Enhancing Fire Detection With Yolov10: Advanced Techniques For Flame And Smoke Recognition

Abstract

Fire detection has become a critical research area due to its importance in protecting human life, infrastructure, and environmental resources. Traditional fire detection systems such as smoke sensors and heat alarms are often limited by delayed response, false alarms, and poor performance in large or complex environments. With the increasing use of surveillance cameras in industries, public spaces, forests, and residential areas, vision-based fire detection has emerged as an effective solution for early fire identification.

This research presents a deep learning-based approach for real-time flame and smoke detection using the YOLOv10 (You Only Look Once) object detection model. The proposed system improves detection accuracy by addressing challenges such as cluttered backgrounds, low visibility, varying fire intensities, overlapping objects, and smoke diffusion patterns. The system processes image and video streams, detects flames and smoke, tracks fire-related regions, and provides accurate localization through improved bounding box regression.

An enhanced feature extraction mechanism and attention-based learning strategy are incorporated to improve the model’s focus on critical fire regions while reducing background interference. Experimental results show that the proposed YOLOv10-based system outperforms existing models such as YOLOv5s and YOLOv8 in terms of accuracy, precision, recall, and F1-score while maintaining real-time performance. The system offers a practical and scalable solution for industrial safety, smart surveillance, forest fire monitoring, and public safety applications.