面向柴油车辆排放黑烟的改进型YOLOv8检测算法研究

doi:10.11996/JG.j.2095-302X.2025020249

图学学报 ›› 2025, Vol. 46 ›› Issue (2): 249-258.DOI: 10.11996/JG.j.2095-302X.2025020249

• 图像处理与计算机视觉 • 上一篇下一篇

面向柴油车辆排放黑烟的改进型YOLOv8检测算法研究

张立立¹^,³(), 杨康¹, 张珂¹, 魏薇¹, 李晶¹, 谭洪鑫²(), 张翔宇³

1.北京石油化工学院信息工程学院，北京 102617
2.中国人民解放军93184部队，北京 100076
3.星禾环保科技有限公司，河南郑州 450001

收稿日期:2024-07-22 接受日期:2024-11-25 出版日期:2025-04-30 发布日期:2025-04-24
通讯作者:谭洪鑫(1992-)，女，工程师，博士。主要研究方向为图像识别与目标检测。E-mail：18809462629@163.com
第一作者:张立立(1988-)，男，副教授，博士。主要研究方向为智能交通。E-mail：zhanglili@bipt.edu.cn
基金资助:
宁夏自然科学基金(2023AAC03889);北京市数字教育研究课题(BDEC2022619048);北京市教育委员会科研计划项目(KM202410017006)

Research on improved YOLOv8 detection algorithm for diesel vehicle emission of black smoke

ZHANG Lili¹^,³(), YANG Kang¹, ZHANG Ke¹, WEI Wei¹, LI Jing¹, TAN Hongxin²(), ZHANG Xiangyu³

1. School of Information Engineering, Beijing Institute of Petrochemical Technology, Beijing 102617 China
2. PLA, No. 93184 Troop, Beijing 100076, China
3. Xinghe Environmental Protection Technology Co., LTD., Zhengzhou Henan 450001, China

Received:2024-07-22 Accepted:2024-11-25 Published:2025-04-30 Online:2025-04-24
First author：ZHANG Lili (1988-), associate professor, Ph.D. His main research interest covers intelligent transportation. E-mail：zhanglili@bipt.edu.cn
Supported by:
Ningxia Natural Science Foundation(2023AAC03889);Beijing Digital Education Research Project(BDEC2022619048);Research and Development Program of Beijing Municipal Education Commission(KM202410017006)

摘要/Abstract

摘要：

柴油车辆排放黑烟是道路交通环保执法的重点和难点。由于受复杂环境条件的影响，针对目前黑烟检测存在精度和速度方面的不足，提出一种基于改进YOLOv8的轻量级柴油车辆排放黑烟的检测模型。首先，在YOLOv8主干网络的基础上，设计一种轻量化特征提取模块C2f-FasterRep提高模型的特征提取能力，同时C2f-FasterRep模块引入上下文锚框注意力机制模块来捕捉长距离的上下文信息，利用全局平均池化和条形卷积增强特征图中心区域的特征，从而提高检测精度；其次，在颈部部分提出一个新的网络结构用于融合主干网络提取的特征，并使用通道注意力模块和维度匹配机制对不同尺度的特征进行融合，增强了模型的多尺度特征融合能力；最后，使用Transformer解码器结构优化YOLOv8模型的检测头，同时，采用交并比感知的查询机制，有助于解码器查询的优化，提高了模型的分类和定位的性能。为保证实验的真实性和有效性，利用部署在河南许昌某道路断面的检测设备采集数据并进行测试验证。实验结果表明，该方法的mAp为95.4%，精确率为94.5%，召回率为97.5%，与现有的黑烟检测方法相比，具有更高的检测精度和更快的检测速度。消融实验结果表明该轻量化特征提取模块、特征融合模块和检测头有利于提高模型检测精度。

关键词: 目标检测, 黑烟检测, YOLOv8, 轻量级模型, 注意力机制

Abstract:

The emission of black smoke from diesel vehicles is a critical and challenging issue in road traffic environmental protection enforcement. Due to the influence of complex environmental conditions, to address the limitations of existing black smoke detection methods in terms of accuracy and speed, a lightweight detection model of black smoke emission from diesel vehicles based on improved YOLOv8 was proposed. Firstly, based on the YOLOv8 backbone network, a lightweight feature extraction module C2f-FasterRep was designed to improve the feature extraction capability of the model, while the C2f-FasterRep module integrated a context-anchored attention mechanism module to capture long-range contextual information. The global average pooling and bar convolution were employed to enhance the features in the central region of the feature map, thereby improving the detection accuracy. Secondly, a new network structure was proposed in the neck part to fuse the features extracted from the backbone network. The channel attention module and dimensional matching mechanism were employed to fuse features of different scales, enhancing the model’s multi-scale feature fusion capability. Lastly, the detection head of the YOLOv8 model was optimized using a Transformer decoder structure. An intersection-parallel-ratio aware query mechanism was incorporated to optimize the decoder query, improving the model’s performance for classification and localization. In order to ensure the authenticity and effectiveness of the experiment, data were collected and tested and verified using the detection equipment deployed on a road section in Xuchang, Henan Province. The experimental results demonstrated that the mAp of the proposed method was achieved at 95.4%, the precision at 94.5%, and the recall rate at 97.5%. Compared with existing black smoke detection methods, the proposed approach exhibited higher detection accuracy and faster detection speed. The results of ablation experiments confirmed that the proposed lightweight feature extraction module, feature fusion module and detection head contributed to improving the accuracy of model detection.

