DEVELOPMENT AND VALIDATION OF AN ECO-FRIENDLY ANALYTICAL METHOD DIGITAL IMAGE-BASED FOR QUALITY CONTROL OF NON-CENTRIFUGAL SUGAR

Non-centrifugal sugar is a natural sweetener obtained by thermal processing of sugarcane juice and it can exhibit wide sensory variability. In this work, the combination of digital image processing with machine learning was proposed as a fast, environmentally friendly, and accurate analytical tool for quality control of non-centrifugal sugar. For this purpose, the RGB patterns and color histograms were extracted of the digital images of non-centrifugal sugar samples and, together with the physicochemical parameters (dependent variables), were used for built predictive regression models using the k-nearest neighbors (K-NN) algorithm. The best results were obtained using the grayscale, RGB, and HSV histogram as analytical information, and the k-NN lazy algorithm had better predictability, indicating good performance and accuracy of the models created. The best-fitting model was determined by calculating the coefficients of determination (R2), which were 100% for the training/calibration models and greater than 80% for the testing/external validation models, and low root-mean-square error values of the prediction (RMSE) and mean absolute error (MAE) were obtained for all models. The use of digital images in combination with the k-NN algorithm can be considered as a fast, non-destructive, and green analytical method, which it can help manufacturers to offer products of different quality levels that meet the requirements of consumers and of food industries.