Development of non-destructive mango assessment using Handheld Spectroscopy and Machine Learning Regression

Abstract

Abstract Quality determines the shelf-life and selling prices of fresh mango, and therefore quality observation and control of fresh mango are of utmost significance in the processing and management of its supply chain. Mango fruit (mangifera indica) quality methods are mostly destructive in nature. Different mechanical, electromagnetic and non-destructive methods are increasingly important nowadays because of the ease of operation, speed, and reliability of the process. This project aims to develop a non-destructive assessment of mango quality using handheld micro NIR (near-infrared) spectroscopic device. NIR spectra data and Brix levels, which indicate the sugar content of the plant, i.e. indicating the sweetness of the mango, were collected from three different types of Mango (Chokanan, Rainbow, and Kai Te), resulting 80 samples (i.e. 60 samples for training and 20 samples for testing) in this project. NIR spectra can be converted mathematically to obtain quantitative information of chemical and physical nature by multivariate calibration. The spectra data is pre-processed using Gaussian smoothing and extended multiplicative signal correction (EMSC) for the elimination of uncontrollable path length or scattering effects. These samples were then used to develop a predictive model using both Support Vector Machine (SVM) regression and Partial Least Squares regression (PLS) methods. The coefficient of determination (R2) obtained from SVM for training/calibration and testing dataset are 0.96 and 0.95 respectively. Meanwhile, the coefficient of determination (R2) obtained from PLS for calibration/training and testing dataset are 0.89 and 0.86 respectively. The results obtained from this project indicate that the handheld NIR has potential use for non-destructive assessment of mango fruits quality.