Article Details

Abstract

To solve the deployment challenge of the forest fire detection model, a lightweight algorithm for enhancing YOLOv8n is put forward: employing GhostConv lightweight convolution to substitute some traditional convolutions;introducing the C2f-Faster module to optimize the feature fusion process, thereby enhancing adaptability to complex scenarios by integrating multi-scale features; combining the C2f-Faster-EMA module to strengthen the model's ability to capture features of faint and blurred smoke and early tiny flames;utilizing the Efficient Detect module to simplify the structure of the detection head. Experimental outcomes demonstrate that the optimized model reduces the parameter quantity by 36.7%, lowers GFLOPs by 51.9%, and raises the mean average precision by 1.4 percentage points, enhancing the detection accuracy on the basis of being lightweight.

Keywords

fire smoke detection; YOLOv8n; lightweighting; Attention Mechanism

References

1. [1] Liu Yueshan Research and Application of Forest Fire Detection Method Based on Improved Yolov7 [D]. Tianjin: Tianjin University of Technology, 2024
2. [2] Zhang Zhi, Feng Shuangwen An improved YOLOv5 video flame real-time detection algorithm [J]. Computer Applications and Software, 2024, 41 (12): 255-260302
3. [3] Pi Jun, Liu Yuheng, Li Jiuhao Research on Lightweight Forest Fire Detection Algorithm Based on YOLOv5s [J]. Journal of Graphic Science, 2023, 44 (1): 26-32
4. [4] Xu Ruijie, Xie Hui, Jiang Wujin, etc Lightweight initial forest fire detection algorithm integrating multi-scale attention [J]. Electronic Measurement Technology, 2025, 48 (15): 80-90
5. [5] Cui Kebin, Geng Jiachang Multi scene fire sign detection algorithm based on EE-YOLOv8s [J]. Journal of Graphic Science, 2025, 46 (1): 13-27
6. [6] Lei Jianyun, Tian Zuohan, Xia Meng, etc Research on Forest Fire Detection Method Based on Improved YOLOv8 [J]. Journal of Central South University for Nationalities: Natural Science Edition, 2026, 45 (1): 97-105
7. [7] LIU G，SHIH K J，WANG T C，et al. Partial Convolution based Padding [J/OL].arXiv:1811.11718 [cs.CV].（2018-11-28）.https://arxiv.org/abs/1811.11718?context=cs.
8. [8] OUYANG D L，HE S，ZHANG G Z，et al. Efficient Multi-Scale Attention Module with Cross-Spatial Learning [C]//ICASSP 2023-2023 IEEE International Conference on Acoustics，Speech and Signal Processing (ICASSP).Rhodes Island：IEEE，2023：1-5.
9. [9] CHEN G J，HOU Y J，CUI T，et al. YOLOv8-CML: A Lightweight Target Detection Method for Color-Changing Melon Ripening in Intelligent Agriculture [J/OL].Scientific reports，2024，14（1）：14400（2024-06-22）.https://doi.org/10.1038/s41598-024-65293-w.
10. [10] HAN K，WANG Y H，TIAN Q，et al. GhostNet: More Features From Cheap OperaTions [C]//Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition.Seattle：IEEE，2020：1577-1586.

Funding

Key Project of Science and Technology Strategy Research in Shanxi Province (202304031401011)

About the authors

Wang Xiaojing (1995-), female, Han ethnicity, from Taiyuan, Shanxi, is a master's student with a research focus on object detection; Wu Junjie (1974-), male, Han ethnicity, born in Shouyang, Shanxi, is a master's supervisor and associate professor. His research interests include data mining and artificial intelligence processing; Guo Ruicheng (2000-), male, Han ethnicity, from Xinzhou, Shanxi, currently pursuing a master's degree in intelligent optimization computing and blockchain technology; Xi Chenran (2000-), male, Han ethnicity, from Fanzhi, Shanxi, is currently pursuing a master's degree in intelligent optimization computing and blockchain technology.