Novel Feature-Extraction Methods for the Estimation of Above-Ground Biomass in Rice Crops

Area of Science:

• Agricultural Remote Sensing
• Plant Physiology
• Geospatial Analysis

Background:

• Traditional methods for measuring above-ground biomass dynamics (AGBD) often rely on vegetation indices derived from multispectral imagery, which can be limited in accuracy.
• Accurate AGBD estimation is crucial for crop management and yield prediction throughout the phenological cycle.

Purpose of the Study:

• To compare two feature extraction methods, GFKuts and Graph-Based Data Fusion (GBF), for non-destructive biomass estimation using aerial multispectral imagery.
• To evaluate the performance of these methods across different rice growth stages.

Main Methods:

• GFKuts: Utilizes Gaussian mixture models, Monte Carlo K-means, and guided image filtering for vegetation index extraction.
• GBF-Sm-Bs: A novel approach that does not rely on calculating vegetation-index image reflectances.
• Experimental comparison using ground-truth biomass measurements from destructive sampling.

Main Results:

• The GBF-Sm-Bs approach achieved a high correlation of 0.995 (R²=0.991) with ground-truth biomass measurements.
• The proposed method demonstrated a Root Mean Square Error (RMSE) of 45.358 g.
• This resulted in a precision increase of approximately 62.43% compared to existing methods.

Conclusions:

• The Graph-Based Data Fusion (GBF-Sm-Bs) method offers a superior approach for non-destructive above-ground biomass estimation.
• This advanced technique significantly enhances the accuracy and precision of biomass measurements in agricultural applications.
• The findings pave the way for more efficient crop monitoring and yield forecasting.