基于卷积自编码网络的织物瑕疵检测方法研究  

Fabric Defect Detection Based on Convolution Autocoding Network

作　　者：刘艳锋 黄惠玲 韩军 LIU Yan-feng;HUANG Hui-ling;HAN Jun(Fujian Institute of Research on the Structure of Matter,Quanzhou Institute of Equipment Manufacturing,Chinese Academy of Sciences,Quanzhou 362200,China;College of Electrical Engineering and Automation,Fuzhou University,Fuzhou 350108,China)

机构地区：[1]中国科学院福建物质结构研究所泉州装备制造研究所,福建泉州362200 [2]福州大学电气工程与自动化学院,福建福州350108

出　　处：《光学与光电技术》2022年第2期47-53,共7页Optics & Optoelectronic Technology

基　　金：福建省科技计划项目(2019T3020、2018T3007、2019T3025);泉州市科技计划项目(2019STS04、2019C097R、2019STS07);中国科学院对外重点合作项目(121835KYSB20180062)资助。

摘　　要：针对织物瑕疵数据集搜集和织物瑕疵检测困难的问题,提出了一种使用深度学习与传统算法相结合的织物瑕疵检测算法。首先提出特征金字塔结构的自编码网络,对正常样本进行学习。其次检测过程中提出同一尺度下进行多模型融合,在降低漏检率的同时移除纹理噪声的干扰。实验结果表明,所提出的学习方法对织物中线状瑕疵检测率高达98%以上,对织物中的面状瑕疵的检测率也达到了84%以上。对于实际生产过程中的瑕疵检测具有应用价值。Aiming at the great difficulties in collecting fabric data-sets and defecting fabric detection,an algorithm of fabric defect detection using deep learning combined with traditional methods is proposed in this paper. Firstly,an autocoding network based feature pyramid structure is proposed,which only needs normal samples for learning. Secondly,in the detection phase,multi-model fusion at the same scale is proposed to reduce the false alarm rate and remove the interference of texture noise. The experimental results show that the learning method proposed in this paper has a detection rate of over 98% for linear defects and over 84% for planar defects. It has more application value in practice.

关 键 词：光照归一化 卷积神经网络 图像重构 残差图融合 瑕疵检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]