A Joint Network of Edge-Aware and Spectral-Spatial Feature Learning for Hyperspectral Image Classification

Area of Science:

• Remote Sensing
• Computer Vision
• Machine Learning

Background:

• Hyperspectral image (HSI) classification is crucial for HSI applications.
• Extracting deep representation features from rich spectral information in HSIs is challenging.
• Existing methods often introduce noise when augmenting edge data and fail to emphasize spectral importance.

Purpose of the Study:

• To propose an edge-aware and spectral-spatial feature learning network (ESSN) for improved HSI classification.
• To address challenges in deep feature extraction, edge representation, and spectral importance weighting.
• To enhance accuracy and robustness against interference in HSI classification.

Main Methods:

• Developed an edge-aware and spectral-spatial feature learning network (ESSN).
• Incorporated an edge feature augment block to adaptively strengthen edge features across spectral bands.
• Implemented a spectral-spatial feature extraction block with adaptive spectral weight adjustment.

Main Results:

• Extensive experiments were conducted on three public hyperspectral datasets.
• The proposed ESSN method demonstrated higher classification accuracy compared to state-of-the-art methods.
• ESSN showed improved immunity to interference.

Conclusions:

• The ESSN network effectively extracts deep representation features for HSI classification.
• Adaptive edge feature augmentation and spectral weighting significantly improve performance.
• ESSN offers a robust and accurate solution for hyperspectral image classification challenges.