Near-Infrared Technology in Agriculture: Non-Destructive Determination of Inner Quality Parameters in Intact Cocoa Beans

Abstract

This present study aimed to apply the near-infrared technology based on reflectance spectroscopy or NIRS in determining 2 main quality attributes on intact cocoa beans namely fat content (FC) and moisture content (MC). Absorbance spectral data, in a wavelength range from 1000 to 2500 nm were acquired and recorded for a total of 110 bulk cocoa bean samples. Meanwhile, actual reference FC and MC were obtained using standard laboratory approaches and Soxhlet and Gravimetry methods. Samples were split onto calibration and validation datasets. The prediction models, used to determine both quality attributes were developed from the calibration dataset using 2 regression methods: Principal component regression (PCR) and partial least square regression (PLSR). To obtain more accurate and robust prediction performance, 4 different spectra correction methods namely baseline shift correction (BSC), mean normalization (MN), standard normal variate (SNV), and orthogonal signal correction (OSC) were employed. The results showed that PLSR was better than PCR for both quality parameters prediction. Moreover, spectra corrections enhanced the prediction accuracy and robustness from which OSC was found to be the best correction method for FC and MC determination. The prediction performance using validation dataset generated a correlation coefficient (r), ratio prediction to deviation (RPD), and ratio error to range (RER) indexes for FC were 0.93, 3.16 and 7.12, while for MC prediction, the r coefficient, RPD and RER indexes were 0.96, 3.43 and 9.25, respectively. Based on obtained results, it may conclude that NIRS combined with proper spectra correction and regression approaches can be used to determine inner quality attributes of intact cocoa beans rapidly and simultaneously. HIGHLIGHTS We study and apply NIRS technology as a fast and novel method to predict inner quality parameters of intact cocoa beans in form of moisture and fat content Prediction models, used to determine both quality attributes were developed using 2 regression methods: Principal component regression (PCR) and partial least square regression (PLSR) To obtain more accurate and robust prediction performance, 4 different spectra correction methods namely baseline shift correction (BSC), mean normalization (MN), standard normal variate (SNV), and orthogonal signal correction (OSC) The best prediction performance was obtained when the models are constructed using PLSR in combination with OSC correction approach The maximum correlation coefficient (r) and ratio prediction to deviation (RPD) indexes for Fat content were 0.93 and 3.16, while for moisture content prediction, the r coefficient and RPD indexes were 0.96 and 3.43, respectively GRAPHICAL ABSTRACT