Project IDeepS

The project Classificação de imagens via redes neurais profundas e grandes bases de dados para aplicações aeroespaciais (Image classification via Deep neural networks and large databases for aeroSpace applications - IDeepS) aims to carry out a large-scale investigation of several existing deep neural networks (DNNs), in order to primarily automate and improve remote sensing image classification to support the land use and land cover analysis accomplished by the Instituto Nacional de Pesquisas Espaciais (INPE). Other computational vision tasks, such as object detection and semantic segmentation, are also addressed in the project. Results of the IDeepS project may contribute to the information issued by INPE regarding deforestation and fire outbreaks.

In addition, the project intends to identify the best DNNs to support autonomous drone flights, for example, to improve the autonomy of these systems with regard to the response to disasters and emergency situations in areas of difficult access. Thus, this project may guide other actions for the increasing dissemination of low-cost unmanned aerial vehicles (UAVs) in civil and military applications.

The IDeepS project is supported by the Laboratório Nacional de Computação Científica (LNCC/MCTI, Brazil) via resources of the SDumont supercomputer. Researchers, professors and post-graduate students from the following organisations are involved in the project: Instituto Nacional de Pesquisas Espaciais (INPE), Instituto de Estudos Avançados (IEAv), Universidade Federal de São Paulo - Campus São José dos Campos (UNIFESP), Instituto Tecnológico de Aeronáutica (ITA), and Universidade Federal de São Carlos - Campus Sorocaba (UFSCar).

The IDeepS project started in October 2021 and is scheduled to end in October 2023. On August 18, 2022, we had the first workshop of the IDeepS project. The event's website is here. More details about the project and the workshop can be accessed here.

Last Update: August/2023