A machine learning methodology for land use/land cover classification in tropical areas using medium resolution satellite imagery, case: Colombia

ABSTRACT : In this work, we first present a methodology for preparing 10 m to 60 m spatial resolution Sentinel-1, Sentinel-2, and ALOS DSM imagery of forest/grassland areas in Colombia to train a DeepLabV3+ convolutional neural network model. Our preprocessing pipeline for the Sentinel-2 imagery com...

Full description

Autores:
Ceballos Arroyo, Alberto Mario
Tipo de recurso:
Tesis
Fecha de publicación:
2021
Institución:
Universidad de Antioquia
Repositorio:
Repositorio UdeA
Idioma:
eng
OAI Identifier:
oai:bibliotecadigital.udea.edu.co:10495/20659
Acceso en línea:
http://hdl.handle.net/10495/20659
Palabra clave:
Remote sensing
Teledetección
Machine learning
Aprendizaje electrónico
Imágenes por satélites
Satellite imagery
Redes de neuronas
Neural networks
Tratamiento de imágenes
Image processing
Deep Learning
Sentinel-2
Convolutional Neural Network
Satellite Imagery
http://aims.fao.org/aos/agrovoc/c_49834
http://aims.fao.org/aos/agrovoc/c_37359
http://aims.fao.org/aos/agrovoc/c_36761
http://aims.fao.org/aos/agrovoc/c_37467
http://vocabularies.unesco.org/thesaurus/concept1557
Rights
openAccess
License
http://creativecommons.org/publicdomain/zero/1.0/
Description
Summary:ABSTRACT : In this work, we first present a methodology for preparing 10 m to 60 m spatial resolution Sentinel-1, Sentinel-2, and ALOS DSM imagery of forest/grassland areas in Colombia to train a DeepLabV3+ convolutional neural network model. Our preprocessing pipeline for the Sentinel-2 imagery comprises cloud and shadow removal, atmospheric correction, and topographical correction, resulting in mostly cloud-free mosaics of tropical areas. At first, we train the network on very low spatial resolution (500 m) labels of the Colombian Amazonas region resampled to 10 m (+100000 samples after augmentation). Then, we fine-tune the network on medium spatial resolution data (30 m) of northern Antioquia, also resampled to 10 m, resulting in faster convergence and higher accuracy despite the limited number of labelled samples (~5000 samples after augmentation). Our results validate recent proposals where low spatial resolution data is used for training neural networks, and motivate us to keep exploring this line of research.