摘要
针对公司现有印制电路板(PCB)⁃自动化光学检测(AOI)设备不易做真假点检测的弊端,提出一种基于深度学习算法库Pytorch对PCB大板表面进行真假点检测的视觉人工智能(AI)系统。运用图像处理中的形态学算法对表面内外孔边界交叉、内孔腐蚀不清、外孔渗铜等缺陷进行检测。通过增加padding、改进优化器、改变激活函数、引入注意力机制4个方案,并对近10万个成品完成数据收集与检测,该系统的平均精度由95.678%提升到98.998%,提高了3.32个百分点,精度较高,运行稳定。
The company's existing automated optical inspection(AOI)equipment for printed circuit board(PCB)is not easy to detect true and false points.In view of this problem,method based on the deep learning algorithm library Pytorch is proposed to detect the large PCB board,with a visual artificial intelligence(AI)system that detects true and false points on the surface.The morphological algorithm in image processing is used to detect surface defects such as intersection of inner and outer hole boundaries,unclear corrosion of inner holes,and copper seepage in outer holes.By adding padding,improving the optimizer,and changing the activation function to introduce an attention mechanism,through data collection and detection of nearly 100k finished products,the system has a detection rate of 99.998%,a false negative rate of 0,and a false detection rate of 0.002%,realizing high precision and stable operation.
作者
胡忠华
许海龙
王甫姜
HU Zhonghua;XU Hailong;WANG Fujiang(Shennan Circuits Co.,Ltd.,Shenzhen 518117,Guangdong,China;Wuxi Shennan Circuits Co.,Ltd.,Wuxi 214028,Jiangsu,China)
出处
《印制电路信息》
2024年第9期57-62,共6页
Printed Circuit Information
关键词
真假点
缺陷检测
视觉人工智能
平均精确度(mAP)
算法
true and false points
defect detection
visual artificial intelligence(AI)
mean accurate precision(mAP)
algorithm
