摘要
为实现自动检测电缆成品上的多个部件引起的产品缺陷,如部件次序错误、部件朝向错误等,设计了一种适用于嵌入式平台的电缆部件缺陷检测方法。该方法以YOLOv5目标检测神经网络作为基础,将其改造为Anchor-free网络使其在嵌入式设备有更快的运行速度,并使用迁移学习及数据增强的方式加快网络训练速度并增加精度。根据实验结果,将该方法应用于NVIDIA Jetson Nano平台,单张图片的平均检测时间仅需76 ms,缺陷检测准确率达99%以上,其中缺陷检出率为100%,可满足工业生产的需求。
In order to automatically detect product defects caused by multiple components on the finished cable product,such as component sequence error,component orientation error,a cable component defect detection method suitable for embedded platform is designed.This method is based on YOLOv5 object detection neural network,which is transformed into Anchor-free network to make it run faster in embedded equipment,and use the methods of transfer learning and data enhancement to speed up the network training speed and increase the precision.According to the experimental results,the method is applied to the NVIDIA Jetson Nano platform,the average detection time of a single picture is only 76ms,and the accuracy of defect detection is more than 99%,the defect detection rate is 100%,which can meet the needs of industrial production.
作者
王庭琛
王宜怀
陈瑞雪
WANG Tingchen;WANG Yihuai;CHEN Ruixue(School of Computer Science and Technology,Soochow University,Suzhou 215006,China)
出处
《传感器与微系统》
CSCD
北大核心
2023年第8期123-126,共4页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(61672369)。