基于深度学习的指针缺陷检测研究  

Research on Pointer Defect Detection Based on Deep Learning

作　　者：黄汉成 赵卫东 夏冰[2] HUANG Hancheng;ZHAOWeidong;XIA Bing(School of Electrical Information and Engineering,Anhui University of Technology,Ma’anshan 243032,China;Equipment Management Department Ma’anshan Iron and Steel Co.Ltd.,Ma’anshan 243000,China)

机构地区：[1]安徽工业大学电气与信息工程学院,安徽马鞍山243032 [2]马鞍山钢铁股份有限公司设备管理部,安徽马鞍山243000

出　　处：《安徽工业大学学报（自然科学版）》2019年第4期375-381,共7页Journal of Anhui University of Technology（Natural Science）

基　　金：国家自然科学基金项目(61601004)。

摘　　要：针对汽车仪表盘烫印指针缺陷种类多、体积小、检测难度较高等特点,设计一种融合SSD(single shot-multibox detector)与RFCN(region-based fully convolutional networks)的指针缺陷检测算法。首先,在现场使用Basler相机采集有缺陷指针的照片,并对其进行标注,使用MobileNet网络作为SSD算法的特征提取网络对烫印和未烫印的指针进行分类;然后,对烫印指针的图片进行裁剪,将裁剪的烫印指针图片进行缺陷标注;最后,结合ResNet网络改进RFCN的特征提取网络,对4种典型缺陷进行检测,且在Ubuntu16.04系统和Tensorflow-gpu深度学习框架下搭建软件系统实验测试。结果表明,该检测算法可准确检测出指针表面的微小缺陷,有效提高产品质量和生产效率。Aiming at the characteristics of multiple types, small size and high detection difficulty of the stamping pointer defect of automobile dashboard, a pointer defect detection algorithm with SSD(single shot-multiboxp detector) and RFCN(region-based fully convolutional networks) was designed. First, Basler camera was used on site to collect photos of defective pointers and annotate them, and mobileNet network was used as the feature extraction network of SSD algorithm to classify stamping and no stamping pointers. Then, the image of stamping pointer was clipped, and the defect of clipped pointer image was marked. Finally, the feature extraction network of RFCN was improved based on ResNet network to detect four typical defects. The software system was built under the Ubuntu16.04 system and Tensorflow-gpu deep learning framework for experimental testing. The results show that the algorithm can accurately detect the tiny defects on the pointer surface and effectively improve product quality and production efficiency.

关 键 词：指针 缺陷检测 SSD RFCN MobileNet 特征提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]