Rapeseed Nitrogen Status Estimation of Vis-NIR Spectra Based on Partial Least Square and BP Neural Network

Abstract

A combined estimation model, artificial neural network (ANN) combined with partial least square regression (PLS) method, has been developed for estimation of rapeseed N status. Spectra tests were performed on the rapeseed canopy of 150 samples in the field using a spectrophotometer (325-1075 nm). 5 optimal PLS principal components were determined by PLS analysis with cross-validation. They were selected as the input of BP neural network to establish the prediction model. The node number of input layer, hidden layer, and output layer was 5, 5, and 1. 110 samples were used as training set and the left 40 samples formed prediction set. The result showed that the prediction performance was excellent with the correlation value of 0.95405, higher than the result (0.8764) obtained only by using PLS method. Most of the relative standard deviation (RSD) was under 5% and the accuracy of prediction reached 95%. Thus, it was concluded that the proposed PLS-ANN model for the spectroscopic estimation of rapeseed N status was superior to other existing spectroscopic methods based on Vis/NIRS.