Weld Segmentation and Defect Detection in Radiographic Images of Pipe Welds

This article dwells upon the process of the automated weld defect detection after segmentation of the weld area in radiographic images. The current software lacks automatic image processing. This paper proposes the algorithm for fully automatic weld segmentation and defect detection. The deep convolutional neural network FgSegNet is applied to weld segmentation while the neural network, consisting of 3 convolutional layers, is applied to defect detection. The experiments show that the prediction accuracy of FgSegNet is 0.9929 and the prediction accuracy of the convolutional neural network for defect detection is 0.8429 by the binary accuracy metric. The next step of automatic system development will be the improvement of accuracy of the convolutional neural network for defect detection and convolutional neural network design for defect classification.