China is developing an AI-powered ultra-fast torpedo designed to dominate underwater warfare.
Researchers from the PLA Navy and China State Shipbuilding Corporation combined physics with machine learning to teach a torpedo to think before striking.
First, the team simulated decoy profiles using bubble-collapse and turbulence models.
They subsequently trained AI with generative adversarial networks (GANs) to refine and detect acoustic fakes.
The researchers then used deep-learning sonar analysis, converting signals into “spectral thumbnails” via a mathematical tool known as Fourier transforms, which breaks down a complex signal into its basic building blocks.
The result:
Boosted detection rates against advanced decoys from 61.3% to over 80%
The new system achieved a 92.2% success rate in identifying real submarines vs. decoys, a Command Control & Simulation report shows