Edge Intelligence for Fire Disaster Mitigation Using IoT and TinyML

Authors:

Abstract
Fire incidents are a major problem around the world. Every year, there are more than 1.3 million fires, which kill more than 3,000 people and cause more than $25 billion in property damage. Traditional fire detection systems don’t work very well in places where resources are limited or far away because they have high false alarm rates, slow response times, and rely on centralized processing. This work presents the development of an autonomous IoT and TinyML-based fire detection system that addresses these limitations through edge-based machine learning and multi-sensor fusion. The proposed system combines an Arduino Nano 33 BLE Sense microcontroller with MQ-2 carbon monoxide, MH-Z19B carbon dioxide, AHT21B temperature, and GPS sensors to get complete fire signatures. The created TinyML Model got 94.71% accuracy on validation and 93.35% accuracy on deployment after Int8 quantization. The TinyML implementation allows for real-time inference with a latency of 1ms and low power use, making it ideal for solar-powered autonomous operation. The performance analysis shows that the quantized model still works well for classification while using less processing power for edge deployment. The system has multiple ways to send alerts, such as local buzzers, integration with a cloud platform, and mobile notifications. This makes sure that emergencies are handled quickly.
Funding Information
  • University of Rwanda