基于改进 YOLOv8 轻量级电力管廊初期火灾检测方法的研究

Early Fire Detection for Lightweight Power Tunnel Based on Improved YOLOv8

摘要

针对于传统城市电力管廊内火灾检测算法模型参数量大、计算量大、推理时间长和可见光对于早期火灾图像识别困难等问题,提出了一种基于图像融合(EMAFusion)的早期火灾检测方法,对 YOLOv8 算法进行改进,用于初期火灾检测.将早期火灾红外图像与可见光图像融合用于火灾隐患的检测,得到热红外信息又补充可见光纹理信息、轮廓特征,调节弥补了红外与可见光特征.相较于单一图像,融合图像可以提升火灾检测的准确率以及模型在复杂场景下的鲁棒性.此外,构建了一种基于改进的 YOLOv8n 的早期火灾检测算法模型,在特征融合阶段 Neck 端引入 AFPN(渐近特征金字塔网络)通过自适应权重融合和多尺度特征融合,AFPN 能够显著提升模型对小目标的检测能力；引入 LADH 检测头在保持轻量化的同时提高模型的检测精度；最后使用 WIoU 损失函数优化网络的边界损失,提高网络模型检测准确性以及加快模型的收敛.实验结果表明,该网络在所使用的模拟电力管廊火灾数据集上的准确率提高了 2.3%,参数量下降了 23.88%,计算量下降了 17.23%,其中检测速度 FPS 相比于基准模型提升了 34.65%.算法精度不仅得到了提升又满足了轻量化的需求,可以满足电力管廊场景下的实时检测.

Abstract

Traditional urban power tunnel fire detection algorithms suffer from several drawbacks,including a large parameter count,high computational load,long inference time,and the difficulty in early fire image recognition using visible light. To address these problems,a novel early fire detection method based on image fusion(EMA Fusion)is proposed,which improves the YOLOv8 algorithm for initial fire detection. This method fuses infrared and visible light images of early fires to detect fire hazards,obtaining both thermal infrared information and visible light texture and contour features. Compared to single-image detection,the fused images enhance the accuracy of fire detection and the robustness of the model in complex scenarios. Moreover,an early fire detection algorithm model is constructed based on the improved YOLOv8n. In the feature fusion stage,the asymptotic feature pyramid network(AFPN)is introduced at the Neck end. Through adaptive weight fusion and multi-scale feature fusion,this network can significantly enhance the model's ability to detect small targets. The LADH detection head is incorporated to improve detection accuracy while maintaining a lightweight structure. Lastly,the WIoU loss function is used to optimize the boundary loss of the network,thereby increasing detection accuracy and accelerating model convergence. Experimental results indicate that on the simulated power tunnel fire dataset,the accuracy of this network increased by 2.3%,while the parameter count decreased by 23.88% and the computational load by 17.23%. The detection speed,measured in frames per second(FPS),was improved by 34.65% compared to that of the baseline model. The algorithm not only achieves higher precision but also meets the requirements for lightweight design,making it suitable for real-time detection in power tunnel scenarios.

关键词

EMAFusion / 渐进特征金字塔 / LADH 检测头 / WIoU / 轻量化

Key words

EMA Fusion / adaptive feature pyramid network(AFPN) / LADH detection head / WIoU / lightweight

参考文献

[1] Ko J. Study on the fire risk prediction assessment due to deterioration contact of combustible cables in underground common utility tunnels[J]. Journal of the Society of Disaster Information, 2015, 11(1): 135-147.

[2] 曾晓亮, 李富祥, 李明伟, 等. 电缆隧道火灾事故特点与灭火系统研究综述[J]. 四川电力技术, 2022, 45(4): 55-59.

[3] Zeng Xiaoliang, Li Fuxiang, Li Mingwei, et al. Review on characteristics of cable tunnel fire accidents and fire—extinguishing systems[J]. Sichuan Electric Power Technology, 2022, 45(4): 55-59(in Chinese).

[4] 杨立中, 叶开. 城市综合管廊消防标准及火灾研究综述[J]. 中国安全科学学报, 2021, 31(8): 132-140.

[5] Yang Lizhong, Ye Kai. Review on fire protection standards and fire research of urban utility tunnels[J]. China Safety Science Journal, 2021, 31(8): 132-140(in Chinese).

[6] 许德胜, 李炎锋, 杨泉. 综合管廊电力舱电缆火灾研究进展[J]. 消防科学与技术, 2024, 43(9): 1213—1217, 1246.

[7] Xu Desheng, Li Yanfeng, Yang Quan. Research progress on cable fires in power cabins of utility tunnels[J]. Fire Science and Technology, 2024, 43(9): 1213—1217, 1246(in Chinese).

[8] 张敏, 张明灿, 李兵兵, 等. 综合管廊电力舱散热计算及通风研究[J]. 地下空间与工程学报, 2024, 20(S2): 916-923.

[9] Zhang Min, Zhang Mingcan, Li Bingbing, et al. Heat dissipation calculation and ventilation research on power cabin of utility tunnel[J]. Chinese Journal of Underground Space and Engineering, 2024, 20(S2): 916-923(in Chinese).

[10] 张晋, 徐大军, 宋文琦, 等. 城市综合管廊电力舱火灾行为试验研究[J]. 地下空间与工程学报, 2020, 16(6): 1818-1825.

