焊接缺陷磁光成像纹理特征GLCM-Gabor识别方法

Texture feature extraction and recognition of magneto-optical images of welded defects based on GLCM-Gabor

一种基于磁光图像纹理特征的焊接缺陷无损检测方法,首先用法拉第磁致旋光效应,结合漏磁场及磁畴理论分析焊接缺陷与磁光图像关系.针对缺陷磁光图像特点,通过灰度共生矩阵(gray level co-occurrence matrix,GLCM)提取磁光图像纹理特征.由于裂纹和凹坑的GLCM纹理特征参数区分度不高,提出用Gabor变换法进一步提取磁光图像纹理特征.将GLCM-Gabor纹理特征作为输入量,用支持向量机(support vector machine,SVM)构造缺陷分类模型.结果表明,该方法可有效识别焊缝表面及亚表面特征(凹坑、裂纹、未熔透、无缺陷),分类模型整体识别率可达89.7%.

A nondestructive testing method for welded defects based on texture features of magneto-optical image was studied. The relation between welded defect and magneto-optical image was analyzed by Faraday magneto-optical effect,leak mag-netic flux and magnetic domain theory. The texture features were extracted by using GLCM to calculate magneto-optical image.Due to the difference of GLCM texture features between crack and sag was subtle,the texture features of magneto-optical image were further extracted by using Gabor filter. A pattern recognition method based on GLCM-Gabor and SVM was proposed to establish the classification model by processing magneto-optical images of welded defects. Classification result showed that seam features( sag,crack,incomplete penetration and no defect) in surface and subsurface could be recognized by this method with an accuracy rate of 89. 7%.