Classification of tomato cultivars for processing with artificial vision and Euclidian distance

Abstract

Industrialized tomato is the vegetable most consumed worldwide and due to its content of bioactive compounds and antioxidants, such as lycopene and carotene, is considered functional food. Color is one of the most important appearance parameters, which defines quality. With the recent advances in computer power and memory of personal computers, the artificial vision system can be applied in the selection or online classification of agricultural products. Thus, the present work proposes a methodology for the classification of different tomato cultivars based on the color model obtained from instrument (colorimeter) and digital image (RGB) of physicochemical characteristics (total soluble solids, pH and total titratable acidity), and pigment content. To this end, two pattern recognition techniques were used and compared: MLP (Multilayer Perceptron) and KNN (K-Nearest Neighbor) neural networks. In the case study, 330 tomato samples were used, 30 fruits of each cultivar. Analyzing the physicochemical characteristics, pigments and instrumental color analysis and digital image, cultivars formed three distinct groups, being H9992 cultivar isolated from the others, cluster II with HY37 and BRSena cultivars and cluster III with the other cultivars, most were grouped due to the presence of similarities. The cross validation results obtained quite high accuracy (%), since cultivars were analyzed in their full maturation stage, when their characteristics are very similar. Statistical models showed remarkable performance in the classification of cultivars. Of the two proposed models, KNN obtained 99.69% accuracy, being the best mathematical model proposed in this study.