Bioinspired Machine-Learning-Assisted Early-Fire Perception System Based on VO2 Optical Switch

The continuation of human civilization has always been accompanied by symbiosis and confrontation with fire. Particularly, humans can comprehensively recognize fire situations based on various sensory receptors in organs (eyes, skin, nose, etc.), further forming a sound fire perception system by in-the-brain recording, modeling, and understanding fire behaviors, leading to the most accurate fire treatments. If a sensing perception system can mimic human perceptual behavior and carry out real-time fire recognition, such an active defense system can achieve real fire safety. Here, inspired by the brain-centered perception system, an early-fire perception system enabled by a VO2-based temperature-flame-modulated optical switch, and a machine-learning-assistant demodulation algorithm is reported. This approach creates real-time monitoring composed of early fire warning (1 s for candle flame and 4 s for 130 degrees C heat flow), fire cause recognition (95.7% accuracy in identification), and evacuation advice provision, advancing the technologies in the perception system that enable future sensors the comprehensive perception capability for fire state.