Evolving multispectral sensor configurations using genetic programming for estuary health monitoring

Area of Science:

• Environmental Science
• Remote Sensing
• Data Science

Background:

• Large-scale ecosystem health assessment is vital for management and regulatory decisions.
• Hyperspectral imaging offers noninvasive estimation of attributes but generates high-dimensional data.
• Real-world applications necessitate models using fewer wavelengths for hyperspectral image analysis.

Purpose of the Study:

• To propose a novel wavelength selection and feature extraction method for hyperspectral image analysis.
• To automatically identify key wavelength regions and informative image features using genetic programming.
• To predict sediment porosity and organic matter content using the developed method.

Main Methods:

• Developed a genetic programming approach for wavelength selection and feature extraction in hyperspectral images.
• Collected field hyperspectral images of sediment paired with ground-truth porosity and organic matter content.
• Proposed two program structures (spectra-based and image-based) for feature extraction tree construction and utilized Support Vector Regression (SVR) models for prediction.

Main Results:

• The proposed spectra-based genetic programming method demonstrated competitive performance against established wavelength selection techniques (SPA, CARS, RC).
• Support Vector Regression models effectively predicted sediment porosity and organic matter content based on extracted features.
• Full-wavelength models showed reliability in predicting sediment organic matter content.

Conclusions:

• The developed genetic programming method offers an effective solution for reducing hyperspectral data dimensionality while retaining predictive power.
• The approach successfully predicted sediment organic matter content across all collected images, demonstrating its practical applicability.
• This method enhances the utility of hyperspectral imaging for large-scale environmental monitoring and assessment.