Poverty incidence identification of cities and municipalities using convolutional Neural Network as applied to satellite imagery

Abstract

Poverty remains to be a hindrance to national growth for many developing countries. This presents a problem in a country's resource management and urban planning of its people. Furthermore, there is a big gap when it comes to data collection and monitoring in developing countries such that the government fails to prioritize this area of responsibility. The authors aim to bridge this gap by classifying wealth through the use of satellite images and emerging technology, particularly Convolutional Neural Networks. The researchers' goal is to test whether AlexNet, a Convolutional Neural Network Architecture, can identify the poverty levels of municipalities in the Philippines, by estimating their poverty incidences, based on satellite images. With this research, it can measure a broader set of society growth indicators. Indicators such as the material of people's houses, the size of their houses, their living conditions, access to roads and also how highly urbanized an area is, are considered important factors to indicate poverty. The results shows an accuracy of 0.84 and average precision-recall of 0.86 and 0.84, respectively. This study will give deeper insight to the government as to which areas to improve on in certain sectors given that a poverty incidence is high in a certain municipality.