Hyperspectral Pansharpening via Mamba-Based Feature Extraction and Cross-Modal Injection

Abstract

Hyperspectral (HS) pansharpening integrates the HS image with a high-spatial-resolution panchromatic (PAN) image, which provides a good solution for improving the spatial resolution of the HS image. However, current fusion methods based on convolutional neural networks (CNNs) and Transformers cannot balance global information modeling and computational efficiency. To address these problems, we propose a Mamba-based HS pansharpening network (MHPN), which integrates Mamba modeling into a multiresolution analysis (MRA) framework. Specifically, a feature extraction Mamba module (MFE) is designed for local and global feature extraction from HS and PAN images with high computational efficiency. A cross-modal injection Mamba module (CMIM) is proposed, in which the high-frequency (HH) spatial information of the PAN image is injected into the HS image via the discrete wavelet transform (DWT) to enable effective cross-modal information injection and fusion. Experimental results on four simulated and real datasets demonstrate the effectiveness of the proposed network.