基于轻量级网络MobileNet V2的二极管玻壳缺陷识别  

Defect recognition of diode glass shells based on lightweight network MobileNet V2

作　　者：哈马友吉 任万春 张秤 张华[1,2] HAMA Youji;REN Wanchun;ZHANG Cheng;ZHANG Hua(School of Information Engineering,Southwest University of Science and Technology,Mianyang 621000,China;Key Laboratory of Robot Technology Used for Special Environment of Sichuan Province,Mianyang 621000,China)

机构地区：[1]西南科技大学信息工程学院,四川绵阳621000 [2]特殊环境机器人技术四川省重点实验室,四川绵阳621000

出　　处：《传感器与微系统》2022年第4期153-155,160,共4页Transducer and Microsystem Technologies

基　　金：国家“十三五”核能开发科研项目(17ZG6102)

摘　　要：针对目前二极管玻壳缺陷检测中存在的自动化程度低、效率低、错误率高和成本高等问题,提出了一种基于MobileNet V2卷积神经网络模型的图像识别算法,实现对二极管玻壳缺陷的准确识别。实验结果表明:该模型能够准确识别二极管玻壳缺陷,网络模型收敛后对玻壳缺陷的总体识别准确率达到93.3%,同时具备了检测速度快(图像识别速度不小于5张/s),模型体量小(不大于26.8 M)等特点,具有一定的实际工业应用价值。In order to solve the problems of low automation,low efficiency,high error rate,and high cost in the detection of diode glass shell defects,an image recognition algorithm based on MobileNet V2 convolution neural network model is proposed to realize the accurate recognition of diode glass shell defects.The experimental results show that the model can accurately identify the defects.After the convergence of the network model,the total recognition accuracy of the glass shell defect can reach 93.3%.At the same time,it has the characteristics of fast detection speed,that is image recognition speed is not less than 5 pieces/s,small model volume,which is not more than 26.8 M and so on.It has a certain value of practical industrial applications.

关 键 词：玻壳 卷积神经网络 缺陷识别 MobileNet V2网络 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]