Spatial mutual information based hyperspectral band selection for classification

Area of Science:

• Remote Sensing
• Computer Vision
• Data Science

Background:

• Hyperspectral imaging generates large datasets, necessitating dimensionality reduction techniques.
• Hyperspectral band selection is crucial for efficient data processing and analysis.
• Existing methods like mutual information overlook spatial dependencies, limiting their effectiveness.

Purpose of the Study:

• To propose a novel hyperspectral band selection method incorporating spatial information.
• To enhance the robustness of band selection by considering spatial dependencies between pixels.
• To validate the proposed method's performance in hyperspectral data classification.

Main Methods:

• Development of a new band selection algorithm based on spatial mutual information.
• Utilizing supervised classification with Support Vector Machine (SVM) as a validation criterion.
• Experimental evaluation on hyperspectral datasets to assess classification accuracy.

Main Results:

• The proposed spatial mutual information method outperforms traditional methods.
• Improved classification accuracy was achieved using the novel band selection approach.
• The method effectively reduces redundancy while preserving important spatial-spectral information.

Conclusions:

• Spatial mutual information is a robust measure for hyperspectral band selection.
• The proposed method offers a significant advancement in hyperspectral data analysis.
• This technique enhances the accuracy and efficiency of hyperspectral image classification.