Content discovery in remote sensing image catalogs

• Grant number: 23/09118-6
• Support Opportunities: Regular Research Grants
• Start date: March 01, 2024
• End date: February 28, 2026
• Field of knowledge: Physical Sciences and Mathematics - Geosciences
• Principal Investigator: Thales Sehn Körting
• Grantee: Thales Sehn Körting
• Host Institution: Instituto Nacional de Pesquisas Espaciais (INPE). Ministério da Ciência, Tecnologia e Inovação (Brasil). São José dos Campos , SP, Brazil
• Associated scholarship(s): 24/22293-4 - SEMI-AUTOMATIC METHODOLOGY FOR DISASTER DETECTION IN IMAGES FROM BRAZILIAN EARTH OBSERVATION SATELLITES, BP.PD; 24/18744-0 - Deep Learning methods for remote sensing image classification, BP.TT; 24/06276-2 - Deep Learning methods for remote sensing image classification, BP.TT

Abstract

According to the seminal definition of the term Big Data, in which the production of new data according to at least 3 factors (increasing data Volume, Velocity in the creation of new data, and Variety/diversity of data, the so called 3 V's) we are able to fit remote sensing images (alsos called satellite images), from the first existing datasets, going back to the Landsat-1 satellite, launched in the 70s, to the PetaBytes generated nowadays, by the Sentinel constellation. To deal with this huge amount of data, automatic processes are required, involving the programming of database algorithms, geoprocessing, image processing, and pattern recognition. The integration of these different areas have been called spatial data science, although it does not present great differences in relation to classical techniques. However, the availability of modern computational structures, related to more powerful and parallel graphics processing, is allowing research in the line of deep neural networks (commonly called Deep Learning). This concept, widely studied nowadays, allows the integration of large databases (in our case, Earth Observation images) for pattern recognition tasks, with accuracies higher than other well known methods(such as decision trees, or support vector machines). This proposal is a sequence from the already completed FAPESP 2017/24086-2 project, which advanced in the generation of algorithms to compute satellite image metadata related to the content present on the images. This continuity lies in the inclusion of Deep Learning methods to refine the target recognition present in the images generated and made available by the remote sensing satellites produced by the National Institute for Space Research (INPE), namely CBERS-4, CBERS-4A and Amazonia-1. Besides the new technique, this proposal aims to study particular targets, such as burned areas, active fires in vegetation, landslides and urban flooding. (AU)

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

QUEVEDO, RENATA PACHECO; MACIEL, DANIEL ANDRADE; REIS, MARIANE SOUZA; RENNO, CAMILO DALELES; DUTRA, LUCIANO VIEIRA; ANDRADES-FILHO, CLODIS DE OLIVEIRA; VELASTEGUI-MONTOYA, ANDRES; ZHANG, TINGYU; KORTING, THALES SEHN; ANDERSON, LIANA OIGHENSTEIN. Land use and land cover changes without invalid transitions: A case study in a landslide-affected area. REMOTE SENSING APPLICATIONS-SOCIETY AND ENVIRONMENT, v. 36, p. 23-pg., 2024-09-13. (23/09118-6)

DE OLIVEIRA, ALISSON CLEITON; SEHN KOERTING, THALES. A multi-temporal dataset for mapping burned areas in the Brazilian Cerrado using time series of remote sensing imagery. BIG EARTH DATA, v. N/A, p. 32-pg., 2025-02-17. (23/09118-6)