Abstract: Firstly, aiming at the problem of traditional single-stage remote sensing image fusion task that required a large number of supervised samples and poor retention of image feature information, we proposed a two-stage panchromatic sharpening method for remote sensing images. The method achieved the fusion of remote sensing images by decomposing the task into two tasks of feature fusion and super-resolution. In the first stage,  the adversarial network feature fusion was generated, and in the second stage,  the super-resolution network generated clearer spatial features,  achieving the goal of high quality remote sensing image fusion. Secondly, the  multiple experiments were conducted by using GaoFen-2 and WorldView-3 satellite datasets to verify the effectiveness of the proposed method, and the fusion results were evaluated by using reference image quality indexes and non-reference image quality indexes, respectively. The experimental results show that the method can better retain the spectral feature information and spatial feature details compared to the traditional methods, and effectively improving the visual effect of the fused image.

Key words: remote sensing image fusion, pansharpening, generative adversarial network, super-resolution