[The estimation model of rice leaf area index using hyperspectral data based on support vector machine].

Area of Science:

• Agricultural Science
• Remote Sensing
• Machine Learning

Context:

• Hyperspectral reflectance data (350-2500 nm) were collected for rice cultivars under varying nitrogen levels.
• Ten vegetation indices (VIs) were derived from hyperspectral data to assess their utility in predicting rice Leaf Area Index (LAI).
• Ground-based hyperspectral measurements were taken at canopy level under controlled field conditions.

Purpose:

• To compare the predictive capabilities of statistical models and Support Vector Machine (SVM) techniques for estimating rice LAI.
• To evaluate the performance of various vegetation indices (VIs) as inputs for both statistical and SVM models.
• To identify the optimal VI and SVM kernel function for accurate rice LAI estimation.

Summary:

• Hyperspectral data were used to derive ten vegetation indices (VIs), which were then employed in statistical models and SVMs to predict rice Leaf Area Index (LAI).
• Three statistical models (exponent, power) and corresponding SVM models with ANOVA, POLY, and RBF kernels were analyzed.
• SVM models consistently outperformed statistical models, with SVM utilizing the TCARI/OSAVI index and a polynomial kernel achieving the highest estimation precision (11% lower RMSE).

Impact:

• Support Vector Machine (SVM) demonstrates high accuracy and robustness for estimating rice LAI from hyperspectral data.
• The study highlights SVM as a valuable tool for enhancing the understanding of relationships between vegetation indices and crop biophysical parameters.
• Findings contribute to improved crop monitoring and precision agriculture applications through advanced data analysis techniques.