Satellite Image Mining for Census Collection: A Comparative Study with Respect to the Ethiopian Hinterland

Census data provides an important source of information with respect to decision makers operating in many different fields. However, census collection is a time consuming and resource intensive task. This is especially the case in rural areas where the communication and transportation infrastructure is not as robust as in urban areas. In this paper the authors propose the use of satellite imagery for census collection. The proposed method is not as accurate as “on ground” census collection, but requires very little resource. The proposed method is founded on the idea of collecting census data using classification techniques applied to relevant satellite imagery. The objective is to build a classifier that can label households according to “family” size. More specifically the idea is to segment satellite images so as to obtain pixel collections describing individual households and represent these collections using some appropriate representation to which a classifier generator can be applied. Two representations are considered, histograms and Local Binary Patterns (LBPs). The paper describes the overall method and compares the operation of the two representation techniques using labelled data obtained from two villages lying some 300km to the northwest of Addis Ababa in Ethiopia.