The world's first large-scale remote sensing basic model with more than 10 billion parameters, "RingMo 3.0," was launched at the first Aerospace Information Technology Conference on Saturday, marking it as an integrated space-air remote sensing model.
The model can accurately extract information from remote sensing images and swiftly convert it into usable products, greatly improving the efficiency of remote sensing data utilization.
RingMo 3.0's core framework uses heat conduction to model visual semantic propagation as heat diffusion. This design overcomes bottlenecks in remote sensing model inference, delivering more than 2.4 times the computing speed of traditional models.
Sun Xian, a researcher at the Aerospace Information Research Institute of the Chinese Academy of Sciences, compared basic models to robots, saying that parameters function as memory and knowledge cells.
"The more parameters a model has, the better it can capture, store and process complex information, allowing it to handle more difficult tasks with greater accuracy," he said.
RingMo 3.0 has been tested on 23 international benchmark datasets across six key tasks, including scene classification, object detection and change detection, where it delivered top-tier performance.
Currently in trial deployment across industries such as emergency response, land management and marine operations, it excels in complex tasks like fine classification and tracking small moving targets, providing an effective solution for integrated space-air-ground applications.
"RingMo 3.0 represents a deep AI application in the field. Unlike many foreign models that are focused on entertainment and writing, China's progress in industrial and societal applications is faster, delivering greater productivity," Ma Jihua, a veteran telecom industry observer, told the Global Times on Sunday.
京公網安備 11010202009201號