Residual channel attention based sample adaptation few-shot learning for hyperspectral image classification.

Area of Science:

• Remote Sensing
• Computer Vision
• Machine Learning

Background:

• Few-shot learning (FSL) is crucial for hyperspectral image classification (HSIC) to reduce the need for extensive labeled data.
• Existing FSL methods often fail to capture cross-domain feature channel correlations, leading to insufficient feature representation.
• Hyperspectral images contain rich spectral information but pose challenges for classification due to high dimensionality and limited labeled samples.

Purpose of the Study:

• To propose a novel FSL method, RCASA-FSL, for improved HSIC.
• To address the limitations of existing FSL techniques in capturing cross-domain feature channel correlations.
• To enhance the feature representation ability for more accurate hyperspectral image classification.

Main Methods:

• Introduced Residual Channel Attention Based Sample Adaptation Few-Shot Learning (RCASA-FSL) for HSIC.
• Developed a Deep Residual Feature Channel Attention Mechanism (DRFCAM) to capture cross-domain dependencies via residual concatenation and stacked residual structures.
• Implemented a Random-based Feature Recalibration Module (RFRM) to reassign feature weights using random matrices for guided sample adaptation.
• Designed a joint loss function combining FSL loss and domain adaptive loss for model optimization.

Main Results:

• The proposed RCASA-FSL method effectively captures and enhances cross-domain dependencies.
• DRFCAM and RFRM modules contribute to improved feature representation and discrimination.
• Experiments on standard hyperspectral datasets show RCASA-FSL outperforms other FSL techniques quantitatively and qualitatively.

Conclusions:

• RCASA-FSL offers a superior approach to few-shot hyperspectral image classification.
• The method's ability to capture cross-domain dependencies and enhance feature representation leads to significant performance gains.
• The proposed mechanisms provide a robust framework for addressing data scarcity in HSIC.