基于非下采样轮廓波变换和朴素贝叶斯分类器的织物缺陷检测

摘要

为检测织物生产过程中产生的缺陷,提出一种非下采样轮廓波变换(nonsubsampled contourlet transform,NSCT)和朴素贝叶斯分类器(naive Bayes classifier,NBC)相结合的缺陷检测算法.该方法分为2个阶段:学习阶段和检测阶段.在学习阶段,分别提取有缺陷和无缺陷织物的子块集合,首先利用NSCT进行滤波去噪;然后提取每个子块的广义高斯分布的混合(mixture of the generalized Gaussion distribution,MoGG)模型,并计算子块之间的相对熵(kullbackleibler divergence,KLD);最后利用得到的数据训练NBC.在检测阶段,将待检测图像分割成子块,利用经过训练的NBC检测子块,输出缺陷检测结果.实验结果表明,该算法对于灰度均匀织物及净色纹理织物的缺陷检测均具有良好效果,并且利用该算法可以检测出多种缺陷类型,检测精度可达到97％,能满足工业生产需求.%In order to detect fabric defects that generated during the production process,the algorithm which combines Naive Bayesian classifier (NBC) and nonsubsampled Contourlet transform (NSCT) is presented.The proposed approach is divided into two phases:the learning phase and detection phase.In the learning phase,sub-blocks of defects and defect-free fabric are extracted,respectively.Firstly,NSCT is used to filter and denoise the image.Then,the mixture of the generalized Gaussian distribution (MoGG) model of each sub-block is extracted,and the Kullback-Leibler divergence (KLD) between the sub-blocks is calculated.Finally,NBC is trained by the obtained data.In the detection phase,the detected image is divided into subblocks.The trained NBC is used to detect the sub-blocks,and the defect detection results are output.Experiment results show that the algorithm of the defect detection has a good effect for the uniform gray and color textured fabric and can detect many kinds of fabric defects.The detection accuracy can reach 97％ which can meet demand of industrial production.