Publication: Estudio, diseño e implementación de un ecosistema IoT
que integra visión artificial multiespectral y su aplicación al sector porcino.
Loading...
Official URL
Full text at PDC
Publication Date
2020
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
En este trabajo se ha evaluado y desarrollado la idea de utilizar un sistema de monitorización no invasiva que permita conocer el estado de salud de un animal a partir de una imagen multiespectral (RGB y térmica) utilizando para ello redes neuronales que permitan generar una máscara de segmentación para cada cerdo de manera que se pueda conocer la temperatura individual de cada uno, todo ello integrado en un ecosistema IoT que permita al ganadero tener monitorizado el estado de salud de sus animales desde su casa sin necesidad de que se persone en la granja con tanta frecuencia, ganando por un lado en productividad ya que controlar la temperatura permite detectar enfermedades y tratarlas con mayor celeridad, reduciendo el uso de antibióticos y aumentando la tasa de supervivencia y, por otro, en calidad de vida, ya que puede hacer todo esto desde su casa y de manera autónoma.
Es por ello que se ha realizado una primera prueba de concepto donde se ha llevado a cabo el estudio de la aplicación y un primer desarrollo que sirva como base para fases futuras. De esta forma, en este trabajo se ha realizado el ajuste fino de un modelo de aprendizaje profundo de segmentación de objetos y tras ello se ha construido un ecosistema IoT completo, partiendo de la captura de los datos y su procesamiento utilizando dicho modelo, el envío a una plataforma en la nube, su almacenaje en una base de datos y finalmente su visualización en una plataforma.
In this work, it has been evaluated and developed the idea of using a non-invasive monitoring system that allows to know the health status of an animal from a multispectral image (RGB and thermal) using neural networks that allow to generate a segmentation mask for each pig in order to know the individual temperature of each one, all this integrated in an IoT ecosystem that allows the farmer to monitor the health status of their animals from home without having to go to the farm so often, gaining on the one hand in productivity as controlling the temperature allows to detect diseases and treat them more quickly, reducing the use of antibiotics and increasing the survival rate and, the other, in quality of life, as it can do all this from home and independently. That is why a first proof of concept has been carried out where the application has been studied and a first development that will serve as a basis for future phases. Thus, in this work a deep learning model of object segmentation has been fine-tuned and after that a complete IoT ecosystem has been built, starting from the data capture and processing using this model, the sending to a cloud platform, its storage in a database and finally its visualization in a platform.
In this work, it has been evaluated and developed the idea of using a non-invasive monitoring system that allows to know the health status of an animal from a multispectral image (RGB and thermal) using neural networks that allow to generate a segmentation mask for each pig in order to know the individual temperature of each one, all this integrated in an IoT ecosystem that allows the farmer to monitor the health status of their animals from home without having to go to the farm so often, gaining on the one hand in productivity as controlling the temperature allows to detect diseases and treat them more quickly, reducing the use of antibiotics and increasing the survival rate and, the other, in quality of life, as it can do all this from home and independently. That is why a first proof of concept has been carried out where the application has been studied and a first development that will serve as a basis for future phases. Thus, in this work a deep learning model of object segmentation has been fine-tuned and after that a complete IoT ecosystem has been built, starting from the data capture and processing using this model, the sending to a cloud platform, its storage in a database and finally its visualization in a platform.
Description
Máster en Internet de las Cosas, Facultad de Informática UCM, Departamento de Ingeniería del Software e Inteligencia Artificial, Curso 2019/2020.