[11] Zhang Jin, Xu Dajun, Song Wenqi, et al. Experimental study on fire behavior of power cabin in urban utility tunnel[J]. Chinese Journal of Underground Space and Engineering, 2020, 16(6): 1818-1825(in Chinese).

[12] 秦瑞, 张为. 一种无锚框结构的多尺度火灾检测算法[J]. 西安电子科技大学学报, 2022, 49(6): 111-119.

[13] Qin Rui, Zhang Wei. A multi—scale fire detection algorithm based on anchor—free structure[J]. Journal of Xidian University, 2022, 49(6): 111-119(in Chinese).

[14] Foggia P, Saggese A, Vento M. Real—time fire detection for video—surveillance applications using a combination of experts based on color,shape,and motion[J]. IEEE TRANSACTIONS on Circuits and Systems for Video Technology, 2015, 25(9): 1545-1556.

[15] Celik T, Demirel H, Ozkaramanli H, et al. Fire detection using statistical color model in video sequences[J]. Journal of Visual Communication and Image Representation, 2007, 18(2): 176-185.

[16] Hashemzadeh M, Zademehdi A. Fire detection for video surveillance applications using ICA K—medoids—based color model and efficient spatio—temporal visual features[J]. Expert Systems with Applications, 2019, 130: 60-78.

[17] 肖霞, 宋卫国, 王彦, 等. 基于类间方差的 MODIS 森林火灾火点检测方法[J]. 光谱学与光谱分析, 2010, 30(8): 2065-2068.

[18] Xiao Xia, Song Weiguo, Wang Yan, et al. MODIS forest fire spot detection method based on inter—class variance[J]. Spectroscopy and Spectral Analysis, 2010, 30(8): 2065-2068(in Chinese).

[19] 冯路佳, 王慧琴, 王可, 等. 基于目标区域的卷积神经网络火灾烟雾识别[J]. 激光与光电子学进展, 2020, 57(16): 83-91.

[20] Feng Lujia, Wang Huiqin, Wang Ke, et al. Fire smoke recognition based on target—region convolutional neural network[J]. Laser & Optoelectronics Progress, 2020, 57(16): 83-91(in Chinese).

[21] 张坚鑫. 基于深度学习的火灾图像分类与检测算法研究[D]. 广州: 广州大学, 2020.

[22] Zhang Jianxin. Research on Fire Image Classification and Detection Algorithm Based on Deep Learning[D]. Guangzhou: Guangzhou University, 2020(in Chinese).

[23] 李悦明, 黄国忠, 王波, 等. 一种基于 CA—SRGAN 的古建筑消防危险品检测模型[J]. 清华大学学报(自然科学版), 2024, 64(11): 1870-1879.

[24] Li Yueming, Huang Guozhong, Wang Bo, et al. Detection model of fire hazard goods in ancient buildings based on CA—SRGAN[J]. Journal of Tsinghua University(Science and Technology), 2024, 64(11): 1870-1879(in Chinese).

[25] 邓力, 周进, 刘全义. 基于改进 YOLOv8 的火焰与烟雾检测算法[J]. 清华大学学报(自然科学版), 2025, 65(4): 681-689.

[26] Deng Li, Zhou Jin, Liu Quanyi. Flame and smoke detection algorithm based on improved YOLOv8[J]. Journal of Tsinghua University(Science and Technology), 2025, 65(4): 681-689(in Chinese).

[27] Chen C, Yu J, Lin Y, et al. Fire detection based on improved PP—YOLO[J]. Signal,Image and Video Processing, 2023, 17(4): 1061-1067.

[28] Xiao L, Li W, Zhang X, et al. EMG—YOLO:An efficient fire detection model for embedded devices[J]. Digital Signal Processing, 2025, 156: 104824.

[29] Yang G, Lei J, Zhu Z, et al. AFPN:Asymptotic feature pyramid network for object detection[C]// 2023 IEEE International Conference on Systems,Man,and Cybernetics(SMC). IEEE, 2023: 2184-2189.

[30] Zhou W, Cai C, Srigrarom S, et al. ADH—YOLO:A small object detection based on improved YOLOv8 for airport scene images in hazy weather[J]. The Journal of Supercomputing, 2025, 81(3): 1-20.

[31] Zheng Z, Wang P, Liu W, et al. Distance—IoU loss:Faster and better learning for bounding box regression[C]// Proceedings of the AAAI Conference on Artificial Intelligence. 2020: 12993-13000.

[32] 毛坤, 朱学军, 赖惠鸽, 等. 多尺度融合优化的印刷电路板缺陷检测算法[J]. 激光与光电子学进展, 2025, 62(10): 141-152.

[33] Mao Kun, Zhu Xuejun, Lai Huige, et al. Defect detection algorithm for printed circuit boards optimized by multi—scale feature fusion[J]. Laser & Optoelectronics Progress, 2025, 62(10): 141-152(in Chinese).

[34] Hu D, Yu M, Wu X, et al. DGW YOLOv8‐:A small insulator target detection algorithm based on deformable attention backbone and WIoU loss function[J]. IET Image Processing, 2024, 18(4): 1096-1108.

[35] Wang J, Chen Y, Sun X, et al. Advancing infrared and visible image fusion with an enhanced multiscale encoder and attention—based networks[J]. iScience, 2024, 27(10): 110915.

基金资助

石家庄市科技计划重点资助项目(241130163A)