Evaluation of a combination of NIR micro-spectrometers to predict chemical properties of sugarcane forage using a multi-block approach

Abstract

Forage quality is essential in livestock farming and has an important role in the functioning of agricultural farms. Access to biochemical variables provides an estimation of the feed value of crop for animal feed at harvest. Near infrared (NIR) spectroscopy provides measurements indirectly related to biochemical variables. In recent years, several micro-spectrometers have been developed that offer the opportunity to predict such biochemical variables at low cost. In this study, the potential of a combination of micro-spectrometers is evaluated to predict crude protein (CP) and total sugar content (TS) of sugarcane. First, each micro-spectrometer with optimal pretreatments was individually compared to a reference laboratory spectrometer. Then, a combination of micro-spectrometers is proposed and prediction models were established by a multi-block method from data fusion called Sequential and Orthogonalised - Partial Least Squares (SO-PLS). For CP, the combination of micro-spectrometers provides model (sep = 0.69%; bias = 0.15%; R t e s t 2 = 0.910) close to those obtained with the reference spectrometer (sep = 0.56%; bias = −0.13%; R t e s t 2 = 0.935). For TS, the results obtained with this combination of micro-spectrometers (sep = 2.38%; bias = −0.52%; R t e s t 2 = 0.983) are better than those obtained with the reference spectrometer (sep = 2.59%; bias = 0.41%; R t e s t 2 = 0.978). For both chemical variables, the combination of the micro-spectrometers significantly increases the performance of the predictive models compared to the models obtained with the micro-spectrometers independently. Using several low-cost micro-spectrometers, combined with a multi-block method would give results as good as a single laboratory spectrometer with a lower cost. • A combination of micro-spectrometers is proposed as a low-cost solution. • The combination significantly increases prediction quality. • The case used is for the prediction of crude protein and sugar content of sugarcane. • A multi-block regression method called SO-PLS is used. • Application to predict chemical properties of sugarcane forage from NIRS.