Super-resolution Reconstruction of Infrared Images of Internal Defective Metal Plates Based on Generative Adversarial Networks

摘要

Infrared thermal imaging non-destructive testing technology has made great progress and has been used in the field of defect detection. However, due to the internal noise of the infrared imaging equipment and the influence of the surrounding environment interference, the infrared images used for defect recognition have the disadvantages of low contrast, low resolution, and low signal-to-noise ratio. Our article first briefly introduces the basic principle of infrared thermal imaging detection technology and the development status at home and abroad, then we builds an infrared image acquisition system that uses long-pulsed thermal method to collect internal defects of a metal plate. In order to solve the problem of noises, our paper first adopts Gaussian blur and homomorphic filtering on the acquired infrared images. Then we use Laplacian operator to process the filtered images and obtained second-order differential images. Finally, we use GAN for super-resolution reconstruction of the filtered second-order differential images. The results show that the super-resolution reconstructed images have higher PSNR and SSIM. What’s more, it retains detailed information about defects in the original infrared images.