Diseño e implementación de un prototipo de automatización para la selección de mangos de exportación utilizando una tarjeta Jetson Nano con un sistema difuso.

Abstract

The objective of this work was the implementation of an automation prototype for the selection of export mangoes using a Jetson Nano card with a fuzzy logic system, for the optimization of the time spent by an operator in selection and classification tasks of mangoes for export in farms. For the development of this prototype, a documentary research was carried out on fuzzy logic systems in the agricultural sector and its impact in the export mango, two main variables were considered for their selection, being these imperfections and size of the mangoes, which are used in the training of the YOLO V3 TINY neural network for its detection, thus fulfilling the different processes within the fuzzy logic system, in the first module the collection of input data by means of a camera is done, the second module performs a fuzzification process converting the digitized data to membership functions, the third module lays out the fuzzy rules thanks to the experiences of an expert selector of mangoes for export, the fourth module uses the method of Mamdani Inference by minima, getting the addition of the output membership functions, having a resulting membership function, finally the fifth module allows to obtain a real value by using the Maximum Defuzzification method Means, so that with the real value the selection process is carried out, thus giving the activator the command to reject or accept the parsed mangoes. The prototype was evaluated by countless tests, resulting in a 30% time saving compared to the expert operator in the selection process of mangoes for export. It is recommended to get a large dataset of mangoes for training the neural network.