Research on estimation models of rubber tree leaf nitrogen content based on hyperspectral and GWO-SVR

Abstract

Natural rubber is an essential economic crop in tropical parts. The quality and yield of rubber can be improved by proper fertilization. Therefore, it is crucial to develop a rapid and accurate model to detect the nitrogen content of rubber trees to improve quality and yield. This paper proposed a new and exemplary method for predicting nitrogen content by support vector regression (SVR) based on the Grey Wolf Optimizer (GWO). The successive projections algorithm (SPA) and competitive adapted reweighted sampling (CARS) were applied to choose the influential bands among the hyperspectral data, and the model was established using SVR. On the test data, the CARS-GWO-SVR model established by the GWO algorithm to optimize the SVR parameter penalty factor c and the kernel function parameter g has a prediction correlation coefficient R2 p=0.8967, and a prediction root mean square error RMSEP=0.2247. Comparative to the CARS-SVR model, The R2 p increased by 10.88%, and the RMSEP decreased by 9.15%. Therefore, GWO-SVR based on hyperspectral can establish a more accurate prediction model for the nitrogen content of rubber trees, providing technical support for the proper fertilization of rubber trees.