KASA to Develop Customized Satellite Image AI Training Datasets for Companies

Reflecting Industry Needs for the First Time

New Datasets to Be Released This Year

The Korea Aerospace Administration (KASA) will begin producing customized satellite image AI training datasets that reflect the needs of domestic satellite imaging and artificial intelligence (AI) companies. The strategy aims to strengthen the competitiveness of the private AI industry by expanding the availability of satellite data.

KASA announced that it will hold an 'Industry Roundtable on the Use of Satellite Image Big Data AI Training Datasets' for domestic satellite imaging and AI-related companies at the Daejeon Convention Center (DCC) on May 15.

Example of 3-dimensional (3D) geospatial analysis using satellite imagery AI training data. This technology enables artificial intelligence (AI) to automatically detect objects such as buildings and roads and precisely extract terrain information, winning the CES 2025 Innovation Award. Provided by Spade and Korea Aerospace Administration (KASA)

The roundtable will discuss the types of satellite image AI training datasets actually needed by the industry, challenges encountered during their use, and areas where policy support is required. KASA plans to incorporate the feedback gathered at the event into future production and provision methods for satellite image AI training datasets.

KASA and the Korea Aerospace Research Institute (KARI) have continued to release relevant training datasets for the development of AI technologies based on satellite imagery. From 2023 to 2024, they released 630,000 datasets in areas such as object detection, building segmentation, road extraction, cloud detection, and land cover classification.

By early this year, an additional 300,000 datasets were released, expanding the range of applications to include marine and glacier monitoring, as well as time-series land cover change detection. These datasets are provided via the national research data platform, DataON.

The released data is being utilized by both industry and academia. SPAIDE has used satellite image AI training datasets to improve the object detection accuracy and terrain information precision of its 3D geospatial AI solution, while a research team at Seoul National University of Science and Technology is applying them to the development and education of ship and aircraft detection models.

Roh Kyungwon, Deputy Administrator of KASA, said, "We will actively promote the opening of satellite data from multipurpose practical satellites and next-generation medium satellites, and enhance the competitiveness of the domestic satellite information industry by reflecting on-site industry feedback in our policies."

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