Underwater image enhancement method based on semi-supervised domain adaptation

Methods based on supervised learning perform well in underwater image enhancement. However, it is difficult to obtain clear labels of underwater images, and there are domain differences among various databases due to different acquisition equipment and waters, resulting in poor generalization ability of these methods. To solve the above problems, this paper proposes an underwater image enhancement algorithm based on semi-supervised domain adaptation, which is composed of domain adaptive module and image enhancement module. A domain adaptive module based on cycle-consistent generative adversarial networks (CycleGAN) is designed to eliminate the domain differences between different datasets. An image enhancement module based on semi-supervised learning strategy is proposed to solve the training problem of unlabeled images by adding physical priors of underwater images. Consistency constraint is used to ensure the stability of training and further improve network performance. The experimental results on the public datasets show that the proposed method is superior to the existing methods. In addition, the algorithm also perform well on the self-collected offshore underwater dataset.