Difference of Gaussian Convolutional Sparse Principal Component Thermography for Defect Signal Enhance in Composite Materials

Abstract

Pulsed thermography (PT) is a well-established non-destructive testing technique for the subsurface defect detection in Carbon Fiber Reinforced Polymer (CFRP). Among the analysis methods for the thermographic data, principal component thermography (PCT) and sparse principal component thermography (SPCT) are recommended for visualization enhancement of defect signals. However, since the methods of PCT and SPCT are performed directly based on the characteristic matrix model of the original thermal images, their results are heavily affected by the noise and uneven background signals inside the images. To solve the problem above, a new method known as difference of Gaussian convolutional sparse principal component thermography (DoG-SPCT) is proposed in this paper. The method first separates defect signals from the interference with a DoG filter, and then extracts features for defective areas by SPCT to enhance visualization of defects. In the experimental part, one CFRP specimen with subsurface defects is detected by PT and the proposed DoG-SPCT is evaluated for the defect visualization enhancing purpose. The result of the experiment shows that the DoG filter can separate the defect components from the noise and uneven background signals, so that the features for defective regions can be effectively extracted in the following SPCT.