Detection of Unmonitored Graveyards in VHR Satellite Data Using Fully Convolutional Networks

Abstract

Lima, Peru, is a highly dynamic urban region home to perpetually evolving informal areas. Earth Observation (EO) studies on these areas focused almost solely on their inhabited parts, the informal housing. In this study, we propose to extend the focus to another component of the informal settlements: informal graveyards. Their emerging morphologies in Lima are simi-lar to informal housing making this particular distinction challenging. Furthermore, both graveyards and housing typically experience joint, intertwined spatial development. The adja-cency of graveyards and informal housing causes social and public health risks. Therefore, detection of boundaries between graveyards and adjacent (in)formal housing is essential, e.g., as an information basis for preventing the spread of diseases and supporting public health and safety in general. However, housing invasions on burial grounds have not yet been systematically investigated. Therefore, this study aims to develop a method for the distinction of informal graveyards from (in)formal housing. We combined anthropological field observa-tions with state-of-the-art Fully Convolutional Networks (FCNs) with dilated convolution of increasing spatial kernels to acquire features of deep level of abstraction on Pleiades optical satellite images. The trained neural network developed reaches good accuracies in mapping informal graveyards and (in)formal housing with a F1-score of 0.878.