Proximal PanNet: A Model-Based Deep Network for Pansharpening

Recently, deep learning techniques have been extensively studied for pansharpening, which aims to generate a high resolution multispectral (HRMS) image by fusing low (LRMS) with panchromatic (PAN) image. However, existing learning-based pansharpening methods directly learn the mapping from LRMS and PAN HRMS. These network architectures always lack sufficient interpretability, limits further performance improvements. To alleviate this issue, we propose novel combining model-based methodology method. Firstly, build an observation model using convolutional sparse coding (CSC) technique design proximal gradient algorithm solve model. Secondly, unfold iterative into network, dubbed as Proximal PanNet, operators neural networks. Finally, all learnable modules can be automatically learned in end-to-end manner. Experimental results on some benchmark datasets show that our performs better than other advanced both quantitatively qualitatively.