基于深度学习的印刷电路板缺陷检测

PCB Defect Detection Based on Deep Learning

为降低传统人工检测的压力,提升检测效率,使用基于深度学习的方法解决印刷电路板制造中的缺陷检测任务。使用YOLOv5s目标检测算法完成印刷电路板快速高效的缺陷检测。搜集具有各类缺陷的印刷电路板样本数据,同时使用数据增强扩充数据集并分别对其执行标注;在Pycharm中配置Pytorch深度学习环境并对以官方YOLOv5s权重作为预训练模型完成印刷电路板缺陷检测任务。结果表明:训练后最优模型平均检测精度为96.3%;经过测试在英特尔i5-8265U中央处理器平台上可以获得约11帧每秒的实时检测速度。基于YOLOv5s的印刷电路板缺陷检测具有良好的检测性能,可以代替传统人工检测,减轻了人力财力压力,具有很高的实用价值。

To reduce the pressure of traditional manual inspection and improve detection efficiency,a method based on deep learning was used to solve the defect detection task in printed circuit board(PCB)manufacturing.The YOLOv5s object detection algorithm was adopted to quickly and efficiently detect defects in PCB.The sample data of PCB was collected with various defects,and the data augmentation was used to expand the dataset and label them separately;a Python deep learning environment was configured in Pycharm and the PCB defect detection tasks were completed by using the official YOLOv5s weight as a pre trained model.The results show that the average detection accuracy of the optimal model after training is 96.3%;after testing,it can achieve a real-time detection speed of about 11 frames per second on the Intel i5-8265U CPU platform.The defect detection of PCB based on YOLOv5s has good detection performance and can replace traditional manual detection,thereby reducing the pressure of manpower and financial resources,which has high practical value.