Most of the images collected by infrared devices have problems such as blurred edges and low signal-to-noise ratio, and it is necessary to enhance their details and remove noise to make up for the lack of hardware conditions. Based on the deep convolutional adversarial generative network, this paper designs an end-to-end lightweight infrared image enhancement network by introducing the residual module, which realizes the mapping of low-quality blurred images to high-quality clear images. At the same time, the high-low quality infrared image pairs dataset is used for experimental comparative analysis. The results show that compared with the existing infrared image enhancement network, the structure designed in this paper can simplify the complexity of the model and accelerate the training speed of the network without reducing the image enhancement quality.