Underwater images suffer from the problems such as color distortion and severe loss of image contrast. Most existing deep learning-based underwater image enhancement methods heavily rely on simulated datasets，and their capability to generalize in actual measurements is limited by the large distribution differences between simulated and real data. To solve this problem，we formulate the challenging underwater image enhancement task into two simpler yet physically explicit sub-problems：color correction and contrast enhancement. and propose a physically disentangled joint intra- and inter-domain adaptation paradigm. First，the intra-domain adaptation corrects the image color. The degraded image is decomposed through learning，and the color degradation is aligned at the scene light level to correct the color distortion while ensuring that other factors are completely unaffected. Furthermore，the decomposition strategy based on underwater scattering model is used again to transfer the underwater degradation factors in a targeted way，so as to achieve mutual migration and interaction between simulation and actual measurement domains and enhance the contrast of underwater images. The results of this method on real underwater image datasets are better than the existing contrast methods in terms of color，texture details and clarity.
