Prediction and classification of sugar content of sugarcane based on skin scanning using visible and shortwave near infrared

Abstract

The potential application of a visible and shortwave near infrared (Vis/SWNIR) spectroscopic technique as a low cost alternative to predict sugar content based on skin scanning was evaluated. Two hundred and ninety one internode samples representing three different commercial sugarcane varieties were used. Each sample was scanned at four scanning points to obtain the spectra data which was later correlated with its °Brix (soluble solids content) values. Partial least square (PLS) model was developed and applied to both calibration and prediction samples. Using reflectance spectra data, the model had a coefficient of determination (R2) of 0.91 and root means square error of predictions (RMSEP) of 0.721 °Brix. The artificial neural network (ANN) was also applied to classify spectra data into five °Brix categories. The ANN has yielded good classification performance, ranging from 50 to 100% accuracy with an average accuracy of 83.1%. These results demonstrated that the Vis/SWNIR spectroscopy technique could be applied to predict sugarcane °Brix in the field based skin scanning method.