Identificación del estado de consumo del queso mantecoso aplicando redes neuronales convolucionales en imágenes digitales
No Thumbnail Available
Date
2018
Authors
Saavedra Sarovio, Walter Jhoan
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Nacional de Trujillo
Abstract
El objetivo de esta investigación fue desarrollar un sistema aplicando redes neuronales convolucionales (CNN) en imágenes digitales de queso mantecoso para identificar su estado de consumo. Para desarrollar el sistema fue necesario contar con una base de imágenes digitales de quesos en buen y mal estado de consumo (aceptable y rechazado) para luego separarlos en grupos de entrenamiento y prueba; para lo cual se hizo la segmentación individual de la imagen de los quesos, del cual se extrajeron sus características con la red neuronal convolucional (CNN) y finalmente se hizo la clasificación con una máquina de soporte vectorial (SVM). Para ser visualizado en la matriz de confusión obteniendo los siguientes resultados; una precisión del 97.297%, exactitud de 99.234% y una tasa de error del 0.766% para las imágenes tomadas con la cámara fotográfica digital Canon (18 Megapixeles) y una precisión del 93.449 %, exactitud de 96.254% y una tasa de error del 3.746% para las imágenes tomadas con la cámara de celular Sony (13 Megapixeles).
ABSTRACT The objective of this research was to develop a system applying convolutional neural networks (CNN) in digital images of buttery cheese to identify its state of consumption. To develop the system it was necessary to have a base of digital images of cheeses in good and poor state of consumption (acceptable and rejected) and then separate them into training and testing groups; for which the individual segmentation of the image of the cheeses was done, from which their characteristics were extracted with the convolutional neuronal network (CNN) and finally the classification was made with a vector support machine (SVM). To be visualized in the confusion matrix obtaining the following results; a precision of 97.297%, accuracy of 99.234% and an error rate of 0.766% for the images taken with the Canon digital camera (18 Megapixels) and an accuracy of 93.449%, accuracy of 96.254% and an error rate of 3.746 % for images taken with the Sony cell phone camera (13 Megapixels).
ABSTRACT The objective of this research was to develop a system applying convolutional neural networks (CNN) in digital images of buttery cheese to identify its state of consumption. To develop the system it was necessary to have a base of digital images of cheeses in good and poor state of consumption (acceptable and rejected) and then separate them into training and testing groups; for which the individual segmentation of the image of the cheeses was done, from which their characteristics were extracted with the convolutional neuronal network (CNN) and finally the classification was made with a vector support machine (SVM). To be visualized in the confusion matrix obtaining the following results; a precision of 97.297%, accuracy of 99.234% and an error rate of 0.766% for the images taken with the Canon digital camera (18 Megapixels) and an accuracy of 93.449%, accuracy of 96.254% and an error rate of 3.746 % for images taken with the Sony cell phone camera (13 Megapixels).
Description
Keywords
Redes Neuronales, Imágenes Digitales, Segmentación, Maquina de Soporte Vectorial y Matriz de Confusión