Estimating poverty using aerial images: South African application

Abstract

Policy makers and the government rely heavily on survey data when making policy-related decisions. Survey data is labour intensive, costly and time consuming, hence it cannot be frequently or extensively collected. The main aim of this research is to demonstrate how deep learning in computer vision coupled with statistical regression modelling can be used to estimate poverty on aerial images supplemented with national household survey data. This is executed in two phases; aerial classification and detection phase and poverty modelling phase. The aerial classification and detection phase use convolutional neural networks (CNN) to perform settlement typology classification of the aerial images into three broad geo-type classes namely; urban,  rural  and farm. This is then followed by object detection to detect three broad dwelling type classes in the aerial images namely; brick house, traditional house, and informal settlement. Mask Region-based CNN (Mask R-CNN) model with a resnet101 backbone model is used to perform this task. The second phase, poverty modelling phase, involves using National Income Dynamics Survey (NIDS) data to compute the poverty measure Sen-Shorrocks-Thon index (SST). This is followed by using ridge regression to model the poverty measure using aggregated results from the aerial classification and detection phase. The study area for this research is eThekwini district in Kwa-Zulu Natal, South Africa. However, this approach can be extended to other districts in South Africa.