Uso de visão computacional para reconhecimento de imagens de frutas em imagens RGB

Abstract

Among the proposals to provide nutritional information to promote healthy eating habits, the use of Digital Image Processing and Computer Vision techniques to identify foods in digital images and provide accurate information based on food composition tables stands out. of reference. In this work, we present a model for detecting and classifying bananas, apples and oranges in Red Green Blue (RGB) images using Convolutional Neural Networks. The proposed model was executed through two experiments. The experiments were carried out with the dataset of (KUANG, 2015), where the images of the fruits have varied backgrounds. The first experiment had an accuracy of 94.33%. The second experiment, using a database expansion through a random transformation function, obtained an accuracy of 98.7%. It is expected that in future works, use the results obtained by the model to create an application, where users can obtain nutritional information, thus promoting a food education channel for the population.