DDA: A dual-domain attention plug-and-play prior for pansharpening

Wenjie Shu, ZiEn Zhang

Keywords: image fusion, pansharpening, hyperspectral image super-resolution, deep learning, plug and play, attention mechanism

Abstract: Pansharpening is an image processing technique that enhances spatial resolution of multispectral images by fusing them with higher-resolution panchromatic images, becoming increasingly critical for remote sensing and geospatial analysis applications. Despite advancements, current deep learning algorithms for pansharpening face limitations: lack of global information extraction in the spatial domain and insufficient interaction across spectral channels. To tackle these challenges, we propose DDA, a dual-domain attention plug-and-play prior, integrating transformer attention with Convolutional Neural Networks, to facilitate better spatial and spectral detail integration.The code is available at Github.

Submission Number: 44