期刊文献+

基于卷积神经网络的PCB缺陷检测与识别算法 被引量:61

PCB defect detection and recognition algorithmbased on convolutional neural network
原文传递
导出
摘要 现有的印刷电路板(PCB)缺陷检测与识别算法大多都采取传统的图像处理和识别过程:即缺陷检测,特征提取和缺陷识别。由于电路板的复杂性,传统方法对于种类较多的缺陷很难达到精确分类,提出一种基于深度学习的PCB缺陷识别算法。首先对参考图像与待测图像进行差分操作找出PCB缺陷区域,然后针对缺陷区域,设计了包括2个卷积层、2个下采样层和4个全连接层的卷积神经网络模型。将PCB缺陷图像批量归一化,选取ReLU作为激活函数,Maxpooling作为下采样方法,并使用Softmax回归分类器训练并优化卷积神经网络。该方法分别与目前生产线上常用的基于方向梯度直方图、尺度不变特征变换特征和支持向量机结合的识别方法进行了比对,实验结果表明,该方法的正确识别显著提高,对于10类PCB缺陷可以得到96.67%的识别准确率,具有较好的应用前景。 Most of the existing PCB defect recognition algorithms adopt traditional image processing and recognition methods,i.e.defect detection,extraction feature and recognition process.Due to the complexity of the circuit board,it is difficult to achieve accurate classification for various defects.A PCB defect recognition algorithm based on deep learning is proposed.Firstly,the defect areas are found by XOR operation between the standard template and the tested template.Then,for the defect area,a convolutional neural network with 2 convolutional layers,2 down sampling layers and 4 fully connected layers is designed.The PCB defect pictures are batch normalized with ReLU is used as the activation function.Max pooling is used as the down sampling method and the Softmax regression classifier is used to train and optimize the convolutional neural network.The proposed method is compared with the recognition methods commonly used on the production line,i.e.,the direction gradient histogram,the scale invariant feature transform feature and the support vector machine.The experimental results show that the recognition rate based on deep learning significantly increase.The identification method can obtain 96.67%recognition accuracy for 10 types of PCB defects.
作者 王永利 曹江涛 姬晓飞 Wang Yongli;Cao Jiangtao;Ji Xiaofei(College of Information and Control Engineering,Liaoning Shihua University,Fushun113001,China;College of Automation,Shenyang Aerospace University,Shenyang110136,China)
出处 《电子测量与仪器学报》 CSCD 北大核心 2019年第8期78-84,共7页 Journal of Electronic Measurement and Instrumentation
基金 辽宁省科学事业公益研究基金(2016002006) 辽宁省自然科学基金(201602557) 辽宁省教育厅科学研究服务地方项目(L201708)资助
关键词 卷积神经网络 支持向量机 图像处理 深度学习 分类识别 convolutional neural network support vector machine machine learning deep learning classification and recognition
  • 相关文献

参考文献11

二级参考文献92

共引文献92

同被引文献456

引证文献61

二级引证文献375

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部