ValidFlow:基于标准化流的无监督图像缺陷检测  

ValidFlow:Unsupervised Image Defect Detection Based on Normalizing Flows

作　　者：张兰尧 陈晓玲 张达敏 岑翼刚[3] 张琳娜[1] 黄彦森 ZHANG Lanyao;CHEN Xiaoing;ZHANG Damin;CEN Yigang;ZHANG Linna;HUANG Yansen(School of Mechanical Engineering,Guizhou University,Guiyang 550025,China;College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China;School of Computer and Information Technology,Beijing Jiaotong University,Beijing 100044,China;College of Civil Engineering,Guizhou University,Guiyang 550025,China;Guizhou Lianjian Civil Engineering Quality Inspection Monitoring Center Co.Ltd.,Guiyang 550016,China)

机构地区：[1]贵州大学机械工程学院,贵阳550025 [2]贵州大学大数据与信息工程学院,贵阳550025 [3]北京交通大学计算机与信息技术学院,北京100044 [4]贵州大学土木工程学院,贵阳550025 [5]贵州联建土木工程质量检测监控中心有限公司,贵阳550016

出　　处：《数据采集与处理》2023年第6期1445-1457,共13页Journal of Data Acquisition and Processing

基　　金：国家自然科学基金(62062021);贵阳市科技计划项目(筑科合同[2023]48-11)。

摘　　要：基于标准化流的CS-Flow方法在缺陷检测领域取得了不错的效果,但其重复堆叠单一耦合块的方式增大了网络的复杂度。为此,本文提出了由特征平行流(Feature advection flow,FA flow)与特征混合流(Feature blending flow,FB flow)两种耦合块堆叠构成的网络ValidFlow,其中FA flow内部的子网络去掉了上下采样的捷径分支,并引入深度可分离卷积;FB flow内部的子网络在3个尺度上进行跨尺度融合。这样的设置使得ValidFlow在参数量减少的同时保证了信息的充分混合。在MVTec AD、MTD和DAGM数据集上与已有方法的对比结果显示,在MVTec AD数据集上,ValidFlow在15个类别中的平均AUROC为99.2%,在4个类别上的AUROC均为100%;在MTD数据集上获得了99.6%的AUROC;相比于CS-Flow,ValidFlow的参数量减少了207.61M,推理速度FPS提升了22;在DAGM数据集上,10个类别的平均AUROC为99.0%,性能上非常接近有监督的方法。The CS-Flow method based on normalizing flows has achieved good results in the field of defect detection,but its way of repeatedly stacking single coupling blocks increases the complexity of the network.Therefore,we propose a network ValidFlow composed of two coupling blocks stacking:Feature advection flow(FA flow)and feature blending flow(FB flow).In the subnetwork of FA flow,the shortcut branch of up and down sampling is removed and depth-separable convolution is introduced.The subnetworks within FB flow are fused across scales at three scales.This allows ValidFlow to reduce the number of parameters while keeping the information well mixed.Compared with the existing methods on MVTec AD,MTD and DAGM datasets,it can be seen that on MVTec AD datasets,the average AUROC of ValidFlow in 15 categories is 99.2%,and the AUROC of ValidFlow in four categories is 100%.On the MTD dataset,AUROC achieves 99.6%.At the same time,compared with CS-Flow,ValidFlow has 207.61M fewer parameters and 22 higher reasoning speed FPS.On the DAGM dataset,the average AUROC of the 10 categories is 99.0%,which is very close to the monitored method in terms of performance.

关 键 词：标准化流 特征平行流 特征混合流 深度可分离卷积 跨尺度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]