Dynamic weighted residual ensemble learning for hyperspectral image classification driven by features and samples.

Area of Science:

• Remote Sensing
• Computer Vision
• Machine Learning

Background:

• Hyperspectral image classification faces challenges in selecting relevant features and informative samples.
• Dynamic ensemble selection offers a promising avenue for improving classification performance.

Purpose of the Study:

• Introduce two novel dynamic residual ensemble learning methods: MF-DWRL and FS-DWRL.
• Address the challenges of feature and sample selection in hyperspectral image classification.

Main Methods:

• MF-DWRL uses multi-feature combinations and K-Nearest Neighbors to identify optimal feature sets and guide residual adjustments.
• FS-DWRL enhances performance by jointly optimizing feature combinations and informative sample selection.
• Both methods employ dynamic ensemble selection with weighted residuals.

Main Results:

• MF-DWRL and FS-DWRL achieve high classification accuracies on three hyperspectral datasets (China-WHU-Hi-HanChuan, WHU-Hi-LongKou, WHU-Hi-HongHu).
• Specific accuracies reached 90.57%, 98.77%, and 91.08% respectively.
• The proposed methods demonstrate significant improvements over existing state-of-the-art techniques.

Conclusions:

• MF-DWRL and FS-DWRL effectively enhance hyperspectral image classification accuracy.
• Joint optimization of features and samples (FS-DWRL) leads to superior performance.
• These dynamic ensemble learning methods represent a significant advancement in the field.