Machine learning based energy-efficient uav trajectory design for site-specific crop spraying

The use of unmanned aerial vehicles (UAVs) in agriculture is extending at a brisk rate due to their huge potential to address problems related to costs, time, productivity and climate change. Motivated by the necessity of leveraging technology to optimize agricultural practices, we propose a methodo...

Full description

Autores:
Acevedo Ramos, Jorge Alfredo
Tipo de recurso:
Trabajo de grado de pregrado
Fecha de publicación:
2020
Institución:
Universidad de los Andes
Repositorio:
Séneca: repositorio Uniandes
Idioma:
eng
OAI Identifier:
oai:repositorio.uniandes.edu.co:1992/51533
Acceso en línea:
http://hdl.handle.net/1992/51533
Palabra clave:
Riego
Drones
Aprendizaje automático (Inteligencia artificial)
Ingeniería
Rights
openAccess
License
https://repositorio.uniandes.edu.co/static/pdf/aceptacion_uso_es.pdf
Description
Summary:The use of unmanned aerial vehicles (UAVs) in agriculture is extending at a brisk rate due to their huge potential to address problems related to costs, time, productivity and climate change. Motivated by the necessity of leveraging technology to optimize agricultural practices, we propose a methodology to determine an efficient trajectory to be followed by an UAV in order to apply pesticides, fertilizers or water to a crop. The main goal is to satisfy the specific resource requirements of the different areas of the crop, while minimizing energy consumption. A sample data generation method considering drone dynamics is presented and used to implement machine learning algorithms and build a black-box model to predict the energy that a UAV would consume by following a complete trajectory. Given this, a reformulation of the Travelling Salesman Problem (TSP) is solved to obtain the desired energy efficient trajectory for crop spraying. The effectiveness of the methodology has been tested for crop irrigation given a CWSI image of a vineyard. Simulations results shown that a trajectory to properly irrigate the crop while significantly reducing water, time and energy consumption can be obtained. Moreover, a neural network was trained to accurately predict energy consumption in just 0.06% of the run time that would require its estimation with a white-box model, thereby reducing the time needed to solve the TSP problem from a week to a few minutes. According to these results, the proposed methodology could be implemented to quickly obtain trajectories for any given crop, contributing to the reduction of resources spent during the spraying process.