The invention discloses a near-infrared and visible light remote sensing image registration method based on reinforcement learning, and the method comprises the following steps: S1, trimming infrared and visible light images to the same size, and carrying out the stacking processing; S2, inputting the stacked images into a residual error improved dense neural network for processing, and outputting a Q value required by registration; S3, reasoning according to the Q value to predict probability distribution of each action in the strategy action space; S4, selecting an action with the maximum probability in the strategy action space according to the probability distribution of each action, and executing the action by the image to be registered; S5, after the to-be-registered image and the reference image reach a set similarity threshold value, carrying out greedy algorithm reasoning sampling on current image registration; and S6, carrying out moving resampling on the current image to be registered, and outputting a final registration result. According to the image registration method, all parts of the registered image are connected naturally and smoothly, multi-modal image registration is more stable, and the effect is better.