Multispectral Images Pan-Sharpening Based on Atrous Convolution Network and Deep Residual Network

Abstract

Pan-sharpening aims to fuse a panchromatic and a multispectral image to enhance the spatial resolution of the latter while retaining its spectral information. Although many algorithms for solving this task have been proposed, there is still room for improvement in spatial detail preservation. In this paper, we propose a network called ARNet to achieve multispectral image pan-sharpening through deep learning. In order to better preserve the spatial details in the multispectral image, we propose to obtain the prior information from the atrous convolution network and then combine it with the residual network (ResNet) to implement pan-sharpening. Experimental results of the quantitative and qualitative evaluation show that the proposed method outperforms state-of-the-art pan-sharpening methods.