Research On Defect Super-resolution Detection Algorithm Based On Infrared Imaging Physics Kernel

With the development of image processing technology,infrared thermography(IRT)has made significant progress in non-destructive testing.This technique has the advantages of large detection area,high speed,and intuitive results.However,its image resolution is generally low,with the resolution of uncooled infrared cameras typically have lower resolution.Additionally,defects in non-destructive testing are often deep or small,making the identification of defects in infrared images more challenging.To address these issues,this paper proposes a super-resolution reconstruction method based on the physical imaging process of IRT,which uses a physical blur kernel that conforms to the infrared special degradation process to reconstruct high-quality images and assist in defect detection,thereby improving the performance of super-resolution algorithms.The specific research contents are as follows:First,the blur mechanism of infrared images is analyzed,and a modulation transfer function of an infrared system is established based on the physical process of imaging,which is represented as a three-dimensional plane of the physical kernel of infrared imaging.Furthermore,a physical blur kernel generation algorithm is proposed,which uses a GAN network to combine low-resolution images with physical kernels and employs adversarial training to enable downsampled images and original low-resolution images to have a similar level of blur,generating a blur kernel.Using the generated blur kernel,the proposed super-resolution reconstruction method achieves PSNR and SSIM values of 27.9d B and 0.76,respectively,which are higher than those obtained using other blur kernels,demonstrating that the proposed kernel is more closely aligned with the degradation process of infrared images and has a gain effect on the super-resolution algorithm.Second,based on the proposed physical blur kernel,an iterative kernel correction(IKC)method for enhancing defects in infrared super-resolution reconstruction is proposed.The algorithm estimates high-resolution images based on the physical blur kernel of infrared images,corrects the generated new blur kernel,and alternately optimizes the estimation and correction processes.Additionally,a feature enhancement module based on the physical shape of defects is added during the estimation process to reconstruct high-quality images for defect detection,further improving the accuracy and coverage of the blur kernel.The effectiveness of the proposed algorithm is verified through longitudinal and cross-sectional comparison experiments on different specimens,algorithm ablation experiments,and object detection experiments.The F-Score indicators of the reconstructed images are 86.1% and 87.7%,respectively,which are higher than those of other state-of-the-art super-resolution algorithms,demonstrating the superiority of the proposed algorithm in improving image quality,enhancing defect information,and reducing missed and false detections.