Multi-scale Fusion Dehazing Algorithm based on Generative Adversarial Network

In haze weather, visible light is scattered and absorbed when it passes through the atmosphere, causing significant degradation in image quality. To address this issue, we propose a multi-scale fusion dehazing algorithm based on an improved Generative Adversarial Network (GAN). In the generator part, we design a multi-scale fusion attention network (Grid-G) with a three rows, multi-column structure. The network first extracts and strengthens local and global features in the image through a feature enhancement module, effectively solving the problem that traditional dehazing methods cannot fully capture image details and global semantic information. Then, channel attention and coordinate attention are introduced to process the dense fog area and high-frequency area of the image from different angles. Finally, a multi-scale fusion module is designed to replace the original ordinary feature splicing operation and enhance the detail fidelity of the image.In the discriminator part, a fusion discriminator (FDG-D) is constructed to enhance the ability to distinguish the source of the image by introducing the high-frequency and low-frequency information of the image as additional priors. The experimental results show that the algorithm proposed in this paper not only has natural colors and clear details subjectively, but also outperforms the existing mainstream algorithms in objective indicators.