基于最小分类误差小波特征的纺织品缺陷分类方法研究

Fabric Defect Classification Using Minimum-classification-error Based Wavelet Features

纺织品缺陷分类是利用计算机视觉技术检测纺织品品质的一个关键环节。提出了一种基于小波框架的纺织品缺陷分类新方法。该方法使用纺织品图像的小波框架来描述缺陷的纹理特征。在最小分类误差训练框架下,通过联合设计一个基于线性变换矩阵的特征提取器和一个分类器,来获取面向缺陷分类的小波框架特征,并最小化分类器的错误概率。该方法对包含9类纺织品缺陷的329个样本,以及328个无缺陷样本进行了分类实验评估,获得了93.1%的分类准确率,相比传统的基于小波变换的分类方法提高了27.2%。

Fabric defect classification plays an important role in computer visionbased fabric quality inspection. In this paper, a novel defect classification method based on wavelet frames is proposed. Defects of texture properties are characterized using the wavelet frames. Minimum classification error training method is used to incorporate the design of a linear transform matrixbased feature extractor and a classifier, which yields classification-oriented wavelet features and minimizes the error rate associate with the classifier. The proposed method has been evaluated on the classification of 329 defect samples containing nine classes of fabric defects, and 328 non-defect samples. A 93.1% classification accuracy has been achieved which is 27. 1% better than the traditional wavelet-based classification method.