TT DATA has partnered with U.S.-based startup Bifrost AI to revolutionize AI model training for satellite imageryusing synthetic data. The collaboration aims to overcome the limitations of costly and time-intensive real-world data collection.
By leveraging Bifrost AI’s synthetic data generation platform, NTT DATA was able to train AI detection models up to hundreds of times faster and about 70% cheaper than traditional methods, achieving a 10-point accuracy improvement in weeks instead of months.
Synthetic satellite imagery—computer-generated but highly realistic—enables AI models to interpret diverse and rare environmental scenarios that are difficult to capture through real images. This approach reduces dependence on large volumes of annotated real-world data.
The proof-of-concept showed strong performance in object and change detection tasks, improving mean average precision and F1 scores while requiring minimal manual labeling.
NTT DATA envisions this technology supporting the rise of “physical AI,” where machines analyze the world through sensors and cameras, similar to human visual learning.
Potential applications include remote infrastructure monitoring, agriculture optimization, and disaster management, all without frequent physical inspections.
This partnership is part of NTT DATA’s broader strategy to enhance AI-driven satellite systems, improving speed, cost-efficiency, and accuracy across Earth observation and industrial applications.
