Nu-Support Vector Classification Training for Feature Identification in "Arepas": A Colombian Traditional Food

The food industry is interest to characterizing the appearance of products that are attractive and popular among the population. For that, it is important to objectively measure certain physical characteristics of food matrices to determine the consumer acceptability. It is known that computer vision has been used to measure physical and chemical characteristics in various foods, obtaining percentages of accuracy higher than 95.00 %. In Colombia, a traditional and highly consumed food is the "arepa", which can be made with corn flour (yellow or white) with different preparation techniques (raw, baked or fried). In this research, the identification of the type of corn flour used and the preparation technique of arepas using computer vision was proposed. For this purpose, information was collected from 90 samples of arepas by training the Nu Support Vector Classification learning algorithm of Scikit Learn. The results obtained showed that the algorithm is able to determine both the type of corn flour and the preparation technique of the arepas. Our results show that the algorithm can determine both the flour type and preparation technique with an accuracy rate of 79 %, for fried arepas made with yellow flour, because the data set color presented similarities with fried arepas made with white flour. Regarding the percentage of white corn flour fried arepas and yellow corn baked arepas, they reflected percentages of 97.05 % and 100.00 % of correct classification of the samples, respectively. These findings suggest that computer vision can be a valuable tool for the food industry in improving the quality and rating arepas and similar food products.