MambaFedCD: Spatial-Spectral-Temporal Collaborative Mamba-Based Active Federated Hyperspectral Change Detection.

Area of Science:

• Remote Sensing
• Computer Vision
• Artificial Intelligence

Background:

• Hyperspectral image (HSI) change detection identifies alterations in geographic areas using bitemporal HSIs.
• Centralized learning methods require data transmission, posing privacy risks for sensitive remote sensing data.
• Manual sample labeling for training is resource-intensive and impractical.

Purpose of the Study:

• To propose a privacy-preserving federated learning framework for hyperspectral image change detection.
• To leverage limited labeled samples from multiple clients for effective change detection.
• To address the challenges of data sensitivity and labeling costs in HSI change detection.

Main Methods:

• A spatial-spectral-temporal collaborative Mamba-based change detection (S2TMamba) model integrates temporal and spatial-spectral information.
• A difference feature diversity correction-based model aggregation (DFDCMA) strategy optimizes global model aggregation.
• A multi-decision federated active learning (MDFAL) strategy efficiently selects informative samples for training.

Main Results:

• The proposed MambaFedCD framework achieves superior performance compared to state-of-the-art methods.
• Experiments demonstrate the effectiveness of the S2TMamba model, DFDCMA strategy, and MDFAL approach.
• The framework successfully ensures data privacy while performing accurate change detection.

Conclusions:

• MambaFedCD offers an effective and privacy-preserving solution for hyperspectral image change detection.
• The synergistic use of spatial, spectral, and temporal information is crucial for accurate change detection.
• Federated learning combined with active learning presents a promising direction for HSI analysis.