Prototipo de sistema de red neuronal Backpropagation cliente/servidor TCP/IP aplicado a fotomonitoreo
Cargando...
Archivos
Fecha
2018
Profesor/a Guía
Facultad/escuela
Idioma
es
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Andrés Bello
Nombre de Curso
Licencia CC
Licencia CC
Resumen
Los algoritmos de estimación de datos usando machine learning cada vez son más
precisos, sobre todo con el incremento de la capacidad de computo de las nuevas
tecnologías, sin embargo, es común que se trabaje en una sola máquina, en vez de
aprovechar el potencial de una red de computadores.
En la actualidad los sistemas de control agrícola dependen mucho de la tecnología, aquí
es donde nace la agricultura de precisión que de la mano de las tecnologías se encarga de
cultivar exhaustivamente los cultivos abarcando todos los factores determinantes para una
cosecha exitosa. (FAO, 2015)
Este proyecto se inspira en crear un prototipo de un hardware y software capacitado para
monitorear un cultivo interior en el cual se pueda confiar cualquier tipo de plantas
alimenticias, medicinales u ornamentales. Con el fin de que en todo su proceso vegetativo
y de floración se encuentre en las condiciones óptimas para su adecuado crecimiento. Con
el objetivo de integrar tecnologías de machine learning al fitomonitoreo, el sistema de
cultivo inteligente tiene una red neuronal, la que se encarga de analizar los datos del
cultivo en su estado actual y predecir posibles estados futuros a partir de las entradas
monitoreadas en la cual podría estar el cultivo.
El proyecto se realiza empezando por su parte teórica, la cual describe, comenta y detalla
todas las bases teóricas del proyecto; donde se pone énfasis en el comportamiento de la
planta en su proceso de respiración el cual permite entregar un feedback sobre el estado
de la planta sin intervenirla, así como la estructura de las redes backpropagation basadas
en perceptrones de Mc Cullcoh-Pitts
Para la implementación del prototipo se utiliza una tarjeta Raspberry pi 3, la cual usando
el lenguaje Python 2.7 con la librería PyBrain donde se diseñó una red neuronal multicapa
con 3 nodos en capa de entrada, 3 nodos en capa oculta y un nodo en capa de salida. Se
utilizan sensores analógicos que permiten hacer lecturas de la humedad, temperatura,
humedad del suelo y el co2 del entorno, para verificar el estado del cultivo.
Se validó el proyecto, primero utilizando tablas prefabricadas con reglas lógicas para las
funciones AND,OR y XOR, para posteriormente probar con sensores de fitomonitoreo,
graficando tanto el puntaje de acierto entre variable estimada y observada además del
error.
The algorithms of data estimation using machine learning are becoming more precise, especially with the increase in computing capacity of new technologies, however, it is common to work on a single machine, instead of taking advantage of the potential of a network of computers. Nowadays, agricultural control systems depend a lot on technology, this is where precision agriculture is born, which at the hands of technologies is responsible for cultivating crops thoroughly, covering all the determining factors for a successful harvest. (FAO, 2015) This project is inspired to create a prototype of hardware and software capable of monitoring an indoor crop in which any kind of food, medicinal or ornamental plants can be trusted. In order that in all its vegetative and flowering process is in the most optimal conditions for proper growth. With the aim of integrating machine learning technologies into phytomomonitoring, the intelligent farming system has a neural network, which is responsible for analyzing crop data in its current state and predicting possible future states from the monitored inputs in which could be the crop. The project is carried out starting from its theoretical part, which describes, comments and details all the theoretical bases of the project; where emphasis is placed on the behavior of the plant in its respiration process which allows feedback on the state of the plant without intervention, as well as the structure of the backpropagation networks based on perceptrons of Mc Cullcoh-Pitts For the prototype implementation a Raspberry Pi 3 card is used, which using the Python 2.7 language with the PyBrain library where a multilayer neural network was designed with 3 nodes in the input layer, 3 nodes in the hidden layer and one node in the layer departure. Analog sensors are used to make readings of the humidity, temperature, humidity of the soil and the co2 of the environment, to verify the state of the crop. The project was validated, first using prefabricated tables with logical rules for the AND, OR and XOR functions, to later test with phytomomonitoring sensors, plotting both the hit score between the estimated and observed variable as well as the error.
The algorithms of data estimation using machine learning are becoming more precise, especially with the increase in computing capacity of new technologies, however, it is common to work on a single machine, instead of taking advantage of the potential of a network of computers. Nowadays, agricultural control systems depend a lot on technology, this is where precision agriculture is born, which at the hands of technologies is responsible for cultivating crops thoroughly, covering all the determining factors for a successful harvest. (FAO, 2015) This project is inspired to create a prototype of hardware and software capable of monitoring an indoor crop in which any kind of food, medicinal or ornamental plants can be trusted. In order that in all its vegetative and flowering process is in the most optimal conditions for proper growth. With the aim of integrating machine learning technologies into phytomomonitoring, the intelligent farming system has a neural network, which is responsible for analyzing crop data in its current state and predicting possible future states from the monitored inputs in which could be the crop. The project is carried out starting from its theoretical part, which describes, comments and details all the theoretical bases of the project; where emphasis is placed on the behavior of the plant in its respiration process which allows feedback on the state of the plant without intervention, as well as the structure of the backpropagation networks based on perceptrons of Mc Cullcoh-Pitts For the prototype implementation a Raspberry Pi 3 card is used, which using the Python 2.7 language with the PyBrain library where a multilayer neural network was designed with 3 nodes in the input layer, 3 nodes in the hidden layer and one node in the layer departure. Analog sensors are used to make readings of the humidity, temperature, humidity of the soil and the co2 of the environment, to verify the state of the crop. The project was validated, first using prefabricated tables with logical rules for the AND, OR and XOR functions, to later test with phytomomonitoring sensors, plotting both the hit score between the estimated and observed variable as well as the error.
Notas
Tesis (Ingeniero en Automatización y Robótica)
Palabras clave
Redes Neurales (Ciencia de la Computación), Fitomonitoreo, Agricultura, Automatización