Single image dehazing algorithm using generative adversarial network based on feature pyramid network

This paper proposes a single image dehazing algorithm using generative adversarial network (GAN) based on feature pyramid network (FPN). This method is an end-to-end image dehazing method, avoiding the physical model dependence. The generator uses MobileNet-V2 as the backbone network, and uses the FPN structure to improve the feature utilization rate of the image, combined with the discriminator formed by the convolutional neural network to form GAN that can improve the training stability and convergence of the generator. The model uses a lightweight MobileNet-V2 network, and the FPN structure also enables multiple-scale feature maps to be obtained while avoiding the use of direct scaling, thus reducing the computational power and memory requirements and allowing the model to operate with limited computational resources. We used the RESIDE training set to train our proposed model and conducted extensive experiments on the test set. The experimental results show that the algorithm has satisfactory results in terms of quality and speed.