Enhancing Fire Detection with YOLOv10: Advanced Techniques for Flame and Smoke Recognition

Fire detection remains a critical challenge in public safety, industrial operations, and environmental monitoring due to the complex and dynamic nature of flames and smoke. This paper presents an advanced fire and smoke detection framework built upon the YOLOv10 object detection architecture, specifically engineered to overcome the limitations inherent in earlier detection methodologies. The proposed system incorporates an enhanced multi-scale feature extraction mechanism, a sophisticated spatial attention module, and an improved bounding box regression approach to achieve robust localization of fire-related hazards under challenging conditions such as cluttered backgrounds, low visibility environments, overlapping objects, and high-resolution video streams. Trained and evaluated on a labelled fire-and-smoke dataset with two target classes, the YOLOv10-based system demonstrates significant improvements over the baseline YOLOv5s model, achieving higher precision, recall, and mean average precision (mAP) scores in comparative experiments. The system processes inputs in real time, rendering it suitable for deployment in diverse settings including industrial plants, residential buildings, public infrastructure, and wildfire surveillance. These findings affirm the viability of YOLOv10 as a practical and scalable solution for next-generation fire safety systems.