China claims AI missile system tracks F-35s with 90% accuracy using heat signatures

Full text

Chinese researchers built an AI targeting infrared traces from F-22 and F-35 engines

Flare decoys reportedly fail to fool this heat-based recognition model

Lab tests show recognition accuracy above 90 percent for simulated targets

Chinese researchers have claimed a lightweight AI system can identify heat signatures resembling those of F-22 and F-35 stealth fighters.

The F-22 and F-35 are designed to make enemy detection and tracking more difficult, particularly by reducing their visibility to radar and other sensors.

That makes any technology capable of recognizing their remaining infrared signatures potentially important for detecting aircraft built around stealth.

AI recognition focuses on aircraft heat

Unlike radar-based detection, infrared systems look for heat produced by an aircraft's engines, exhaust, and heated surfaces during flight.

The researchers claim that AI can analyze those patterns and distinguish simulated F-22 and F-35 signatures from other airborne objects.

Pilots often use flares to confuse conventional heat-seeking sensors during an engagement.

However, the researchers say that aircraft-generated heat differs from flare emissions, giving an AI system another basis for separating an aircraft from decoys.

The researchers tested their recognition model using simulated targets representing the thermal characteristics associated with F-22 and F-35 aircraft.

According to the study's lead author, the approach is intended to combine rapid processing with sufficient recognition capability for missile applications.

“Lightweight recognition models could become widely used in future air-to-air missiles because they can provide high-speed recognition while maintaining strong identification capabilities,” said An Jiangshan, first author of the study.

Laboratory testing reportedly produced recognition accuracy exceeding 90%, although operational performance against real aircraft remains unverified.

That distinction matters because real aircraft generate changing thermal patterns under different speeds, altitudes, maneuvering conditions, and environmental circumstances.

The researchers also describe the system as lightweight, suggesting that its computing requirements could be suitable for missile-mounted hardware with limited space and processing capacity.

Limits remain around real-world performance

Stealth aircraft are designed primarily to reduce radar visibility, but infrared emissions remain an important consideration for infrared-guided weapons.

Heat-seeking missiles already use infrared sensors, while machine-learning systems could potentially assist those sensors with identifying complex thermal patterns.

However, the reported research does not establish that the system can reliably track operational F-22 or F-35 aircraft under combat conditions.

Actual engagements would introduce atmospheric conditions, changing viewing angles, aircraft maneuvers, background temperatures, and electronic countermeasures that laboratory testing may not fully reproduce.

Even without proving the system can defeat operational aircraft, the research points toward a growing challenge for American stealth platforms.

The United States cannot assume that reducing radar visibility will remain sufficient as AI systems become better at recognizing infrared patterns.

Future F-22 and F-35 upgrades could therefore require greater attention to heat management, exhaust signatures and other infrared characteristics.

The same concern applies to future stealth bombers, which could face increasingly capable AI-assisted infrared sensors.

Stealth designs may need to consider how machine-learning systems interpret heat patterns rather than simply minimizing the strength of those emissions.

That could become increasingly important as lightweight AI processors become easier to integrate into missiles and other airborne weapons.

American defence planners have little reason to wait until such systems demonstrate their capabilities during actual combat.

Via SCMP