基于主成分分析法的本色布疵点分类算法

Grey fabric defect classification based on principal component analysis

特征提取是本色布疵点分类最关键的问题。针对本色布疵点类型多、形态变化大,使得本色布疵点特征提取算法很难实现的问题,从分析本色布编织方法出发,分析本色布图像的特点,发现本色布图像中存在极强的相关性。采用主成分分析法(PCA)对本色布图像进行去相关性处理,将图像压缩到前k个最大方差的子空间,作为图像的特征向量。在含有457幅训练样本,795幅测试样本的平纹本色布数据集上,最大分类准确率达99.11%。

Feature extraction is the key process of grey fabric defect classification, however, due to the large numbers of defect type and large variety of defect appearance, it is difficult to find an effective feature. Started from the wave pattern analysis to find out characteristics of fabric image, and then strong linear correlation in grey fabric image was found out in this paper, and PCA algorithm was adopted to process fabric image and the image was compressed into k-dimension subspace to be the first k-th variance, and the compressed image was flattened as the final feature vector. Experiments were carried out on a plain fabric dataset of 457 training samples and 795 testing samples, results showed that the algorithm in this paper can gain an accuracy of 99.11%.