基于SSD模型的PCB外观缺陷检测

PCB Appearance Defect Detection Based on SSD Model

目前工业生产手机主板过程中,对于贴片区域以及元器件的外观检测主要还是由人工目检来完成,但人工目检漏检率高,疏漏大,人力成本高,耗时大,耽误生产。针对这些难点,本文提出将深度学习模型应用在工业生产中,代替人工目检,提高检测率和生产效率。采用SSD模型,减少和优化了网络参数,减少了网络层数,加快了网络推理时间。根据现场生产自做数据集,对缺陷进行实时标注,用Halcon对输入网络的图片进行统一处理。使得整体对缺陷的检测满足客户对CT和直通率的需求。

At present,in the production process of mobile phone motherboard in industrial production,the appearance inspec⁃tion of patch area and components is mainly completed by manual visual inspection.However,due to the high rate of missing in⁃spection,large omission,high labor cost and time-consuming,the production is delayed.In view of these difficulties and pain points,the deep learning model is proposed to be applied to industrial production instead of manual visual inspection to improve the detection rate and production efficiency.SSD model is used to reduce and optimize network parameters,reduce some network layers and speed up network reasoning time.According to the on-site production data set,the defects are marked in real time,and the pictures input into the network are uniformly processed by Halcon.Make the overall defect detection meet the needs of custom⁃ers for CT and through rate.