When light travels in water, it attenuates and causes phenomena such as absorption and scattering so that underwater images have problems with color distortion, low contrast, and fuzzy details. Since the attenuation of light varies with the wavelength and largely depends on the type of water body and structure of the underwater scene, it is difficult to remove various interference factors to achieve the best effect. We analyzed the current underwater enhancement and restoration algorithms, finding that the effectiveness of most methods requires further improvement and some methods have limited application in terms of scope. Therefore, we innovatively adopted a reinforcement learning method combined with underwater presentation characteristics to adaptively learn the most reasonable enhancement strategy and used evaluation criteria for underwater images as the feedback function of the reinforcement learning. This method further modifies the color, contrast, and detail of the image. It is also well-adapted to images with different turbidity conditions, water areas, and underwater environments, demonstrating robustness.