基于计算机视觉的电子元器件表面缺陷检测  

The Surface Defect Detection of Electronic Components Based on Computer Vision

作　　者：刘竞升 邱宝军[1] 吕宏峰[1] LIU Jingsheng;QIU Baojun;LV Hongfeng(CEPREI,Guangzhou 511370,China)

机构地区：[1]工业和信息化部电子第五研究所,广东广州511370

出　　处：《电子产品可靠性与环境试验》2023年第3期91-96,共6页Electronic Product Reliability and Environmental Testing

基　　金：广州市科技计划项目-基础研究计划-基础与应用基础研究项目(202102080285)资助。

摘　　要：电子元器件表面缺陷(如划痕、气泡和毛刺等)会在一定程度上影响其质量。现阶段主要是通过人工的方法对表面缺陷进行检测。随着人工智能技术的成熟,通过计算机视觉的方法对电子元器件的表面缺陷进行检测已经成为可能。因此,提出了一种基于计算机视觉的电子元器件表面缺陷检测方法。首先,搭建了用于检测的硬件平台,并建立了电子元器件外观图像的样本库;其次,分析了图像滤波和图像初步筛选的方法;然后,分析了卷积神经网络的基本原理,并对比各种成熟神经网络模型的识别准确率;最后,选择VGGNet19网络来对电子元器件的外观图像进行分类识别,识别的准确率高达90%,从而证明了所提出的方法的有效性,具有一定的实用价值。The surface defects(such as scratches,bubbles and burrs)of electronic component will affect their quality to some extent.At present,surface defects are mainly detected by artificial methods.With the mature of artificial intelligence,it has become possible to detect surface defects of electronic components by means of computer vision.Therefore,a surface defect detection method for electronic components based on computer vision is proposed.Firstly,a hardware platform for detection is built,and a sample library of appearance images of electronic components is established.Next,the methods of image filtering and image preliminary screening are analyzed.Then,the basic principle of convolutional neural network is analyzed,and the recognition accuracy of various mature neural network models is compared.Finally,VGGNet19 network is selected to classify and recognize the appearance images of electronic components.The recognition accuracy is as high as 90%,which proves the feasibility of the proposed method,and the method has certain practical value.

关 键 词：电子元器件 表面缺陷 图像滤波 卷积神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]