Hyperspectral Image Classification Based on Improved Rotation Forest Algorithm.

Area of Science:

• Remote Sensing
• Machine Learning
• Image Processing

Background:

• Hyperspectral image classification is crucial in remote sensing.
• Achieving high accuracy and generalization requires effective classification methods.
• Existing methods may face challenges in efficiency and multiclass handling.

Purpose of the Study:

• To propose an efficient hyperspectral image classification method named ROF-KELM.
• To improve classification accuracy and generalization capabilities.
• To leverage the strengths of Kernel Extreme Learning Machine and Rotation Forest.

Main Methods:

• Feature segmentation using Non-negative Matrix Factorization (NMF).
• Utilizing Kernel Extreme Learning Machine (KELM) as the base classifier.
• Employing an improved Rotation Forest (ROF) ensemble with Q-statistic for base classifier selection and voting for final classification.

Main Results:

• The proposed ROF-KELM method demonstrates effectiveness in hyperspectral image classification.
• Validation on AVIRIS, ROSIS, and UCI datasets shows promising results.
• The method achieves high accuracy and strong generalization.

Conclusions:

• ROF-KELM offers an efficient and effective approach for hyperspectral image classification.
• The integration of NMF, KELM, and ROF provides a robust classification framework.
• The method is suitable for complex remote sensing data analysis.