This repository contains code for the multi-patch self-attention network created to tackle the problem of Pansharpening.
Pansharpening is the problem of increasing the spatial resolution of Multi-Spectral images by fusing it with the corresponding Panchromatic images of the same scene.
In this paper, we present a novel architecture inspired from Transformer Networks, for the problem of Pansharpening. We introduce a 'Multi-Patch Attention' mechanism which calculates channel-wise attention on non-overlapping smaller patches of the image to capture local-level details.
We evaluated our model on 2 datasets created from IKONOS and LANDSAT-8 imagery, provided access to us under the Third Party Missions by the European Space Agency, and by Space Applications Centre (SAC) of Indian Space Research Organization (ISRO), respectively.