改进YOLOv7的带钢表面缺陷检测算法
Improved YOLOv7 Strip Surface Defect Detection Algorithm
孙卫波 1丁卫1
作者信息
- 1. 上海大学机电工程与自动化学院,上海 200444
- 折叠
摘要
针对带钢缺陷检测中存在的计算复杂度高、对小目标检测效果差等问题,提出了一种基于改进YOLOv7 的小目标检测算法,用于区分带钢在生产过程中产生的表面缺陷.在YOLOv7 的基础上,首先,将loss部分的损失函数CIOU替换为MPDiou,可以更好地处理目标框之间的交叉、遮挡等情况;接着,使用PConv(Partial Convolution)替换YOLOv7 中的Backbone部分的卷积,可以减少算法的冗余计算量;最后,在YOLOv7 的Head部分引入SimAM(Simple Attention Mechanism)注意力机制、动态蛇形卷积(Dynamic Snake Convolution)和BiFPN特征融合模块,以增强卷积神经网络的特征表达能力.
Abstract
In order to solve the problems of high computational complexity and poor detection effect of small targets in strip defect detection,a small target detection algorithm based on improved YOLOv7 was proposed to distinguish the sur-face defects generated by strip steel in the production process.On the basis of YOLOv7,firstly,the loss function CIOU of the loss part is replaced by MPDiou,which can better deal with the intersection and occlusion between the target boxes.Then,PConv(Partial Convolution)is used to replace the convolution of the backbone part in YOLOv7,which can reduce the redundant computation of the algorithm.Finally,the SimAM(simple attention mechanism),dynamic snake convolution and BiFPN feature fusion modules are introduced into the Head part of YOLOv7.
关键词
带钢/缺陷检测/YOLOv7/PConv/SimAM/BiFPNKey words
strip steel/defect detection/YOLOv7/PConv/SimAM/BiFPN引用本文复制引用
出版年
2024
NETL
NSTL
万方数据