Key words: object detection, black smoke detection, YOLOv8, lightweight model, attention mechanism

中图分类号:

张立立, 杨康, 张珂, 魏薇, 李晶, 谭洪鑫, 张翔宇. 面向柴油车辆排放黑烟的改进型YOLOv8检测算法研究[J]. 图学学报, 2025, 46(2): 249-258.

ZHANG Lili, YANG Kang, ZHANG Ke, WEI Wei, LI Jing, TAN Hongxin, ZHANG Xiangyu. Research on improved YOLOv8 detection algorithm for diesel vehicle emission of black smoke[J]. Journal of Graphics, 2025, 46(2): 249-258.

图/表 16

图1 YOLO-BSD模型结构

Fig. 1 The structure of YOLO-BSD model

图2 Conv结构

Fig. 2 The structure of Conv

图3 C2f-FasterRep和Faster Block结构

Fig. 3 The structure of C2f-FasterRep ((a) C2f-FasterRep; (b) Faster Block)

图4 SPPF结构

Fig. 4 The structure of SPPF

图5 CA模块结构

Fig. 5 The structure of CA module

图6 SFF模块结构

Fig. 6 The structure of SFF module

图7 解码器结构

Fig. 7 The structure of decoder

表1 实验环境配置

Table 1 Experimental environment configuration

配置	参数
CPU	Intel Core i7-7700K @4.20 GHz
GPU	NVIDIA GeForce RTX 4090
内存	24 GB
操作系统	Ubuntu 18.04
深度学习框架	PyTorch 2.0
编程语言	Python3.9

图8 数据采集装置

Fig. 8 Data collection device

图9 黑烟样本

Fig. 9 Black smoke sample

图10 数据增强的黑烟图片

Fig. 10 Data augmentation black smoke image

表2 不同检测模型的对比实验

Table 2 Comparison experiment of different detection networks

Model	Param/M	FLOPs/G	Precision/%	Recall/%	mAp(0.50)/%	mAp(0.50∶0.95)/%	FPS
Faster-RCNN	165.00	199.0	82.0	87.0	85.3	41.3	24
SSD	24.50	87.9	79.0	82.0	81.5	46.8	32
YOLOv5s	7.10	16.5	75.9	82.0	83.7	50.1	40
YOLOv6s	18.50	45.3	83.7	84.1	89.1	52.5	42
YOLOv8s	11.10	28.8	85.5	87.3	88.1	57.9	45
YOLO-BSD	9.27	22.2	94.5	97.5	95.4	69.7	57

图11 热力图检测结果示例((a)原始图像；(b)热力图检测结果)

Fig. 11 The example of heat map detection result ((a) Original image; (b) Heat map detection result)

表3 TT100K数据集对比实验

Table 3 Comparison experiments on the TT100K dataset

Model	Param/ M	Precision/ %	Recall/ %	mAp(0.50)/ %
YOLOv8s	11.10	82.4	78.6	81.4
YOLO-BSD	9.27	85.8	82.9	83.5

表4 自建数据集的消融实验

Table 4 Ablation experiment on self-collection dataset

Model	C2f-FasterRep	HLS-PAN	New Head	Param/M	mAp(0.50)/%
				11.10	88.10
YOLOv8s	√			10.20	93.44
		√		9.79	92.97
			√	13.10	91.33
YOLO-BSD	√	√		7.81	94.18
	√	√	√	9.27	95.40

表5 不同错误类型的比较/%

Table 5 Comparison of different error types/%

Model	Class	Location	Both	Duplicate	Bkgd	Missed
YOLOv8s	2.34	1.17	0.21	0.05	1.35	0.18
YOLO-BSD	1.10	0.32	0.07	0.01	0.87	0.02

