基于残差模型改进的带钢缺陷分类网络  

Improved Defect Classification Network for Steel Strip Based on Residual Model

作　　者：张恒 付伟 杨钰婷 何新龙 王晓明 ZHANG Heng;FU Wei;YANG Yuting;HE Xinlong;WANG Xiaoming(School of Computer and Software Engineering,Xihua University,Chengdu Sichuan 610039,China;College of Chinese&Asean Arts,Chengdu University,Chengdu Sichuan 610106,China;College of and Law,Harbin University,Harbin Heilongjiang 150080,China;School of Civil Engineering,Harbin University,Harbin Heilongjiang 150076,China)

机构地区：[1]西华大学计算机与软件工程学院,四川成都610039 [2]成都大学中国-东盟艺术学院,四川成都610106 [3]哈尔滨学院文法学院,黑龙江哈尔滨150080 [4]哈尔滨学院土木建筑工程学院,黑龙江哈尔滨150076

出　　处：《信息与电脑》2023年第11期189-191,共3页Information & Computer

摘　　要：为了进一步提高带钢缺陷图像分类的准确性,文章基于ResNet18进行了改进,得到了一种新的网络结构,并命名为FusionNet。在FusionNet模型上添加卷积块注意力模块(Convolutional Block Attention Module,CBAM)后,验证集的准确性达到了99.50%,较基于ResNet18网络的CBAM模块及DenseNet121算法,准确率分别提高了0.94个百分点和1.51个百分点。设计的实验证明了改进后的网络具有一定的实用价值。In order to further improve the accuracy of defect image classification in strip steel,an article was written that presented improvements based on ResNet 18,resulting in a new network architecture,named FusionNet.After adding a Convolutional Block Attention Module(CBAM)to the FusionNet model,the accuracy of the validation set was found to have been increased to 99.50%.This was compared to the performance of the CBAM module based on the ResNet18 network,and the DenseNet121 algorithm.The accuracy improvement was found to be 0.94%and 1.51%,respectively.The designed experiments have demonstrated that the improved network has practical value.

关 键 词：残差网络 图像分类 缺陷分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]