Data on key measures of socioeconomic development can be a powerful tool to assist government policies decisions. However, reliable data on such measures at the local level is still lacking in many parts of the world, especially in developing countries. This mainly occurs due to the difficulties of scaling up traditional data collection efforts, such as government surveys, which can be expensive, time-consuming, and overlook hard-to-reach areas. In the Brazilian context, the main census

more »

... e is known as IBGE (Brazilian Institute of Geography and Statistics) and the data collection is done usually at 10-year intervals. This periodicity produces a data gap that prejudices the interpretability of the collection results, since they may not reflect the Brazilian reality in real-time. The goal of this project is to develop an inexpensive and scalable deep learning method for estimating socioeconomic indicators to fill this data gap using satellite imagery. This is very promising due to the low consumption of resources and the great availability of satellite images, allowing to cover larger and more remote areas and making the process automated and more efficient, without the necessity of on-site collection. To this end, the project aims to replicate the work done by Yeh et al. (2020) [1] in the Brazilian scenario, more specifically, the region Vale do Ribeira will be used as a case study. Yeh et al. uses deep neural networks to predict socioeconomic criteria in African countries, based on satellite images of their territories.