Recognition of Weld Defect Types

To improve the recognition accuracy of weld defects in the radiographic image,a method based on direct multiclass support vector machine (SVM) is proposed to recognize the defect types,where the recognition of weld defects is regarded as a constrained optimization problem,and the edge-based features and region-based features of the weld defect are employed as the feature vector. This method solves the difficulty of achieving higher accuracy under a small training set. The experimental results demonstrate that the recognition accuracy of the method gets 94.25%,higher than that of the one-versus-one SVM and multi-layer perceptron (MLP) neural network,and the introduced region-based features improve the characterization capability of the feature group.