tread image.Then,deep learning target detection algorithm-YOLOv3 is adopted to extract the damage area in the tread image,so as to complete the damage detection of wheelset tread.The simulation results show that the algorithm can detect the location and area of damage accurately under the background of large disturbance of similar images,and when the IoU threshold is set at 0.5, the AP value can reach 83.19%.
作者
何静
余昊宇
张昌凡
刘建华
罗学明
He Jing;Yu Haoyu;Zhang Changfan;Liu Jianhua;Luo Xueming(Technology College of Electrical and Information Engineering,Hunan University of Technology,Zhuzhou 412007,China)
出处
《电子测量与仪器学报》
CSCD
北大核心
2019年第12期25-30,共6页
Journal of Electronic Measurement and Instrumentation
基金
国家自然科学基金(61773159)
湖南省自然科学基金(2019JJ40067,2018JJ4066)资助项目。
