摘要
针对冷轧带钢表面缺陷图像特征提取的特点,提出了基于类距离可分离性判据的混合特征提取方法。该方法以小波变换的L1范数特征和灰度共生矩阵二次统计特征为基础,运用基于类距离的可分离性判据原理提取出可分离性特征向量。对几种生产现场出现频率较高、危害严重的典型缺陷进行了计算机实验研究,实验结果表明,运用基于类距离可分离性判据的混合特征提取方法提取的特征向量具有较大的可分离性,很大程度上提高了特征的分类有效性,使缺陷识别取得了较高的正确识别率。
Aiming at the characteristic in feature extraction of cold steel strip surface defect images,a mixed feature extraction method of separable criterion based on class distance is proposed.The method is based on wavelet transform L1 norm feature and secondary statistic feature of gray level co-occurrence matrix,extracts the separable feature vector according to the separable criterion theory based on class distance.Experimental investigations are carried out on computer aiming at several typical defects which are serious and excessive at the locale,the results show that the mixed feature extraction method of separable criterion based on class distance can get the more separable feature vectors, increase validity of classification of feature greatly,and get a higher correct recognition rate of defects.
出处
《计算机工程与应用》
CSCD
北大核心
2006年第27期184-186,190,共4页
Computer Engineering and Applications
基金
国家自然科学基金资助项目(编号:50574019)
国家科技部重大基础研究前期研究专项资金资助项目(编号:2003CCA03900)
关键词
冷轧带钢
表面缺陷
特征提取
可分离性判据
混合特征
cold steel strip,surface defect,feature extraction,separable criterion,mixed feature
