Early Detection of Wildfire using IoT Sensors without Labelled Fire Data

Abstract

Wildfires, commonly referred to as bushfires in Australia and forest fires in Europe, pose significant threats to life, property, and the environment. Prompt identification of wildfires is crucial to prevent them from escalating into catastrophic events. This paper considers the use of wireless sensors to measure environmental conditions, in particular carbon dioxide, to detect wildfire at its early stages. We propose a detection algorithm that leverages exclusively non-fire measurement data, without relying on labelled fire data. This differs significantly from earlier research in this field, which generally presumes the existence of labelled fire data. In addition, our algorithm takes into account the temporal and spatial correlations in the sensor measurements to determine the existence of fire. Experimental burns are conducted in a campsite by igniting a small fire from a 0.7 square metres fire pit. The result shows that our algorithm applied to the experimental data can be tuned to minimise false alarms, at the same time detecting small fires within 50 metres and within 4 to 14 minutes after ignition. Show more