Lock-in Thermography-based Resolution Improvement Using Gaussian Fourier Transform and Deep Learning

In the era of nondestructive testing 4.0, the post-processing of digital infrared thermography technology is an essential element. In this study, the lock-in thermography technique among digital infrared thermography technique was used to detect circular defects existing on the back side of metal specimens. Generally, a halogen lamp is used as an external heat source device, which has the problem of not providing uniform heat. To address this problem, the uniformity of the heat pattern was ensured by applying Gaussian filtering to fast Fourier transform. In addition, a very-deep super-resolution algorithm among deep learning techniques was applied to improve the resolution of noisy images. It was confirmed that SNR was improved in VDSR images. In addition, 11 detection rates were confirmed at 0.02 Hz, and a tendency for the detection rate to decrease as the frequency increased was confirmed. Accordingly, this study presents a thermal-homogenization process method for phase and amplitude images and an algorithm for improving resolution.