A New GAN-Based Approach to Data Augmentation and Image Segmentation for Crack Detection in Thermal Imaging Tests

Abstract As a popular nondestructive testing (NDT) technique, thermal imaging test demonstrates competitive performance in crack detection, especially for detecting subsurface cracks. In test, the temperature of area is higher than that non-crack during NDT process. By extracting features image sequences, curve each spatial point employed detection. Nevertheless, quality images influenced by noises due to complex environment NDT. this paper, modified generative adversarial network (GAN) improve segmentation performance. To feature extraction ability and alleviate influence noises, penalty term put forward loss function conventional GAN. A data preprocessing method developed where principle component analysis algorithm adopted extraction. The argumentation technique utilized guarantee quantity training samples. validate its effectiveness NDT, GAN applied detect cracks on eddy current pulsed thermography dataset.