Graph Information Aggregation Cross-Domain Few-Shot Learning for Hyperspectral Image Classification.

Area of Science:

• Computer Vision
• Machine Learning
• Remote Sensing

Background:

• Domain adaptation (DA) in hyperspectral image (HSI) classification often struggles with new classes in target data.
• Existing DA methods primarily use local spatial information, neglecting nonlocal spatial relationships.
• Cross-scene HSI classification faces performance degradation due to domain shift.

Purpose of the Study:

• To develop a robust domain adaptation method for cross-scene HSI classification that handles new classes in target data.
• To integrate nonlocal spatial information into domain alignment for improved classification performance.
• To combine few-shot learning (FSL) with graph-based domain alignment for enhanced HSI classification.

Main Methods:

• A graph information aggregation cross-domain few-shot learning (Gia-CFSL) framework is proposed.
• Intradomain distribution extraction (IDE) block and cross-domain similarity aware (CSA) block are designed to capture nonlocal relationships and interdomain similarities.
• Feature-level and distribution-level cross-domain graph alignments are employed to mitigate domain shift impacts.

Main Results:

• The proposed Gia-CFSL framework significantly improves classification performance on HSI datasets with new classes in the target domain.
• Incorporating nonlocal spatial information enhances the effectiveness of domain alignment.
• Experimental results demonstrate the superiority of the proposed method over existing DA techniques.

Conclusions:

• The Gia-CFSL framework offers a promising solution for cross-scene HSI classification, particularly in few-shot learning scenarios with domain shift.
• Integrating graph-based nonlocal information aggregation is crucial for robust domain adaptation.
• The method effectively addresses challenges posed by new classes and domain variations in HSI data.