基于Gabor小波的TOFD图像缺陷识别研究

Defect recognition of TOFD image based on Gabor wavelet

针对运用超声衍射时差法(TOFD)法对焊缝进行检测时,图像缺陷人工定性主要受检验人员经验和专业知识影响缺乏可靠性的问题,提出了一种TOFD图像缺陷自动定性的方法。该方法首先提取TOFD缺陷图像的Gabor小波特征,并依据这些特征,采用主成分分析技术(PCA)对Gabor特征进行降维,然后采用Fisher线性判别分析方法对其进行了判别分析,最后完成了缺陷的自动定性分析;同时,建立了一个实际系统,并在测试样本上进行了试验验证,试验在109幅人工试块缺陷及自然缺陷训练样本及25幅测试样本中进行,采用Gabor小波特征及原始图像像素特征所构建的缺陷分类器识别率比较。研究结果表明,基于Gabor小波特征的缺陷识别方法识别率达到72%,比原始图像特征的缺陷识别方法更优。

Aiming at the problems of time of flight diffraction （TOFD） method which credibility of defect recognition influenced by personnel＇s experience and professional knowledge,an automatic recognition approach was proposed for TOFD image defects.Firstly,the Gabor wavelet of TOFD defects image characteristics was extracted,to reduce the dimension Gabor features,the principal component analysis （PCA） technique has been used based on these characteristics.Then,the Fisher linear discriminant analysis method was used to analysis,at last,the actual system and the test sample on the test verification has been established for testing.Experiment in the 109 training samples and 25 test samples which comes from artificials and natural defects,compared Gabor wavelet features with characteristics of the original image pixel defect classifier recognition rate in test.The results indicate that the rate of recognition based on Gabor wavelet feature is 72％,it has higher rate of recognition than the method based on the original image pixel.