基于改进YOLOv4的PCB裸板缺陷检测方法研究

Research on YOLOv4 Based on PCB Bare Board Defect Detection Method

PCB(Printed Circuit Board)裸板缺陷采用人工目视和电气特性等传统检测方法存在检测效率低、误检率高、成本高、劳动强度大、接触式飞针易造成损伤等问题。提出将YOLO(You Only Look Once)v4引入到PCB裸板缺陷检测,采用Canopy+K-means聚类改进YOLOv4中K-means聚类获取先验框。将改进的YOLOv4应用到PCB裸板的短路、开路、缺口、毛刺、焊点漏焊、余铜等缺陷检测中,开发了软硬件平台,形成了高精度、高检测速度的PCB裸板缺陷检测系统。基于PCB_DATASET数据集将该文提出的方法与已有的基于深度学习的PCB裸板缺陷检测方法进行了对比实验和分析。实验结果表明,该文提出方法的缺陷检测mAP(mean Average Precision)值达到了99.48%,检测速度为37.09帧/秒,相较于目前已有的基于深度学习的PCB裸板缺陷检测方法,不仅检测速度更快,而且检测精确度更高。

This paper proposes to introduce YOLO(You Only Look Once)v4 into PCB bare board defect detection,and use Canopy+K-means clustering to improve K-means clustering in YOLO v4 to obtain a priori frames.The improved YOLOv4 is applied to the detection of defects such as short circuit,open circuit,notch,burr,solder joint leakage and residual copper on bare PCBs,and a software and hardware platform is developed to form a PCB bare board defect detection system with high accuracy and high detection speed.Based on the PCB_DATASET dataset,the proposed method is compared with the existing deep learning-based PCB bare board defect detection method.The experimental results show that the mAP(mean Average Precision)value of the proposed method reaches 99.48%and the detection speed is 37.09 frames/second.