Deforestation monitoring based on machine learning techniques

Machine Learning (ML) represents one of the most dynamical fields in contemporary technology. In conjunction with remote sensing for Earth observation data, it enables groundbreaking advancements in areas like environmental monitoring and disaster response, revolutionizing the analysis and use of th...

Descripción completa

Detalles Bibliográficos
Autor: Manresa Román, María-Isabel
Tipo de recurso: tesis de maestría
Fecha de publicación:2024
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/421466
Acceso en línea:https://hdl.handle.net/2117/421466
Access Level:acceso abierto
Palabra clave:Machine learning
Artificial intelligence
Remote sensing
Machine Learning
Earth observation
Artificial Intelligence
Aprenentatge automàtic
Intel·ligència artificial
Teledetecció
Àrees temàtiques de la UPC::Enginyeria de la telecomunicació::Radiocomunicació i exploració electromagnètica::Teledetecció
Descripción
Sumario:Machine Learning (ML) represents one of the most dynamical fields in contemporary technology. In conjunction with remote sensing for Earth observation data, it enables groundbreaking advancements in areas like environmental monitoring and disaster response, revolutionizing the analysis and use of this data. This project has been carried out at UPC and Tracks CO2, a startup company specialized in Artificial Intelligence (AI) and using remote sensing for Earth observation data to monitor forests. In this case, the project proposal is to build a change detection system that focuses on classifying types of changes in forests during different instances of time with the final objective of monitoring deforestation. For this project, we have built the databases, and trained random forest, XGBoost and Multilayer Perceptron \ac{MLP} models to reach a 0,486 accuracy in validation and test when using specific classes. The accuracy goes up to 0,705 when merging classes to detect Forest growth or forest shrinkage.