COMPUTER VISION CLASSIFICATION OF POTATO CHIPS BY COLOR

Abstract

ABSTRACTIn this research the automatic classification of commercial potato chips by computer vision was studied. The general objective was to design a tool that would be able to classify objectively potato chips according to their color in different categories. For this purpose, sensory measurements of color in 100 potato chips were correlated with the corresponding objective measurements obtained by computer vision system. Potato chips with and without ruffles of different brands were used for training and validation experiments. Sensory evaluations were done with a special chart that classifies potato chips in seven color categories. Simultaneously, the color of the same potato chips classified by the sensory panel, was determined objectively by a computer vision system in L*, a*, b* units. A linear regression model was good enough to predict potato chip sensory color values from the corresponding instrumental measurements by computer vision. The linear model after following the process of crossed validation crossed presented an error of ∼4% for smooth chips (without ruffles) and ∼7% for chips with ruffles.PRACTICAL APPLICATIONSThe automatic classification methodology presented for potato chips is general and has a wide range of potential uses. It could be applied not only to other potato cultivars and frying conditions but also to other less heterogeneous raw materials and unit operations different than potato and frying, respectively. The computer vision system used in this research could as well be very useful in the food industry as a large amount of information can now be obtained from measurements at the pixel level that allows a better characterization of foods and thus improves quality control.