SSMIF: Enhanced Spatial-Spectral Mamba Interactive Fusion Network for Hyperspectral Change Detection

Change detection (CD) in remote sensing images is a complex task. Recent advancements in convolutional neural networks (CNNs) and transformer-based methods have significantly improved CD accuracy. However, hyperspectral image (HSI) CD remains particularly challenging, especially in terms of change feature extraction and fusion. In this paper, we present an enhanced spatial-spectral mamba interactive fusion (SSMIF) network specifically designed for HSI CD. This network incorporates the state space model (SSM) and flow alignment techniques to extract and fuse features more effectively. In particular, the spatial and spectral mamba (SASM) model is employed to capture spatial and spectral features from bi-temporal HSIs. Furthermore, we introduce a bi-temporal flow alignment (BFA) method to resample images for improved feature alignment. An additional long short-term memory (LSTM) module is used to filter important features and reduce redundancy. Experimental results on HSI CD datasets show that the proposed SSMIF network consistently outperforms several state-of-the-art approaches. The source code of the proposed SSMIF will be released at https://github.com/creativeXin/SSMIF.