基于改进YOLOv7的钢材表面缺陷检测  

Steel surface defect detection based on improved YOLOv7

作　　者：张琴[1] ZHANG Qin(Department of Information Engineering,Fuzhou Polytechnic,Fuzhou 350108,China)

机构地区：[1]福州职业技术学院信息工程系,福州350108

出　　处：《智能计算机与应用》2024年第10期182-188,共7页Intelligent Computer and Applications

基　　金：2022年度福建省中青年教师教育科研项目(JAT220652)。

摘　　要：针对钢材表面缺陷纹理特征不明显、不同缺陷类间差异不明显和缺陷尺度变化剧烈等问题,本文设计了一个纹理信息增强模块(Texture Information Enhancement Module,TIEM)来保留主干网络上采样丢失的细节纹理特征信息和加强主干网络对不规则缺陷的空间建模能力;在颈部网络融入跳跃连接的多尺度自适应卷积模块(Multi-scale Adaptive convolution mod-ule with Skip Connections,MASC)来增强网络对不同尺度缺陷目标的感知能力,进而增强小目标的细粒度特征和大目标的高层语义信息,增强检测器的全局感知能力。以YOLOv7为基线模型,在公开数据集NEU-DET上,改进后的模型比基线模型mAP_(50)和mAP_(50:95)分别提高了3.0%和2.1%,并优于现阶段其他主流目标检测器。Aiming at the problems of inconspicuous texture characteristics of steel surface defects,inconspicuous differences between different defect categories,and severe defect scale changes,this paper designs a Texture Information Enhancement Module(TIEM)to preserve the loss of detailed texture feature information in the previous stage of the backbone network and strengthen the spatial modeling ability of the backbone network for irregular defects;a multi-scale adaptive convolution module(multi-scale Adaptive convolution module with Skip Connections,MASC)is integrated into the neck network to enhance the perception ability of the network to targets of different scales,and then enhance the fine-grained features of small targets and the high-level semantic information of large targets,and enhance the global perception ability of the detector.Taking YOLOv7 as the baseline model,on the public dataset NEU-DET,the improved model has improved by 3.0%and 2.1%compared with the baseline model mAP_(50) and mAP_(50:95),respectively,and is better than other mainstream target detectors at this stage.

关 键 词：目标检测 YOLOv7 缺陷检测 纹理信息 多尺度自适应卷积 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]