Urban poverty patterns in Pyongyang (North Korea): A deep-learning-based spatial analysis

This study attempts to identify the location and characteristics of poor urban neighborhoods in Pyongyang using deep-learning-based remote sensing (RS), geographic information systems (GIS), and big data analyses. To this end, we first classify the residential areas of Pyongyang City into five classes. We then analyze each class's spatial and inter-variable characteristics to identify the location and characteristics of poor residential areas. Due to the scarcity of statistical data and the restricted access to the city, this study introduces a research methodology termed Machine-Human Sequential Design (MHSD). The results of the study show that low levels of Housing environment, Economic status, Accessibility, and Block activation characterized poor neighborhoods in Pyongyang. In comparing individual variables, the density of single-story slate-roofed houses (tangjip) was the highest, and the density of night illumination, number of cars, and points of interest (POIs) was the lowest. This study illustrates the distinctions in urban poverty in Pyongyang compared to developing countries and introduces a novel research methodology tailored for a highly closed city.