Abstract: Aiming at the problems of single detection parameter of smoke detectors in aircraft cargo compartment, high false alarm rate and no visual interaction, a composite fire detection system based on machine learning using Raspberry Pi is designed and implemented. Raspberry Pi is used as control center and to connect CO sensor, TVOC sensor, PM10 sensor for data collection. The KNN algorithm is used for multi-sensor data fusion, the output result is fire and no fire, the accuracy rate reaches 98%, and the processed data is stored in the SQLite database. The Raspberry Pi builds a Web server and connects to the onboard server local area network, the device realizes the interaction of the visual interface by accessing the HTML web page. Through the experimental test, the system can generate an alarm indication within 46 s in a cargo hold with a height of 2 m. The functional indicators meet the design requirements of cargo hold fire detection, and the underreport rate is 0 and the false alarm rate is less than 1%, and provides a reliable solution for the design of the airborne fire detection system.

Key words: Raspberry Pi, multi-sensor, KNN algorithm, visual interface