YOLOv11-based Flame Recognition Algorithm Utilizing a Fusion Dual-stream Attention Mechanism

Rui Gong, Qiang Li, Jingyu Li
Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, PMLR 278:77-86, 2025.

Abstract

Flame change characteristics are affected by ignition source, air pressure, wind direction and other factors, and the traditional method describes the problems of leakage, false alarm and poor real-time performance. Vision-based image detection is one of the important means to solve the above problems. Therefore, a YOLOv11 optimized flame detection algorithm is proposed. First, the feature extraction PCF module is designed to enhance the characterization of different layers of feature maps. Second, the model incorporates the dual-stream mechanism attention mechanism to improve the attention to different scale features. Finally, the model introduces an improved Focal Loss function to optimize the regression accuracy and network robustness in the prediction region. The model is subjected to comparative experiments on the Flame public dataset. The results show that the improved model performs well for flame smoke detection in complex scenarios, reaching 49.6% on mAP50 and 18.9% on mAP75, which is an improvement of 2% and 8% in accuracy compared to the original YOLOv11 model.

Cite this Paper

BibTeX
@InProceedings{pmlr-v278-gong25a, title = {YOLOv11-based Flame Recognition Algorithm Utilizing a Fusion Dual-stream Attention Mechanism}, author = {Gong, Rui and Li, Qiang and Li, Jingyu}, booktitle = {Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing}, pages = {77--86}, year = {2025}, editor = {Zeng, Nianyin and Pachori, Ram Bilas and Wang, Dongshu}, volume = {278}, series = {Proceedings of Machine Learning Research}, month = {25--27 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v278/main/assets/gong25a/gong25a.pdf}, url = {https://proceedings.mlr.press/v278/gong25a.html}, abstract = {Flame change characteristics are affected by ignition source, air pressure, wind direction and other factors, and the traditional method describes the problems of leakage, false alarm and poor real-time performance. Vision-based image detection is one of the important means to solve the above problems. Therefore, a YOLOv11 optimized flame detection algorithm is proposed. First, the feature extraction PCF module is designed to enhance the characterization of different layers of feature maps. Second, the model incorporates the dual-stream mechanism attention mechanism to improve the attention to different scale features. Finally, the model introduces an improved Focal Loss function to optimize the regression accuracy and network robustness in the prediction region. The model is subjected to comparative experiments on the Flame public dataset. The results show that the improved model performs well for flame smoke detection in complex scenarios, reaching 49.6% on mAP50 and 18.9% on mAP75, which is an improvement of 2% and 8% in accuracy compared to the original YOLOv11 model.} }
Endnote
%0 Conference Paper %T YOLOv11-based Flame Recognition Algorithm Utilizing a Fusion Dual-stream Attention Mechanism %A Rui Gong %A Qiang Li %A Jingyu Li %B Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing %C Proceedings of Machine Learning Research %D 2025 %E Nianyin Zeng %E Ram Bilas Pachori %E Dongshu Wang %F pmlr-v278-gong25a %I PMLR %P 77--86 %U https://proceedings.mlr.press/v278/gong25a.html %V 278 %X Flame change characteristics are affected by ignition source, air pressure, wind direction and other factors, and the traditional method describes the problems of leakage, false alarm and poor real-time performance. Vision-based image detection is one of the important means to solve the above problems. Therefore, a YOLOv11 optimized flame detection algorithm is proposed. First, the feature extraction PCF module is designed to enhance the characterization of different layers of feature maps. Second, the model incorporates the dual-stream mechanism attention mechanism to improve the attention to different scale features. Finally, the model introduces an improved Focal Loss function to optimize the regression accuracy and network robustness in the prediction region. The model is subjected to comparative experiments on the Flame public dataset. The results show that the improved model performs well for flame smoke detection in complex scenarios, reaching 49.6% on mAP50 and 18.9% on mAP75, which is an improvement of 2% and 8% in accuracy compared to the original YOLOv11 model.
APA
Gong, R., Li, Q. & Li, J.. (2025). YOLOv11-based Flame Recognition Algorithm Utilizing a Fusion Dual-stream Attention Mechanism. Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, in Proceedings of Machine Learning Research 278:77-86 Available from https://proceedings.mlr.press/v278/gong25a.html.

Related Material