Weld defect classification based on texture features and principal component analysis

In the traditional process used to classify weld defects in radiographic images, the precision of the weld defect region (WDR) segmentation and the validity of the defect feature extraction significantly influence the classification accuracy. To address the abovementioned issues, a feature extraction and weld defect classification method based on texture features and principal component analysis (PCA) is proposed in this paper. Firstly, the weld seam region (WSR) in the radiographic image, which contains several WDRs, is segmented, rather than the individual WDRs. Then, the texture features of the WSR are extracted for use in the weld defect classification process. Since the extracted texture features have high dimensions and are partially redundant with each other, PCA is then used to reduce the dimensions of these features. Finally, a multiclass support vector machine (SVM) is used to classify the defects based on the obtained principal components. The experimental results demonstrate that the proposed method can effectively extract the general features of weld defect types and can achieve a classification accuracy of up to 90.4%. Moreover, the proposed method can classify slag (SL) and porosity (PO) defects more accurately than traditional classification methods.