Assessment of CNN-Based Methods for Poverty Estimation from Satellite Images

One of the major issues in predicting poverty with satellite images is lack fine-grained and reliable indicators. To address this problem, various methodologies were proposed recently. Most recent approaches use a proxy (e.g., nighttime light), as an additional information, to mitigate problem sparse data. They consist building training CNN large set images, which then used feature extractor. Ultimately, pairs extracted vectors labels are learn regression model predict indicators.First, we propose rigorous comparative study such based on unified framework common images. We observed that geographic displacement spatial coordinates observations degrades prediction performances all methods. Therefore, present new methodology combining grid-cell selection ensembling improves handle coordinate displacement.