基于YOLOv5的电路板焊接缺陷检测优化算法

Circuit Board Defect Detection Equipment Based on YOLOv5

随着近年来深度学习目标检测算法的发展,使用目标检测算法对各种产品进行瑕疵缺陷检测受到广泛关注。主要提出了一种优化后的YOLOv5目标检测算法,用于提高对电路板焊接缺陷的检测成功率。优化后的算法在原有YOLOv5网络框架的基础上,新增了一个提升小目标检测精度的检测头。该检测头基于CNN构成,与原有的3个检测头结合,能更加有效地识别小目标。集中采集了两种主要焊接缺陷的图片进行焊接缺陷模型训练,经过实验,最后得出改进的算法相比于原有模型,在平均精确率AP值上能够提升约2%。

With the development of object detection algorithms in recent years,using object detection algorithms for defect detection in various products has received widespread attention.This article proposes an optimized YOLOv5 object detection algorithm to improve the detection success rate of welding defects on circuit boards.The optimized algorithm adds a detection head to improve the detection accuracy of small targets based on CNN.This detection head,combined with the original 3 detection heads of the YOLOv5 network framework,can effectively identify small targets.This paper collects 5 main welding defect images for welding defect model training.After experiments,the improved algorithm can improve the average precision(AP)value by about 2%compared to the original model.