Pansharpening via Double Nonconvex Tensor Low-Tubal-Rank Priors

In this letter, based on the tensor representation modeling, we propose a novel and strict tensor-based pansharpening model via double nonconvex tensor low-tubal-rank (DNTLTR) priors for the fusion of low resolution multispectral (LRMS) and panchromatic (Pan) images to produce the high-resolution MS (HRMS) images. By modeling the MS image as a third-order tensor for better modeling its spatial-spectral structural correlations, we particularly exploit the tensor low-tubal-rank properties of HRMS as well as the difference of HRMS and Pan at the same time, and then propose a novel unified log tensor nuclear norm-based DNTLTR prior term. Moreover, for the spectral preservation of LRMS image, we also impose the spatial degradation-based spectral fidelity constraint between HRMS and LRMS. Then, we apply the alternating direction method of multiplier to optimize the proposed DNTLTR model. Finally, we show both the reduced-scale and full-scale fusion experiments to validate the effectiveness of DNTLTR visually and quantitatively.

I. Introduction

Pansharpening is to fuse the low-resolution multispectral (LRMS) and panchromatic (Pan) images to produce the high-resolution MS (HRMS) images by preserving the spectral information of LRMS and the spatial structures of pan simultaneously, which is an important and hot topic of resolution enhancement in remote sensing [1].