A domestic research team has developed artificial intelligence (AI) that can locate slum areas on its own using only satellite images.
KAIST said on the 6th that a joint team led by Computer Science Department Professor Cha Mi-young and Management Engineering Department Professor Kim Ji-hee, together with Chonnam National University Geography Department Professor Yang Jae-seok, developed a satellite image-based general-purpose slum detection AI technology.
There have been studies using satellite images to detect slums, but building shapes and density vary greatly from city to city, causing accuracy to drop sharply in new areas. In most developing countries, there is a lack of data that marks slum locations one by one, making AI training itself difficult.
To solve this, the team introduced a mixture of experts (MoE) structure in which multiple AI models learn different regional characteristics and, when a new city is input, automatically select the most suitable model. They also used test-time adaptation (TTA), a technique in which the AI compares and verifies the predictions of multiple models and trusts only the commonly matching areas to reduce errors on its own.
When the technology was applied to major cities such as Kampala and Maputo in Africa, it distinguished slum areas more precisely than the latest existing techniques. It is expected to be used in a range of policy areas, including planning for urban infrastructure expansion in developing countries, pre-identifying areas vulnerable to disasters and infectious diseases, selecting targets for housing environment improvement projects, and monitoring implementation of the United Nations Sustainable Development Goals (SDGs).
Professor Cha Mi-young said, "This study shows that AI can go beyond a simple analysis tool to contribute to solving real social problems even in regions with limited data." Professor Kim Ji-hee noted, "It will help complement on-site surveys that require enormous expense and effectively allocate limited resources to the areas that need them most."
The study won the best paper award in the social impact AI institutional sector at the AAAI 2026, one of the world's most prestigious AI conferences.
References
AAAI 2026, LINK: https://aaai.org/about-aaai/aaai-awards/aaai-conference-paper-awards-and-recognition/