Improved technique for retrieval of forest parameters from hyperspectral remote sensing data

Area of Science:

• Remote Sensing
• Forestry
• Machine Learning
• Biomass Estimation

Background:

• Accurate land surface object recognition and biological parameter retrieval are crucial for forest monitoring.
• Hyperspectral imaging offers rich spectral and spatial information for detailed analysis.
• Existing methods may lack precision in complex forest environments.

Purpose of the Study:

• To develop and apply machine learning pattern recognition for land surface object identification using hyperspectral data.
• To retrieve biological parameters, specifically forest phytomass and biomass, for recognized forest classes.
• To enhance spatial and spectral analysis through a modified Bayesian classifier.

Main Methods:

• Utilized machine learning pattern recognition on spectral and textural features from hyperspectral images.
• Employed a modified Bayesian classifier for improved spatial and spectral domain analysis.
• Solved atmospheric optics problems using modeled forest canopy projective cover and density.

Main Results:

• Successfully detected and delineated forest object classes within high spectral and spatial resolution images.
• Enabled retrieval of phytomass (leaves/needles) and total canopy biomass for identified forest types.
• Demonstrated the effectiveness of the modified Bayesian classifier in improving recognition procedures.

Conclusions:

• The proposed machine learning approach effectively identifies forest objects and quantifies biomass from hyperspectral imagery.
• The modified Bayesian classifier enhances the accuracy of spatial and spectral analysis in forest remote sensing.
• This technique provides valuable data for forest management and ecological studies.