Sistema Clasificador de Ciruela Horvin Usando Técnicas de Inteligencia Artificial

Abstract

The development of this research, a prototype of a conveyor belt for the selection of plums is implemented, in which a sensing section composed of a camera contained in an environment of controlled luminosity is installed. The images acquired by the sensing system were submitted, on the one hand to Computer Vision and Deep Learning algorithms, oriented to the extraction of characteristics and, on the other hand to Machine Learning and Deep Learning algorithms, aimed at classifying the fruit into three categories, defined by their morphological characteristics and nutritional deficiencies (visual impairments). The first tests to reach a satisfactory classification were carried out by applying multiple Computer Vision techniques to the images, such as: Canny edge detection, morphological operations, background segmentation by thresholds, among other techniques that allow engineering characteristics, the which gave way to a conditional classification structure. Subsequently, tests were made with decision trees, vector support machines, multilayer perceptron and K-Nearest Neighbor (KNN). Finally, a convolutional neural network (CNN) structure was implemented.