The Korea AeroSpace Administration said it will newly release about 150,000 entries of the "2025 satellite information big data artificial intelligence (AI) training dataset," built using national satellite information, on Oct. 21.
The dataset was pursued as part of the Korea AeroSpace Administration's project to develop a satellite information big data utilization support system, with the Korea Aerospace Research Institute (KARI), the lead research institution, in charge of building the data. Designed with a focus on ocean monitoring and land cover change detection, it is expected to be used for AI training.
Since 2023, the Korea AeroSpace Administration has systematically built and released more than 630,000 AI training datasets based on satellite information, and this additional release is expected to further expand use by industry and academia. In particular, based on imagery captured by the Korea Multipurpose Satellite-3, -3A and -5, it is expected to help improve the accuracy of satellite image analysis and the development of related new technologies.
A total of 150,000 entries will be released, comprising six ocean monitoring datasets—fish farms, oil spills, coastlines, vessels, landfast ice, and sea ice—and a land cover time-series change dataset.
In addition, the Korea AeroSpace Administration has already built and released large-scale datasets on various topics, including object detection (880,000 images), building partitioning (350,000 images), roads (more than 10,000 km), cloud detection (7,500 images), and land cover classification (1,200 images). All data are easily accessible to anyone through the Korea Institute of Science and Technology Information (KISTI)'s national research data platform DataOn (https://dataon.kisti.re.kr).
Han Chang-heon, director general for space and aerospace industry at the Korea AeroSpace Administration, said, "AI training datasets based on national satellite imagery will increase the utilization of satellite information by domestic researchers and corporations and help secure global competitiveness," adding, "We will continue to build and release new datasets that reflect industry demand."