Aplicación de deep learning para la clasificación y reconocimiento de enfermedades en las hojas de las plantas.
No hay miniatura disponible
Fecha
2023-09-19
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad de Guayaquil. Facultad de Ingeniería Industrial. Carrera de Ingeniería en Teleinformática.
Resumen
El presente trabajo surge de la necesidad encontrada en el sector agronómico y el actual
reconocimiento de la inteligencia artificial para solución de tareas. Se creó un modelo de red
neuronal convolucional que clasifica las enfermedades de las plantas de tomate, maíz y soya
mediante el preprocesamiento de un conjunto de datos (imágenes). Para la creación del
modelo se realizaron pruebas con varios modelos entre ellos, modelos densos con y sin
aumento de datos. Se probó aplicando Dropout para evitar el sobreajuste, quedando como el
modelo que dio mejores resultados el CNN con aumento de datos y Dropout.
Adicionalmente se creó un mini invernadero con las muestras de las plantas seleccionadas
para tener predicciones en tiempo real y visualizarlas en una interfaz visual que muestra el
índice de probabilidad y la clase a la que pertenece la imagen mostrada. Las métricas
utilizadas para la evaluación del modelo fueron: accuracy, F1_Score, precisión y la matriz
de confusión.
This research arises from the need found in the agricultural sector and the current recognition of artificial intelligence for solving tasks. A convolutional neural network model that classifies tomato, corn and soybean plant diseases was created by preprocessing a data set (images). For the creation of the model, tests were carried out with several models, among them, dense models with and without data increase. It was tested by applying ‘Dropout’ to avoid overfitting, leaving the CNN with increased data and Dropout as the model that gave the best results. Additionally, a mini greenhouse was created with the samples of the selected plants to have predictions in real time and visualize them in a graphical interface that shows the probability index and the class to which the displayed image belongs. The metrics used for the evaluation of the model were: accuracy, F1_Score, precision and the confusion matrix.
This research arises from the need found in the agricultural sector and the current recognition of artificial intelligence for solving tasks. A convolutional neural network model that classifies tomato, corn and soybean plant diseases was created by preprocessing a data set (images). For the creation of the model, tests were carried out with several models, among them, dense models with and without data increase. It was tested by applying ‘Dropout’ to avoid overfitting, leaving the CNN with increased data and Dropout as the model that gave the best results. Additionally, a mini greenhouse was created with the samples of the selected plants to have predictions in real time and visualize them in a graphical interface that shows the probability index and the class to which the displayed image belongs. The metrics used for the evaluation of the model were: accuracy, F1_Score, precision and the confusion matrix.
Descripción
Este documento contiene archivo en PDF.
Palabras clave
INTELIGENCIA ARTIFICIAL Y CIENCIA DE DATOS, DEEP LEARNING, ENFERMEDADES EN PLANTAS, RED NEURONAL CONVOLUCIONAL