参考文献 26

[1]	丁怡婷. 力争2030年前实现碳达峰, 2060年前实现碳中和——打赢低碳转型硬仗[R/OL]. (2021-04-02) [2024-04-21]. https://www.gov.cn/xinwen/2021-04/02/content_5597403.htm.
	DING Y T. Strive to achieve carbon peak before 2030 and carbon neutrality before 2060-win the tough battle of low-carbon transformation[R/OL]. (2021-04-02) [2024-04-21]. https://www.gov.cn/xinwen/2021-04/02/content_5597403.htm. (in Chinese).
[2]	中华人民共和国生态环境部, 中华人民共和国国家发展和改革委员会, 中华人民共和国科学技术部, 等. 关于印发《深入打好重污染天气消除、臭氧污染防治和柴油货车污染治理攻坚战行动方案》的通知[R/OL]. (2022-11-10) [2024-04-21]. https://www.gov.cn/zhengce/zhengceku/2022-11/17/content_5727605.htm.
	Ministry of Ecology and Environment of the People’s Republic of China, National Development and Reform Commission, Ministry of Science and Technology of the People’s Republic of China, et al. Notice on the issuance of the in-depth fight to eliminate heavy pollution weather, ozone pollution prevention and control and diesel truck pollution control action plan[R/OL]. (2022-11-10) [2024-04-21]. https://www.gov.cn/zhengce/zhengceku/2022-11/17/content_5727605.htm (in Chinese).
[3]	郭韬远, 任明武. 基于注意机制的双分支黑烟车辆检测网络[J]. 计算机与数字工程, 2022, 50(1): 147-151.
	GUO T Y, REN M W. Dual branch network for black smoke and vehicle detection based on attention mechanism[J]. Computer and Digital Engineering, 2022, 50(1): 147-151 (in Chinese).
[4]	TAO H J, LU X B. Contour-based smoky vehicle detection from surveillance video for alarm systems[J]. Signal, Image and Video Processing, 2019, 13(2): 217-225.
[5]	江东, 王晓龙, 陈政, 等. 基于深度学习算法的机动车尾气排放黑烟监测[J]. 现代电子技术, 2023, 46(17): 66-69.
	JIANG D, WANG X L, CHEN Z, et al. Monitoring of black smoke from motor vehicle exhaust based on deep learning algorithm[J]. Modern Electronics Technique, 2023, 46(17): 66-69 (in Chinese).
[6]	吴丙芳, 叶兵, 汪仕铭. 一种双流卷积神经网络的黑烟车检测算法[J]. 合肥工业大学学报(自然科学版), 2022, 45(2): 198-202.
	WU B F, YE B, WANG S M. Smoky vehicle detection based on two-stream convolutional neural networks[J]. Journal of Hefei University of Technology (Natural Science), 2022, 45(2): 198-202 (in Chinese).
[7]	杨哲, 李威. 基于ResNet34改进的黑烟检测算法[J]. 长江信息通信, 2023, 36(4): 88-90.
	YANG Z, LI W. Improved black smoke detection algorithm based on ResNet34[J]. Changjiang Information Communication, 2023, 36(4): 88-90 (in Chinese).
[8]	CAO Y C, LU X B. Learning spatial-temporal representation for smoke vehicle detection[J]. Multimedia Tools and Applications, 2019, 78(19): 27871-27889. DOI
[9]	TAO H J, LU X B. Automatic smoky vehicle detection from traffic surveillance video based on vehicle rear detection and multi‐feature fusion[J]. IET Intelligent Transport Systems, 2019, 13(2): 252-259.
[10]	TAO H J, LU X B. Smoke vehicle detection based on spatiotemporal bag-of-features and professional convolutional neural network[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2020, 30(10): 3301-3316.
[11]	井振威, 张团善. 基于GSS-YOLOv8n的轻量化织物疵点检测算法[J/OL]. (2024-07-18) [2024-05-22]. http://kns.cnki.net/kcms/detail/61.1132.ts.20240716.1140.002.html.
	JING Z W, ZHANG S T. Lightweight fabric defect detection algorithm based on GSS-YOLOv8n[J/OL]. (2024-07-18) [2024-05-22]. http://kns.cnki.net/kcms/detail/61.1132.ts.20240716.1140.002.html. (in Chinese).
[12]	叶大鹏, 景均, 张之得, 等. MSH-YOLOv8: 融合尺度重建的蘑菇小目标检测方法[J]. 智慧农业, 2024, 6(5): 139-152.
	YE D P, JING J, ZHANG Z D, et al. MSH-YOLOv8: mushroom small object detection method with scale reconstruction and fusion[J]. Smart Agriculture, 2024, 6(5): 139-152 (in Chinese). DOI
[13]	肖振久, 严肃, 曲海成. 基于多重机制优化YOLOv8的复杂环境下安全帽检测方法[J]. 计算机工程与应用, 2024, 60(21): 172-182. DOI
	XIAO Z J, YAN S, QU H C. Safety helmet detection method in complex environment based on multi-mechanism optimization of YOLOv8[J]. Computer Engineering and Applications, 2024, 60(21): 172-182 (in Chinese). DOI
[14]	LIU J F, YANG L J, CHENG H X, et al. Three-dimensional convolutional vehicle black smoke detection model with fused temporal features[J]. Applied Sciences, 2024, 14(18): 8173.
[15]	CHEN K, WANG H, ZHAI Y C. A lightweight model for real-time detection of vehicle black smoke[J]. Sensors, 2023, 23(23): 9492.
[16]	WANG H, CHEN K, LI Y F. Automatic detection method for black smoke vehicles considering motion shadows[J]. Sensors, 2023, 23(19): 8281.
[17]	WANG S P, HAN Y, YU M M, et al. Research on black smoke detection and class evaluation method for ships based on YOLOv5s-cmbi multi-feature fusion[J]. Journal of Marine Science and Engineering, 2023, 11(10): 1945.
[18]	李珍峰, 符世琛, 徐乐, 等. 基于MBI-YOLOv8的煤矸石目标检测算法研究[J]. 图学学报, 2024, 45(6): 1301-1312. DOI
	LI Z F, FU S C, XU L, et al. Research on gangue target detection algorithm based on MBI-YOLOv8[J]. Journal of Graphics, 2024, 45(6): 1301-1312 (in Chinese). DOI
[19]	胡凤阔, 叶兰, 谭显峰, 等. 一种基于改进YOLOv8的轻量化路面病害检测算法[J]. 图学学报, 2024, 45(5): 892-900. DOI
	HU F K, YE L, TAN X F, et al. A refined YOLOv8-based algorithm for lightweight pavement disease detection[J]. Journal of Graphics, 2024, 45(5): 892-900 (in Chinese). DOI
[20]	苑朝, 赵亚冬, 张耀, 等. 基于YOLO轻量化的多模态行人检测算法[J]. 图学学报, 2024, 45(1): 35-46. DOI
	YUAN C, ZHAO Y D, ZHANG Y, et al. Lightweight multi-modal pedestrian detection algorithm based on YOLO[J]. Journal of Graphics, 2024, 45(1): 35-46 (in Chinese). DOI
[21]	IOFFE S, SZEGEDY C. Batch normalization: accelerating deep network training by reducing internal covariate shift[EB/OL]. [2024-05-22]. https://dl.acm.org/doi/10.5555/3045118.3045167.
[22]	ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: faster and better learning for bounding box regression[C]// The 34th AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2020: 12993-13000.
[23]	LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]// 2017 IEEE International Conference on Computer Vision. New York: IEEE Press, 2017: 2999-3007.
[24]	HO Y, WOOKEY S. The real-world-weight cross-entropy loss function: modeling the costs of mislabeling[J]. IEEE Access, 2020, 8: 4806-4813.
[25]	BERGER M A. Random affine iterated function systems: mixing and encoding[M]// PINSKYM A, WIHSTUTZV. Diffusion Processes and Related Problems in Analysis, Volume II:Stochastic Flows. Boston: Birkhäuser, 1992: 315-346.
[26]	BUSLAEV A, IGLOVIKOV V I, KHVEDCHENYA E, et al. Albumentations: fast and flexible image augmentations[J]. Information, 2020, 11(2): 125.

