Deep back-projection networks have excellent performance in the super-resolution reconstruction of visual images. This paper explores the application of deep back-projection networks to the super-resolution reconstruction of infrared images. In view of the characteristics of low infrared image contrast and low image quality, the following improvements were made in the framework of the deep back-projection network: adding a concatenation layer before the upsampling module, cascading the previous downsampling output and the original low-resolution preprocessed image as the input of the upsampling module. This was designed to improve the network's ability to obtain high-frequency information of the image and enhance the detail of the generated image. The experimental results proved that the proposed algorithm could create infrared super-resolution reconstructed images with richer details and improved visual effects.