Hyperspectral Image Classification via Discriminant Gabor Ensemble Filter.

Area of Science:

• Remote Sensing
• Computer Vision
• Machine Learning

Background:

• Hyperspectral image (HSI) classification is crucial for remote sensing applications.
• Convolutional neural networks (CNNs) show promise but require extensive labeled data.
• Traditional Gabor filters extract spatial information but may miss discriminative features.

Purpose of the Study:

• To propose a novel Gabor ensemble filter (GEF) for efficient deep feature extraction in HSI classification.
• To develop a CNN architecture utilizing GEF that requires fewer trainable parameters and less training data.
• To enhance feature discriminability and achieve end-to-end learning for HSI classification.

Main Methods:

• Introduced the Gabor ensemble filter (GEF) combining fixed Gabor filters and learnable filters for feature extraction.
• Designed a CNN architecture incorporating GEF for HSI classification with reduced parameter count.
• Integrated local discriminant structure with triplet hard loss for improved feature learning.

Main Results:

• The proposed GEF-based method achieved significantly higher classification accuracy on three HSI datasets compared to state-of-the-art methods.
• The network demonstrated effective deep feature extraction and learning from limited training samples.
• The method exhibited speedy training and testing performance.

Conclusions:

• The GEF method offers a powerful and efficient solution for hyperspectral image classification.
• This approach effectively addresses the challenge of limited labeled data in deep learning for HSI.
• The proposed method provides a promising direction for advancing remote sensing data analysis.