Abstract: Aiming at the problem that underwater image degradation. This paper proposed an adaptive dense feature fusion underwater image enhancement algorithm based on conditional generative adversarial network (CGAN). The algorithm proposed a novel adaptive dense feature fusion (ADFF) module, which could prompt the network to learn more effective features from previous and current features for fusion by adaptively learning the spatial importance weights of different levels of features. In the experiment, the U-Net structure generator was used, the ADFF module was integrated at each level of the generator, and the WGAN-GP (Wasserstein GAN with gradient penalty) adversarial loss and combined loss of L₁ and L₂ loss was used to constrain the network model. Experimental results show that, compared with other underwater image enhancement algorithms, this algorithm achieves superior performance on both synthetic and real data sets, and can generate clear underwater images with better visual effects.