"Estimating Economic Indicators of North Korea and Others Using Artificial Intelligence"

A technology has been developed that can estimate economic indicators of regions such as North Korea, where it has been difficult to accurately grasp poverty conditions, by interpreting satellite images using artificial intelligence. This technology is significant in that it can gauge the economic situation of specific areas based on detailed images even without basic statistics.

KAIST announced on the 21st that a research team led by Professors Cha Mi-young and Kim Ji-hee, in an international joint study with the Institute for Basic Science, Sogang University, Hong Kong University of Science and Technology (HKUST), and National University of Singapore (NUS), developed an AI technique that analyzes economic conditions using daytime satellite images.

(From the top left) Andonghyun, Ph.D. candidate, Department of Computer Science, KAIST (1); Yang Jaeseok, Ph.D. candidate, Department of Geography, National University of Singapore (2); Cha Miyoung, Professor, KAIST / IBS CI (3); Kim Jihee, Professor, KAIST (4); Park Sangyun, Professor, Hong Kong University of Science and Technology (5); Yang Hyunju, Professor, Sogang University (6). Provided by KAIST

During the development process, the research team focused on creating a universal model capable of monitoring even the poorest countries, where basic statistical data is scarce, rather than the usual environment of learning based on existing statistical data.

They utilized Sentinel-2 satellite images operated and freely provided by the European Space Agency (ESA), finely dividing the satellite images into small areas of 6 km² (2.5×2.5 km²), and quantified visual information such as buildings, roads, and greenery using AI techniques to derive economic indicators for each area.

The distinguishing feature of the research team's model compared to previous studies is the implementation of a 'human-machine collaborative algorithm' that incorporates human-provided information into AI predictions to estimate economic indicators in regions lacking basic data.

Humans interpret satellite images to judge the scale of economic activity (assessing whether it is large or small), and the machine learns from the information provided by humans to assign economic scores to each image.

As a result of verifying this approach, the research team confirmed that collaboration between humans and AI produces significantly better outcomes than estimating economic indicators through machine learning alone.

Building on this, the team expects to expand the scope of economic analysis to regions where existing statistical data is insufficient. They explained that it was possible to apply this technology to North Korea and five Asian countries (Nepal, Laos, Myanmar, Bangladesh, Cambodia) to derive and disclose detailed economic indicator scores.

Compared to South Korea, where lights are bright in nighttime light images primarily used for economic scale estimation (background photo at the top left provided by NASA Earth Observatory), most areas in North Korea, except Pyongyang, appear as dark spaces without lights. In contrast, the model developed by the research team allows for more detailed economic predictions for North Korea (top right) and five Asian countries (background photo at the bottom provided by Google Earth). Provided by KAIST

In particular, the economic indicators presented in this study showed a high correlation with existing socioeconomic indicators such as population density, employment numbers, and number of businesses, and the team reaffirmed that it is applicable to underdeveloped countries with limited data.

The research team also emphasized that “the greatest advantage of the proposed model is its ability to detect annual changes in economic activity.” This strength can contribute to quickly monitoring trends in the Sustainable Development Goals (SDGs) targeted by the international community, such as 'ending poverty' and 'reducing inequality.'

Furthermore, the methodology proposed by the research team is expected to be utilized not only for economic conditions but also for measuring various social and environmental indicators.

For example, if the technology (model) developed by the team is trained to identify regions highly affected by climate change and disaster damage, it can map out areas requiring humanitarian aid after disasters, enabling prioritized support.

Professor Cha Mi-young of KAIST’s Department of Computer Science and IBS Data Science Group CI, who participated in the research, stated, “The important significance of this study lies in its ability to address global poverty issues by integrating computer science, economics, and geography,” adding, “The research team plans to apply the developed AI algorithm to international social issues such as carbon dioxide emissions, disaster damage detection, and overall impacts of climate change.”

Meanwhile, the development of SDGs indicators using satellite images and AI and their policy applications are among the technology fields attracting international attention, and this research area is increasingly important as Korea can take the lead and drive it forward.

Accordingly, the research team plans to publicly release the developed model code for free to the public, enabling the measured indicators to be used in policy design and evaluation in various countries.

This content was produced with the assistance of AI translation services.