Design of an intelligent fire warning system based on STM32 microcontroller and edge AI

Abstract

To address the problems of slow response, high false alarm rate, single detection dimension, and lack of intelligent linkage in traditional fire warning systems, this paper designs and implements an intelligent fire early warning system based on a dual STM32F103C8T6 master-slave architecture and a K230 edge computing unit. The system integrates smoke, temperature-humidity, and flame sensors along with the YOLO vision model. It adopts a weighted voting-based multi-source data fusion decision mechanism. Under interference conditions such as low illumination and occlusion, the system enhances robustness through image enhancement and a temporal voting mechanism. Data is wirelessly transmitted to the OneNET cloud platform and a WeChat mini-program via the ESP8266 module using the MQTT protocol. When a fire is detected, the system can automatically execute emergency responses including power cutoff, smoke exhaust fan activation, and water pump startup. Test results show that the system operates stably, responds quickly, achieves high recognition accuracy and low false alarm rate, making it suitable for early fire warning in homes and small spaces.