Spectral-Spatial-Temporal Kolmogorov-Arnold Network for Hyperspectral Change Detection.

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Artificial Intelligence in Earth Observation

Background:

• Hyperspectral change detection (HCD) is vital for monitoring Earth's surface using remote sensing data.
• Existing Convolutional Neural Network (CNN) and transformer models struggle with intricate spectral-spatial relationships and feature distribution discrepancies in hyperspectral images (HSIs).

Purpose of the Study:

• To propose a novel Spectral-Spatial-Temporal Kolmogorov-Arnold Network (SSTKAN) for improved HCD.
• To effectively model complex spectral-spatial relationships and mitigate feature distribution discrepancies in bitemporal HSIs.

Main Methods:

• A spectral-spatial Kolmogorov-Arnold network (KAN) extracts spectral-spatial features.
• A 3-D KAN captures temporal and difference features.
• Second-order statistical alignment reduces feature distribution discrepancies, and a multiscale feature enhancement (MSFE) module strengthens representations.

Main Results:

• The proposed SSTKAN demonstrates superior performance on four public hyperspectral datasets.
• Both qualitative and quantitative results show advancements over existing change detection (CD) approaches.
• The method effectively handles spectral-spatial complexities and environmental variations.

Conclusions:

• The SSTKAN offers a significant advancement in hyperspectral change detection.
• The network's ability to model complex relationships and align feature distributions leads to enhanced detection accuracy.
• This research provides a robust framework for future HCD studies.