Research of Image Dehazing Algorithm Based on CNN

In response to the problems of fog residue and color distortion in current dehazing methods, this paper combines the advantages of generative adversarial networks in image super-resolution reconstruction and proposes an image dehazing algorithm based on channel attention and conditional generative adversarial networks. This algorithm adopts an encoding decoding structure, and the generator designs a multi-scale residual module and an efficient channel attention module to expand the receptive field, extract multi-scale features, and dynamically adjust the weights of different channels to improve feature utilization. The accuracy of image discrimination has been improved by using Markov discriminator for image segmentation evaluation. At the same time, the loss function increases content loss, reduces pixel and feature level losses in dehazing images, preserves more image detail information, and achieves high-quality image dehazing. The experimental evaluation on the RESIDE dataset indicates that our proposed model outperforms other advanced algorithms, achieving an average improvement of 36.36% in peak signal-to-noise ratio and 8.80% in structural similarity index. It effectively improves the problems of color distortion and incomplete dehazing, and is an effective image dehazing algorithm.