基于轮胎X光图像样本重采样图像缺陷检测  

Defect detection based on resampling of tire X-ray image samples

作　　者：刘韵婷 刘鑫 LIU Yunting;LIU Xin(School of Automation and Electrical Engineering,Shenyang Ligong University,Shenyang 110159,China)

机构地区：[1]沈阳理工大学自动化与电气工程学院,辽宁沈阳110159

出　　处：《通信与信息技术》2024年第5期19-23,共5页Communication & Information Technology

基　　金：辽宁省自然科学基金项目(项目编号:2022-KF-14-02);辽宁省教育厅面上项目(项目编号:LJKMZ20220617)资助。

摘　　要：针对生成对抗网络轮胎X光图像缺陷检测,训练阶段生成器会丢失部分图像特征,并且难以确定样本的潜在空间维度,导致部分不必要的图像特征重建。为了解决这些问题,构建了样本重采样生成对抗网络SRGAN(Sample Resampling Generate Adversarial Networks),生成器以VQ-VAE为基本框架,利用注意力特征融合模块(Atten-tion Feature Fusion,AFF)搭建了新的跳连层,并在SRGAN的生成器中加入了转换损失函数LVQ。最后,使用自制的轮胎X光图像数据集对SRGAN和已经提出的部分生成对抗网络模型进行训练和测试,并将得到的AUC值进行对比,进一步证明了SRGAN具有更好的图像缺陷检测能力。For defect detection in tire X-ray images generated by adversarial networks,during the training phase,the generator may lose some image features and find it difficult to determine the potential spatial dimensions of the samples,resulting in unnecessary image feature reconstruction.In order to solve these problems,a sample resampling generation adversarial network(SRGAN)was con⁃structed.The generator uses VQ-VAE as the basic framework and utilizes the Attention Feature Fusion(AFF)module to build a new hop layer.The conversion loss function LVQ was added to the SRGAN generator.Finally,the self-made tire X-ray image dataset was used to train and test SRGAN and the proposed partially generated adversarial network model,and the obtained AUC values were com⁃pared,further proving that SRGAN has better image defect detection ability.

关 键 词：生成对抗网络 VQ-VAE AFF 轮胎X光图像缺陷检测 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]