Data interpretation for spectral sensors with correlated bands

Area of Science:

• Remote Sensing
• Data Science
• Functional Analysis

Background:

• Emerging spectral sensors exhibit high correlation between band spectral response functions, exceeding 50% in some cases.
• Conventional geometrical models for spectral data analysis fail with such high correlations.
• Traditional sensors like Multispectral Thermal Images (MTI) and Landsat have lower band overlap.

Purpose of the Study:

• To present a generalized geometrical model for spectral data analysis using functional analysis.
• To define sensor and scene spaces for characterizing sensor suitability for spectral sensing tasks.
• To evaluate the performance of different classifiers with highly correlated spectral bands.

Main Methods:

• Development of a generalized geometrical model based on functional analysis.
• Definition of sensor space and scene space.
• Application and comparison of first-order distance/angle metrics and second-order statistical classifiers.
• Analysis of spectral data with high band correlation.

Main Results:

• Traditional geometrical models and first-order distance/angle classifiers fail with highly correlated spectral bands without preprocessing.
• Second-order statistical classifiers demonstrate robustness against problems introduced by correlated band responses.
• The generalized geometrical model and defined spaces aid in characterizing sensor suitability.

Conclusions:

• A generalized geometrical model and second-order statistical classifiers are essential for analyzing data from new spectral sensors with high band correlation.
• Appropriate preprocessing is necessary for first-order classifiers to function with correlated bands.
• The proposed framework enhances the suitability assessment of sensors for specific spectral sensing applications.