The Integration Approach of LSTM and Yolo V8 for Fire Propagation Detection

Fire accidents are among the most common disasters, posing significant risks of severe personal injury and property damage. Rapid detection of fire incidents is crucial to prevent these damages effectively. Visual surveillance plays a vital role in ensuring accurate fire detection, minimizing false alarms while maintaining prompt responsiveness. In this paper, we present an advanced fire detection system that integrates YOLO v8 with an LSTM network to not only detect fires accurately but also predict fire propagation. Initially, YOLO v8 is used to process the dataset and develop a pre-trained model. This model is then re-trained using sequential fire incident videos to extract key features and fire locations, which serve as inputs for the LSTM to learn the fire's spread dynamics. In the final stage, the LSTM predicts fire propagation, providing an additional confirmation of fire emergencies. The proposed method achieves promising and convincing results, with a mean Average Precision (mAP) of 91.9% and a precision of 94.4%.