Mulit-Stage Dual-Domain Guided Attention Network for Pansharpening

Available deep learning (DL) based pansharpening methods primarily extract spectral information and spatial information from panchromatic (PAN) and multispectral (MS) images, respectively. This hinders the efficient derivation and exploitation of the potential spectral-spatial information, thereby degrading the quality of generated high-resolution multispectral (HRMS) images. In this paper, we propose a novel multi-stage dual-domain guided attention pansharpening network (MDGAPN), aiming to fully capitalize on the spectral and spatial information contained in both the PAN and MS images. Firstly, to enhance the feature extraction, we introduce two subnetworks: the progressive spectral feature extraction subnetwork (PSPeN) and progressive spatial feature extraction subnetwork (PSPaN), which are devised to extract both spatial and spectral information from the input PAN and MS images. Wherein, the frequency domain (FD) and intensity domain (ID) features of the PAN and MS image are leveraged as the guidance to improve the efficiency of feature extraction. Then, a joint spatial-spectral attention feature fusion module and a multi-stage residual reconstruction module are devised to efficiently harness the extracted spatial and spectral information. Finally, experiments are conducted to assess the effectiveness of our proposed MDGAPN.