Hyperspectral image classification using functional data analysis.

Area of Science:

• Remote Sensing
• Data Science
• Computer Vision

Background:

• Hyperspectral imaging provides abundant spectral bands for detailed object and material discrimination.
• Traditional hyperspectral image classification methods often operate within a multivariate analysis framework.
• There is a need for advanced methods to fully leverage the rich spectral information and spatial context in hyperspectral data.

Purpose of the Study:

• To propose a novel hyperspectral image classification method based on functional data analysis (FDA).
• To demonstrate the effectiveness of treating spectral data as continuous functions for improved classification.
• To enhance the utilization of spectral information and inter-pixel relationships in hyperspectral images.

Main Methods:

• The proposed method utilizes functional data analysis (FDA) by viewing each pixel's spectral curve as a continuous function.
• Functional principal component analysis (FPCA) is employed to classify these spectral functions.
• The approach considers the relevance between adjacent pixel elements to enhance classification performance.

Main Results:

• Experimental results on three distinct hyperspectral image datasets were evaluated.
• The proposed FDA-based method achieved higher classification accuracies compared to existing state-of-the-art techniques.
• The method effectively leverages the continuous nature of spectral data and spatial correlations.

Conclusions:

• Functional data analysis offers a powerful framework for hyperspectral image classification.
• The proposed method demonstrates superior performance by capitalizing on spectral continuity and pixel relationships.
• This approach represents a significant advancement in accurate hyperspectral image analysis and classification.