Comparison of YOLOv5 and YOLOv8 Models' Performance in Early Fire and Smoke Detection Applications

Abstract

Abstract: In this paper, we conducted a comparative study on the performance of the YOLOv5 and YOLOv8 models for fire and smoke detection across various scenarios. Utilizing a dataset of 9,756 images capturing diverse fire incidents under different environmental conditions, both models were trained using identical hyperparameters (learning rate = 0.001, batch size = 16, and 40 epochs). YOLOv8 consistently outperformed YOLOv5 in terms of accuracy, precision, and recall across various evaluation metrics, highlighting the architectural improvements introduced in YOLOv8. Notably, YOLOv8 performed better in detecting fires in complex scenarios, such as small fires and challenging lighting conditions, where YOLOv5 faced difficulties. However, both models faced challenges in detecting transparent smoke, particularly in daylight. The results suggest that while YOLOv8 holds promise for further improvements in fire detection, expanding the dataset and exploring more advanced configurations could lead to better performance in real-world applications. This research emphasizes the importance of model selection based on specific applications and the potential of the latest YOLO versions to enhance early fire detection.