In machine vision systems based on optical techniques, the degradation of hazy images poses difficulties for many applications, especially monitoring and early warning in the field of public safety, such as motion monitoring of drone flight, fixed video surveillance of police, border monitoring system, etc. The existing dehazing algorithms based on deep learning trained by synthetic datasets are also difficult to be applied to the above environment. To this end, a single image dehazing algorithm based on unsupervised learning is proposed. At first, by improving the extraction module of the unsupervised dehazing algorithm YOLY, the quality of the obtained image is improved. And then, the brightness of the generated image is improved by adjusting the module parameters. At last, various loss function adjustments are added to further improve the quality of the image. Experimental results show that the image processed by this method has bright colors and complete details, which is closer to the real image, and the image noise is significantly reduced, which has practical application value.