Consumer-oriented Pineapple Gold Classification Using Neural Networks and Partial Least-Squares Discriminant Analysis

Abstract

The criterion used by the final consumer for the selection of edible pineapple, which bases on the color of the peel, often leads to errors because the color does not typically correspond to the fruit’s internal maturity index (MI). Therefore, the objective of this study was to predict the internal maturation index of the pineapple by using as classifying feature the texture extracted from the RGB color spaces and classifying this fruit into two categories (IM edible and IM inedible)by means of digital images and the convolutional neural networks. 10 structures of convolutional neural networks including the VGG16 and Inception V3 and partial least squares discriminant analysis to foresee the maturity were trained and validated. 168 Golden variety pineapples were purchased at random from supermarkets and marketplaces in Bogotá, Colombia. The maturity index was experimentally quantified and, with the obtained value the pineapples were placed into a two categories range : edible or inedible and, as a result only the Only 33% of the pineapples fell within the edible range (MI: 19-34). The network that generated the best prediction was the VGG16 network with an accuracy of 0.90, while partial least squares discriminant analysis showed an accuracy of 0.88