CSI-Based Flame Detection Method Based on Antenna and Signal Propagation Characteristics

Indoor fire poses a critical threat to life and property, yet traditional detection methods face notable limitations. Smoke detectors may cause false alarms or missed detections due to dust accumulation, temperature sensors respond slowly and are influenced by ambient fluctuations, and video-based methods require optimal lighting, struggle with occlusions, and raise privacy concerns. More importantly, these sensors are unable to detect flames in non-line-of-sight (NLoS) spaces, such as fires in adjacent apartments of high-rise residential buildings at night, which can lead to casualties due to the lack of warning before the fire spreads. To address those problems, this paper introduces an NLoS flame detection method based on antenna impedance and wireless signal propagation characteristics. Since Channel State Information (CSI) can reflect the relationship between antenna impedance and signal propagation under the influence of environmental factors, CSI data extracted from widely deployed home Wi-Fi networks with through-the-wall capability is used as the input data of the proposed method. Finally, an NLoS flame detection method based on machine learning has been proposed, and the relationships between detection performance, signal propagation characteristics and flame cross-sectional area have been also deeply analyzed. The experimental results show that the flame detection method in NLoS scenarios can overcome the limitations of traditional methods, and provide a new solution for fire detection in NLoS environments. The related dataset for this paper is posted at https://github.com/T-bjq/WCFD-CSI-dataset.