Adaptive affinity propagation with spectral angle mapper for semi-supervised hyperspectral band selection.

Area of Science:

• Remote Sensing
• Computer Science
• Data Science

Background:

• Dimensionality reduction is crucial in hyperspectral imagery analysis.
• Affinity Propagation (AP) is a clustering algorithm with potential for hyperspectral band selection.
• A limitation of AP is its inability to determine a fixed number of exemplars, hindering its application.

Purpose of the Study:

• To propose an adaptive AP (AAP) algorithm for semi-supervised hyperspectral band selection.
• To investigate the impact of distance metrics on band selection performance.
• To enhance the AP algorithm for more controlled and effective band selection.

Main Methods:

• Developed an adaptive AP (AAP) algorithm incorporating an exemplar number determination algorithm and bisection method.
• Established relationships between selected exemplar numbers and user preferences.
• Evaluated AAP using various distance metrics for hyperspectral band selection.

Main Results:

• The proposed AAP algorithm significantly improves hyperspectral band selection.
• AAP demonstrates superior performance compared to other popular band selection methods.
• The method achieves lower computational cost and robust results in experiments.

Conclusions:

• The adaptive AP algorithm offers an effective solution for semi-supervised hyperspectral band selection.
• AAP overcomes the limitations of traditional AP by enabling fixed exemplar selection.
• The proposed approach provides a more computationally efficient and robust method for hyperspectral data analysis.