Image Courtesy: USAF

China has developed a compact artificial intelligence system designed to help air-to-air missiles identify the infrared signatures of stealth fighters such as the F-22 and F-35, with researchers reporting accuracy as high as 97.1% in simulated laboratory testing.

The system was developed by researchers from the Beijing Institute of Technology and China Airborne Missile Academy for use with missile-mounted infrared imaging systems. The researchers say the technology is designed to overcome the limited computing power, weight and space available inside modern missile seekers. The findings were reported by the researchers, according to South China Morning Post.

Unlike radar-based detection, infrared systems search for heat produced by aircraft. Engines, exhaust systems and aerodynamic friction all generate infrared emissions, creating signatures that can potentially be detected even when an aircraft is designed to minimize its radar cross-section.

That presents both an opportunity and a challenge for missile seekers. Aircraft can deploy flares to create additional heat sources and confuse infrared-guided weapons, meaning a missile must rapidly determine whether the signal it detects belongs to the intended aircraft or a decoy.

To train the AI system, the researchers used 3,245 infrared images collected by a missile-borne scanning system. The dataset contained three categories of airborne targets, including simulated representations of the F-22 and F-35. Testing produced a reported recognition accuracy of 97.1%, while another evaluation recorded accuracy of roughly 90% for the simulated fighter targets.

The results do not demonstrate that the system can identify operational F-22 or F-35 aircraft in combat. Lead researcher An Jiangshan said the relevant real-world test data is confidential and cannot be publicly released.

The researchers also focused heavily on making the AI small enough for missile applications. They reduced the model to 16.1% of the parameters used by previous approaches and cut its computational requirements to 19.2%.

A dedicated AI accelerator further improved performance. During hardware testing, the system reportedly achieved 96.4% accuracy while processing an infrared image in approximately 1.5 milliseconds. Power consumption was about 2.2 watts.

Such processing speeds could be important in an engagement, where a missile may have only moments to classify what its seeker detects. Performing the analysis onboard would also eliminate the need to rely on external computing systems.

The research does not establish that Chinese missiles can reliably track operational stealth fighters. The experiments relied on simulated targets, and the researchers said additional work is needed to improve both speed and recognition accuracy.

Still, the technology highlights a persistent challenge for stealth aircraft. Radar-evading designs can reduce detection from certain sensors, but they cannot completely eliminate the heat produced by an aircraft’s engines and movement through the atmosphere. More capable infrared sensors paired with lightweight AI could make those unavoidable signatures increasingly useful for future missile guidance systems.