Research Assistant Remote Sensing Lab, Saint Louis University
Disclosure: Disclosure information not submitted.
Estimation of leaf chlorophyll concentration (LCC) is an important indicator of plant health, vigor, physiological status, productivity, and nutrient deficiencies. Hyperspectral spectroscopy at leaf level has been widely used to estimate LCC; however, the model estimates may suffer from the noise incurred within raw spectral information. In this study, we used the fractional derivative (FD) analysis of raw spectral reflectance to extract the most useful features for estimating LCC of Sorghum. We derived FDs from 0.2-2.0 orders with 0.2 order increment. The models were built using partial least squares regression (PLSR), random forest regression (RFR), support vector regression (SVR), and extreme learning regression (ELR) from FD augmented wavelengths. Different numbers of sensitive features were selected based on Pearson’s correlation coefficients, partial least square-based variable importance in the projection, and random forest-based feature importance before performing the model training. The models were optimized using different parameters while training and test results were yielded from the best performing training parameters. Results showed that (1) FD-augmented wavelengths yielded improved model performance compared to raw spectral data; (2) SVR resulted in the best performance with different feature selection methods and FD orders; (3) partial least square-based variable importance in the projection was found as the best feature selection method with 50 features.