基于深度学习的PCB缺陷检测方法研究

Research on Defect Detection Method of PCB Based on Deep Learning

针对在PCB生产过程中出现漏孔、鼠咬、开路、短路、毛刺、余铜、灰尘、划痕等缺陷而影响其后期使用的问题,提出了一种基于Faster R-CNN算法的PCB缺陷检测方法。该方法以ResNet-101为基础骨干网络构建特征金字塔网络,采用Soft-NMS算法对预选框进行筛选,然后使用在线困难样本挖掘方法,将损失值较高的困难样本集中进行处理,提高网络对复杂PCB缺陷样本检测的精确度。实验结果表明,改进后的Faster R-CNN缺陷检测方法可以对各类PCB缺陷进行准确定位和分类,平均检测精度达到93.76%,相较于传统Faster R-CNN方法提高了24.5个百分点,对PCB缺陷全自动检测的研究有一定参考价值。

A PCB defect detection method based on the Faster R-CNN algorithm is proposed to address the problem of defects such as leaky holes,mouse bites,open circuits,short circuits,burrs,residual copper,dust and scratches that occur in the PCB production process and affect its later use.The method uses ResNet-101 as the basic backbone network to construct a feature pyramid network,adopts the Soft-NMS algorithm to filter pre-selected boxes,and then uses an online difficult sample mining method to concentrate difficult samples with high loss values for processing to improve the accuracy of the network in detecting complex PCB defect samples.The experimental results show that the improved Faster R-CNN defect detection method can accurately locate and classify various types of PCB defects with an average detection accuracy of 93.76%,an improvement of 24.5 percentage points compared to the traditional Faster R-CNN method,which is of reference value for the study of fully automated PCB defect detection.