a comparative study concerning linear and nonlinear models to determine sugar content in sugar beet by near infrared spectroscopy (nir)

This paper reports on the use of artificial neural networks (ann) and partial least square regression (pls) combined with nir spectroscopy (900-1700 nm) to design calibration models for the determination of sugar content in sugar beet. in this study a total of 80 samples were used as the calibration set, whereas 40 samples were used for prediction. three pre-processing methods, including multiplicative scatter correction (msc), first and second derivatives were applied to improve the predictive ability of the models. models were developed using partial least squares and artificial neural networks as linear and nonlinear models, respectively. the correlation coefficient (r), sugar mean square error of prediction (rmsep) and sdr were the factors used for comparing these models. the results showed that nir can be utilized as a rapid method to determine soluble solid content (ssc), sugar content (sc) and the model developed by ann gives better correlation between predictions and measured values than pls.