Defect classification for ultrasonic inspection in weld seam based on LBP-KPCA feature extraction

Weld defects affect the structural safety and the defect classification is important for structural safety evaluation. This paper proposes a method which combines one-dimensional local binary pattern (1-D LBP) algorithm and kernel principal component analysis (KPCA) to extract the characteristics of weld defect echo signal. The 1-D LBP algorithm is used to extract the LBP features of the defect echo signal, and the principal component analysis of the LBP features set is carried out by KPCA. The top N principal components with the contribution rate of more than 90% are selected as the feature vectors for defect classification. The automatic classification of defect types is realized by support vector machine (SVM) based on radial basis function. The experiments of defect features extraction and classification were carried out with the weld defects of slag, porosity and non-penetration. The results show that the accuracy of defect classification is 96.7% when the LBP-KPCA features are used, which is superior to the conventional features. The proposed method provides an important reference for defect classification and nondestructive evaluation of weld defects.