Artificial Intelligence for Early Fire State Detection: Training Neural Networks on Environmental Sensor Data

View at publisher Summarise with VIRAT
Authors: Samoel Ortu, Chiara Matta, Marco Pireddu, Giuseppe Fancello, Daniele Giusto, Matteo Anedda
Cited by: 0
References: 9

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This paper presents a lightweight and reproducible framework for fire-state classification based on environmental data collected from Bosch BME688 sensors. The proposed system combines four primary physical parameters — temperature, relative humidity, pressure, and gas resistance — into a compact neural network for accurate detection of different fire phases. Controlled ignition and extinction experiments were conducted to build a labeled dataset covering steady and transitional conditions. Experimental results show that the model effectively discriminates between No Fire, Start, Fire, and End states, achieving reliable performance with minimal computational complexity. The approach demonstrates robustness and suitability for embedded deployment on low-power IoT nodes. This work contributes to the development of data-driven, edge-compatible fire monitoring systems capable of supporting early detection and situational awareness in autonomous environments.

Fields of Study & Subjects

Fire Detection and Safety Systems Fire dynamics and safety research Smart Agriculture and AI

Publication details

Journal / source
International Conference on Consumer Electronics
Pages
1–6
Publication date
3 February 2026
Language
EN