基于ResNet的焊接缺陷分类识别  

Welding defect classification based on ResNet

作　　者：翁佩纯 李蓉 张远海[2] WENG Peichun;LI Rong;ZHANG Yuanhai(College of Computer,University of Electronic Science and Technology of China,Zhongshan Institute,Zhongshan Guangdong 528400,China;College of photoelectric information,Zhongshan Torch Polytechnic,Zhongshan Guangdong 528403,China)

机构地区：[1]电子科技大学中山学院计算机学院,广东中山528400 [2]中山火炬职业技术学院光电信息学院,广东中山528403

出　　处：《智能计算机与应用》2023年第5期103-106,共4页Intelligent Computer and Applications

基　　金：中山市科技计划项目(2019B2074)。

摘　　要：本文针对称重传感器上应变片焊接缺陷问题,以ResNet50为基础模型,构建一种焊接缺陷分类模型,实现对正常、少焊、短路、掉线缺陷的分类识别。为解决缺陷样本分布不平衡的问题,通过缺陷分析和数据增强处理,构建应变片缺陷样本数据集。实验结果表明,本文搭建的焊接缺陷分类模型能有效地识别和分类焊接缺陷图像,在测试集上的分类准确率、精确率、召回率、F_(1)-score分别达到97.4%、97.2%、96.2%、96.7%。Aiming at the welding defects of strain gauge on weighing sensor,a welding defect classification model is constructed based on resnet50 to classify types of normal,less welding,short circuit and weld line defects.To solve the problem of unbalanced distribution of defect samples,the welding defect sample dataset is constructed through defect analysis and data enhancement processing.The experimental results show that the welding defect classification model built in this paper can effectively identify and classify the weld joint defects and weld line defects.The classification accuracy rate on the test set is 97.4%,the Precision is 97.2%,the recall rate is 96.2%,and the F_(1) score is 96.7%.

关 键 词：焊接缺陷 分类模型 数据增强 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]