基于改进Faster RCNN的PCB表面缺陷检测研究  

Research on PCB Surface Defect Detection Based on Improved Faster RCNN

作　　者：龚陈博 南卓江 陶卫[1] GONG Chenbo;NAN Zhuojiang;TAO Wei(School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240

出　　处：《自动化仪表》2024年第7期99-103,109,共6页Process Automation Instrumentation

摘　　要：印刷电路板(PCB)在制造过程中不可避免地存在焊点缺焊、短路、毛刺、缺口、开路、余铜等微小缺陷。传统的基于机器视觉检测的缺陷检测方法存在检测速度慢、误检率和漏检率高、抗干扰能力弱等问题。为解决上述问题,提出一种基于改进快速区域卷积神经网络(Faster RCNN)的PCB表面缺陷检测方法。首先,在传统Faster RCNN框架的基础上,融入扩展特征金字塔网络(EFPN)以实现特征提取与融合,并进行多尺度检测,从而尽可能保留图像细节信息以提高检测性能。其次,利用K-means算法结合交并比(IoU)优化区域建议网络(RPN)结构中的锚框参数,使得生成的锚框方案更有针对性。试验结果表明,改进Faster RCNN在PCB缺陷数据集上的全类平均正确率(mAP)值达到93.4%、检测速度达到每秒21.79帧。所提方法可推广应用至芯片、光学器件表面微小缺陷在线检测,从而提升工业生产效率。Printed circuit board(PCB) inevitably have tiny defects such as solder joints missing solder,shorts,burrs,nicks,open circuits,residual copper and so on during the manufacturing process.The traditional defect detection methods based on machine vision inspection have problems such as slow detection speed,high false detection and leakage rates,and weak anti-interference ability and so on.To solve the above problems,a PCB surface defect detection method based on improved faster region convolutional neural network (Faster RCNN) is proposed.Firstly,on the basis of the traditional Faster RCNN framework,an extended feature pyramid network (EFPN) is incorporated to achieve feature extraction and fusion for multi-scale detection to maximize the retention of image detail information to improve the detection performance.Secondly,the K-means algorithm combined with the intersection over union(IoU) is used to optimize the anchor parameters in the structure of the region proposal network (RPN),which makes the generated anchor scheme more targeted.The experimental results show that the improved Faster RCNN achieves a mean average precision(mAP) value of 93.4% and a detection speed of 21.79 frame per second on this PCB defect dataset. The proposed method can be generalized to the online detection of tiny defects on the surface of chips and optical devices to improve the efficiency of industrial production.

关 键 词：印刷电路板 缺陷检测 快速区域卷积神经网络 扩展特征金字塔网络 K-MEANS 小目标检测 机器视觉 

分 类 号：TH39[机械工程—机械制造及自动化]