< Home      < Project Home       

Detail Injection-based Deep Convolutional Neural Networks for Pansharpening

Liang-Jian Deng, Gemine Vivone*, Cheng Jin, Jocelyn Chanussot


Paper & Code


The fusion of high spatial resolution panchromatic data with simultaneously acquired multispectral data with lower spatial resolution is a hot topic, which is often called pansharpening. In this paper, we exploit the combination of machine learning techniques and fusion schemes introduced to address the pansharpening problem. In particular, deep convolutional neural networks are proposed to solve this issue. These latter are combined first with the traditional component substitution and multi-resolution analysis fusion schemes in order to estimate the non-linear injection models that rule the combination of the upsampled low resolution multispectral image with the extracted details exploiting the two philosophies. Furthermore, inspired by these two approaches, we also developed another deep convolutional neural network for pansharpening. This is fed by the direct difference between the panchromatic image and the upsampled low resolution multispectral image. Extensive experiments conducted both at reduced and full resolutions demonstrate that this latter convolutional neural network outperforms both the other detail injection-based proposals and several state-of-the-art pansharpening methods.


The flowchart of FusionNet method for multispectral image pansharpening




Bib Citation

author={Liang-Jian Deng, Gemine Vivone, Cheng Jin, Jocelyn Chanussot},
journal={IEEE Transactions on Geoscience and Remote Sensing},
title={Detail Injection-Based Deep Convolutional Neural Networks for Pansharpening},
pages={DOI: 10.1109/TGRS.2020.3031366},