Hybrid Attention Based Residual Network for Pansharpening

Pansharpening aims at fusing the rich spectral information of multispectral (MS) images and spatial details panchromatic (PAN) to generate a fused image with both high resolutions. In general, existing pansharpening methods suffer from problems distortion lack detail information, which might prevent accuracy computation for ground object identification. To alleviate these problems, we propose Hybrid Attention mechanism-based Residual Neural Network (HARNN). proposed network, develop an encoder attention module in feature extraction part better utilize features MS PAN images. Furthermore, fusion is designed improve contour image. A series ablation contrast experiments are conducted on GF-1 GF-2 datasets. The results less distorted pixels more demonstrate that HARNN can implement task effectively, outperforms state-of-the-art algorithms.