Deep features based on a DCNN model for classifying imbalanced weld flaw types

Feature extraction and feature selection are vital steps to construct an intelligent diagnosis system for classifying the weld flaws from an X-ray image. Deep learning has been successfully used in image analysis and automatic object recognition. It has good performance for learning more representative hierarchical features that are more sensitive to classification. However, there are still few applications of deep learning in feature learning for classifying different weld flaws, and few studies have been performed to compare the feature classification ability of different feature extraction methods. In this paper, we developed a model based on a deep convolutional neural network (DCNN) to extract the deep features directly from X-ray images. To validate the effectiveness of deep features, we cropped patches from the X-ray images as the learning dataset. Furthermore, considering the imbalance of the number of patches with different weld flaws, we used 3 types of resampling methods for 3 balanced datasets. Using the datasets, we compared the classification ability of 5 types of features extracted using traditional methods and deep learning. The best results were obtained for the deep features from the proposed DCNN model, achieving an accuracy of 97.2%, which is considerably higher than that obtained using the traditional feature extraction methods. We believe that the proposed model could be used to help workers evaluate X-ray images more intelligently.