YOLOv5-GoldSOSA: Integrating Gold-YOLO and RCS-OSA for an Enhanced Flame Detection System

With the increasing risk of fire incidents in industrial and living environments, timely and accurate flame detection is crucial for preventing fire accidents and reducing casualties. This study introduces a novel flame detection system named YOLOv5-GoldSOSA, which integrates the advanced target det...

Full description

Saved in:
Bibliographic Details
Published in2025 International Conference on Electrical Drives, Power Electronics & Engineering (EDPEE) pp. 1025 - 1029
Main Author Zhao, Zichen
Format Conference Proceeding
LanguageEnglish
Published IEEE 26.03.2025
Subjects
Accuracy
Computer architecture
Fire safety
Fires
Flame Detection
Gold-YOLO
Lighting
Neck
Object Detection
RCS-OSA
Robustness
Thermal sensors
Training
YOLO
YOLOv5
Online AccessGet full text

Cover

More Information
Summary:With the increasing risk of fire incidents in industrial and living environments, timely and accurate flame detection is crucial for preventing fire accidents and reducing casualties. This study introduces a novel flame detection system named YOLOv5-GoldSOSA, which integrates the advanced target detection models Gold-YOLO and RCS-OSA to enhance detection speed and accuracy. YOLOv5-GoldSOSA, based on the YOLOv5 architecture, leverages the multi-scale feature fusion capabilities of Gold-YOLO and the efficient information exchange mechanism of RCS-OSA to achieve rapid flame recognition. Extensive experiments conducted on various real-world flame datasets demonstrate that YOLOv5-GoldSOSA significantly improves inference speed while maintaining high detection accuracy. Specifically, after 218 epochs of training, YOLOv5-GoldSOSA achieved a precision (P) of 0.66, recall (R) of 0.571, mAP50 of 0.622, and mAP50-95 of 0.311, showing noticeable performance improvements over the baseline YOLOv5N (which after 268 epochs achieved a precision of 0.615, recall of 0.568, mAP50 of 0.584, and mAP50-95 of 0.303). Additionally, YOLOv5-GoldSOSA excels in detecting small-sized flames and flames in complex backgrounds, demonstrating its potential for practical applications. This study not only provides a new technical solution for the field of flame detection but also offers new insights for future research in deep learning-based target detection.
DOI:10.1109/EDPEE65754.2025.00185