基于管道闭路电视(CCTV)检测系统,提出5种不同的管道结构性缺陷特征提取方法,对缺陷图像进行纹理特征提取,并利用二分类支持向量机(BSVM)对特征提取数据进行一对一投票分类。结果表明,Gabor法提取得到的特征维数远大于GLCM、B_GLCM、GGCM以及LBP四种方法得到的;B_GLCM法相比GLCM法,特征提取效率略有提高,但随着分区的减小,提取耗费的时间会逐渐增大;GLCM和B_GLCM提取方法对裂纹、错口和腐蚀三种缺陷均具有较好的分类效果,分类正确率都在90%以上,其他三种提取方法的分类正确率较低;综合考虑分类准确率和分类效率,建议采用分区大小为3×3的B_GLCM方法对管道缺陷纹理特征进行提取,以获得最佳的检测结果。

Based on the detection technology of CCTV (closed-circuit television) pipeline detection system, 5 different pipeline structural defect feature extraction methods are proposed to extract the texture features of the defect image, and the BSVM classifier is used to classify the feature extraction data one-to-one. The results show that the feature dimension extracted by Gabor method is much larger than GLCM and B_ GLCM, GGCM and LBP, compared with GLCM method, the efficiency of feature extraction of B_GLCM is slightly improved, but with the reduction of partition, the extraction time will gradually increase; for crack, staggered joint and corrosion, GLCM and B_GLCM have good classification effect, and the classification accuracy is more than 90%. Under the other three extraction methods, the classification accuracy is low. Considering the classification accuracy and efficiency, the B_GLCM method of partition size of 3×3 is recommended and it can extract the texture features of pipeline defects and obtain the relatively best detection results.