Deep Learning-Based Fire Detection System Using Image-Preprocessing Techniques on Thermal Imaging Data

Abstract

Fires can spread rapidly within a short period of time, causing significant human casualties and property damage, thus necessitating early fire detection technologies. However, conventional sensor-based fire detection methods suffer from detection delays and false alarms and exhibit performance limitations under various environmental conditions. Therefore, this paper proposes a fire detection system based on thermal imaging camera data, image preprocessing techniques, and the YOLO v8 object detection deep learning model. Through preprocessing, flame regions were effectively emphasized, while non-fire elements were removed, improving the detection performance of the model. In addition, to overcome the limitations of acquiring actual thermal imaging data, easily obtainable RGB flame images were processed using an HSV color conversion algorithm and utilized as training data by comparing them with thermal imaging camera data. The experimental results demonstrated that the proposed method improved fire detection performance and exhibited stable convergence characteristics while facilitating flame detection in various environments.