Chinese military scientists have developed an AI system that can accurately distinguish real submarines from decoys, even during high-speed underwater combat. The system achieved a 92.2% success rate using advanced neural networks and sonar analysis.

China’s military researchers have unveiled a powerful artificial intelligence system that can detect real submarines even in waters filled with high-tech decoys, an innovation that could reshape underwater warfare.

The announcement came just months after the release of the Chinese blockbuster Operation Leviathan, which featured a dramatic scene where an American submarine uses acoustic illusions to escape incoming torpedoes, reports South China Morning Post. Now, researchers from the PLA Navy Armament Department and China State Shipbuilding Corporation claim to have developed the real-world answer to such deception.

In a peer-reviewed study published in April in the Chinese-language journal Command Control & Simulation, the scientists said their AI system was able to correctly identify genuine targets with 92.2% accuracy, even when torpedoes were travelling at high speeds and surrounded by decoys.

Beating the bubble trail

Modern submarine warfare increasingly relies on misleading torpedoes using acoustic holograms, false bubble trails, and sonar decoy swarms. These tricks, especially when used alongside supercavitating torpedoes, weapons that travel at extreme speeds using vapor bubbles to reduce drag, make detection difficult.

These ultra-fast torpedoes often create so much noise and distortion that older systems fail to distinguish real targets.

But the Chinese research team, led by engineers Wu Yajun and Liu Liwen, stated, "Existing systems are inadequate when facing decoy-heavy environments. Only long-range, high-accuracy recognition can ensure mission success."

AI trained on decoy deception

To develop a solution, the scientists combined machine learning with hydrodynamic physics. They started by simulating how real and fake submarines behave underwater, including bubble collapse and turbulence patterns, using classified data from China’s high-speed torpedo test ranges.

They then trained their AI model using a generative adversarial network (GAN), a system where one neural network creates decoy signals and another learns to spot the flaws.

This dueling process allowed the system to build a large collection of realistic submarine and decoy acoustic profiles. The AI then used Fourier transforms to turn sonar signals into “sonic thumbnails,” and passed them through convolutional layers which is a method adapted from image recognition to detect subtle differences.

When tested against the most advanced fake signals, the AI’s recognition rate jumped from 61.3% to over 80%, and up to 92.2% in real-time torpedo conditions.

Racing against the world

China’s research comes amid an international push to build smarter undersea weapons. The Russian VA-111 Shkval and similar Western torpedoes already use supercavitation, but all face challenges in distinguishing real targets during high-speed engagements.

The Chinese team notes that today’s underwater battlefield is more complex than ever, with multiple simultaneous threats, decoys, jammers, and electronic countermeasures, crowding the sonar screen.

“Since these torpedoes are autonomous and cannot rely on real-time communication, everything must be decided instantly and independently,” the paper added. That makes high-level AI not just useful, but essential.

A step toward deployment

The researchers concluded that their deep-learning recognition system, backed by GAN-generated training data, provides the technical foundation for real-world deployment in Chinese naval forces.

By addressing one of the most complex problems in underwater warfare, discerning fact from illusion, China’s new AI technology might be the beginning of a new era where machines, not humans, make the split-second decisions beneath the sea.