Original Source
KAIST Develops AI 'Poverty Map,' Detecting Slums with 88% Accuracy from Satellite Images
Development and Features of the AI 'Poverty Map'
A South Korean research team has developed an AI-powered 'Poverty Map' that rapidly and accurately identifies urban slum areas using satellite images. This technology enables differentiation between densely packed general residential areas and slums, which are difficult to distinguish with the naked eye. To achieve this, the research team trained the AI model on diverse characteristics of slum areas from 12 developing countries. For example, African slum areas typically feature a high concentration of homes with rusty corrugated iron roofs, whereas South American slum areas are characterized by irregular formations along hillsides.
AI Model Operation and Performance
When a new satellite image is input, the developed AI technology independently seeks the most suitable expert information for that image through 'customized path selection.' Furthermore, it minimizes errors by classifying only those regions where the judgments of multiple expert models commonly converge as slums. Test results showed high performance, with a slum detection accuracy of 88% even on previously unseen satellite images. Professor Kim Jee-hee of KAIST's School of Technology Management explained that the system utilized a combination of 12 experts to create even more country-specific experts globally, enhanced by AI technology.
Expected Impact and International Recognition
The research team expects this 'AI Poverty Map' to play a crucial role in addressing poverty issues in developing countries with limited data. Recognizing its significance, this research achieved the remarkable feat of winning the Best Paper Award at the International Joint Conference on Artificial Intelligence (IJCAI). This technology is anticipated to provide practical assistance in policy formulation and resource allocation in impoverished regions in the future.
*Source: YouTube: YTN (2026-03-11)*
