Effect of different regression algorithms on the estimating leaf parameters based on selected characteristic...

Area of Science:

• Plant physiology
• Remote sensing
• Biophysical modeling

Background:

• Accurate monitoring of leaf biochemical contents (LBC) is crucial for understanding plant health and productivity.
• Traditional methods for LBC assessment are often time-consuming and labor-intensive.
• Developing efficient, non-destructive methods for LBC estimation is a significant research objective.

Purpose of the Study:

• To identify characteristic wavelengths sensitive to key leaf biochemical parameters.
• To evaluate the performance of various regression algorithms for estimating LBC using these wavelengths.
• To determine optimal parameters for regression models and compare their accuracy across different datasets.

Main Methods:

• Sensitivity analysis based on a physical model to determine characteristic wavelengths for carotenoid, chlorophyll, dry matter, water thickness, and leaf structure.
• Performance evaluation of six regression algorithms (Random Forest, BPNN, SVR, RBFNN, PLSR, GPR) with varying parameters.
• Optimization of regression algorithm parameters for LBC estimation.

Main Results:

• Ten characteristic wavelengths were identified as optimal for monitoring LBC.
• Gaussian Process Regression (GPR) demonstrated superior performance in estimating LBC compared to other algorithms.
• Optimal parameters for each regression algorithm were determined, enhancing their predictive capabilities.

Conclusions:

• The combination of 10 selected characteristic wavelengths and the GPR model provides an efficient and accurate method for estimating LBC.
• This approach offers a promising non-destructive technique for plant physiological status assessment.
• The findings contribute to advancements in remote sensing applications for agriculture and ecological monitoring.