Recently, deep learning-based algorithms have emerged in the image defogging field. A cyclic generative adversarial network (CycleGAN) was used to create an image defogging algorithm in this study. Further, the desired processing effect was achieved by optimizing the generator in the CycleGAN. In the encoding and decoding networks of the generator, Leaky ReLU and Tanh activation functions were used, and the residual blocks of the conversion network were optimized by reducing the number and weighting optimization. The use of fog in the design of a single image can result in improved clarity and detail. Peak signal-to-noise ratio, structural similarity, and information entropy are some of the objective evaluation indexes that were enhanced.