Moran’s Index-Based Tensor Decomposition for Eddy Current Pulsed Thermography Sequence Processing

Abstract

Eddy current pulsed thermography (ECPT) is widely used in the nondestructive testing (NDT) of metal surface defects. Because defect information is sometimes affected by the edge effect, it is necessary to segment the ECPT image sequence, that is, separate the defected part from the background, to improve the detection effect. Tensor robust principal component analysis has been widely used in many image segmentation fields, such as medical imaging, face recognition, and so on, which also includes NDT, but it faces the problems of high time cost, overfitting, and insufficient physical interpretability in industrial applications. In this article, a new variable, global Moran’s index, is introduced to tensor decomposition for ECPT image sequences processing since the defect signal always gathers around the defects and shows spatial cohesion. The proposed method, named Moran’s index-based tensor decomposition (MITD), can significantly reduce the iterations, by up to 97%, and remove the influence of background noise. To demonstrate the performance of MITD, several experiments are carried out on three artificial samples and three natural samples from a nuclear power plant. Furthermore, a detailed comparison is drawn between MITD and other existing methods. The experimental results show that MITD not only reduces the time cost but also improves the image contrast.