Preliminary Study on the Method of Infrared Image Reconstruction Based on SRGAN Network

With the continuous improvement of the resolution of detectors and experimental equipments, the original low-resolution real samples can not meet the training requirements of simulation and recognition algorithm. Simply Stretching the image can cause the blur and Mosaic of the image. The image areas and blocks cannot truly represent the optical characteristics. During the enhancement process, the details are destroyed. Therefore, more efficient and accurate methods should be explored for infrared scene generation, so as to improve the authenticity and diversity of infrared imaging simulation. This paper mainly aims at the above problems, introduces super-resolution generated adversarial network, and draws the advantages of its jumping connection and deep residual network structure, defining new loss function to improve the output image. Exploring ways to improve the quality of infrared images and solve the problems of limited number of samples and low resolution in the process of infrared imaging simulation.