Caffeine content calibration model on green beans arabica Mandailing Natal coffee using NIRS and artificial neural network

Abstract

Generally, Near Infrared Reflectance (NIR) spectroscopy is a non-destructive method that can identify the chemical content of agricultural products. In this study, several pretreatment data were used to reduce some noises from original NIR spectra, namely normalization (0-1) and Multiple Scatter Correction (MSC) and calibration model of caffeine content using Artificial Neural Network (ANN). Result from wavelength analysis showed that the caffeine content of Mandailing Natal green bean arabica coffee was in the wavelengths of 1,208.9 nm and 1,728.91 nm. A good ANN architecture result was obtained with normalization/MSC which was 13-3-1 at 1,000 iterations, which was indicated by r = 0.9845, CV = 4.80% and RMSEC-RMSEP= 0.0383.