How to Research on Fire Smoke Detection Algorithm in Enclosed Spaces Based on Improved YOLOv8

Abstract Recognizing the critical importance of timely fire detection to minimize property damage and casualties, traditional sensor-based fire detection systems are expensive and are not suitable for large or outdoor environments. Herein, we introduce a novel fire smoke detection algorithm that utilizes multi-scale attention fusion to enhance the early detection and reporting of fire in enclosed spaces. The algorithm uses grouped sparse convolution (GSConv), a SwinTrans precursor, and a VoVGSCSP module to reconstruct the neck structure of the YOLOv8 network. This integration allows for efficient and accurate detection of both smoke and flames, even in the early stages of a fire when these signs are subtle and challenging to identify. Our approach is validated through the creation and use of three high-quality datasets, including real fire, real smoke, and two-object datasets with both smoke and fire. The model’s performance is evaluated using precision, recall, F1-score, and Average Precision (AP) metrics, demonstrating significant improvements over existing YOLOv5s and YOLOv8s models. The proposed algorithm achieves a notable F1-score of 95.00% and a recall rate of 90.34%, underscoring its effectiveness for real-world applications. An ablation study further confirms the contributions of each component in the network architecture. The research accomplishment has significant practical value for smart building systems, industrial safety monitoring, and public safety, offering a robust technological solution for early fire detection and prevention.