基于改进YOLOv4的铝材表面缺陷检测方法

Defect detection method in aluminum material surface based on improved YOLOv4

针对铝材表面缺陷检测精度不高,容易漏检的问题,提出基于改进YOLOv4的缺陷检测方法。在CSPResblock模块中引入注意力机制SE模块,赋予各个通道相应的权重,加强网络对于重要信息的训练,提升网络的特征提取能力;改进SPP模块,使用不同宽高比的池化核,有利于保留更多的短边信息,提高网络对大宽高比缺陷的检测能力;对PANet结构进行改进,在对应特征层级上拼接输入信息,主要融合主干网络的三层输出,获得更多较浅的特征信息,提升对小目标的检测能力;实验结果表明,改进后的YOLOv4算法在铝材表面缺陷数据集上的精度(mAP)达79.27%,优于其他常见目标检测算法。

Aiming at the problem of low accuracy and easily missing defects in aluminum surface defect detection,a defect detection method of improved YOLOv4 is proposed.SE is introduced into CSPResblock giving channels weights,which can increase training effect for important information and improve the ability of feature extraction.Using pooling kernels with different aspect ratios is beneficial to retain more short-side information,so SPP is revised to improve the network's ability to detect large aspect ratio defects.PANet is improved to fuse more input shallow feature information from the three outputs of backbone,increasing the ability in detecting small objects.The experiment result shows that mAP of improved YOLOv4 algorithm achieves 79.27%in aluminum surface defect data set,better than other common object detection algorithms.