面向柴油车辆排放黑烟的改进型YOLOv8检测算法研究

Research on improved YOLOv8 detection algorithm for diesel vehicle emission of black smoke

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 16

参考文献 26

相关文章 15

编辑推荐

Metrics

本文评价

[1]	翟永杰, 王璐瑶, 赵晓瑜, 胡哲东, 王乾铭, 王亚茹. 基于级联查询-位置关系的输电线路多金具检测方法[J]. 图学学报, 2025, 46(2): 288-299.
[2]	赵振兵, 韩钰, 唐辰康. 基于改进YOLOv8的配电线路绝缘子缺陷级联检测方法[J]. 图学学报, 2025, 46(1): 1-12.
[3]	程旭东, 史彩娟, 高炜翔, 王森, 段昌钰, 闫晓东. 面向域自适应目标检测的一致无偏教师模型[J]. 图学学报, 2025, 46(1): 114-125.
[4]	崔克彬, 耿佳昌. 基于EE-YOLOv8s的多场景火灾迹象检测算法[J]. 图学学报, 2025, 46(1): 13-27.
[5]	王志东, 陈晨阳, 刘晓明. 基于轻量化改进YOLOv8的通信光缆缺陷检测[J]. 图学学报, 2025, 46(1): 28-34.
[6]	陈冠豪, 徐丹, 贺康建, 施洪贞, 张浩. 基于转置注意力和CNN的图像超分辨率重建网络[J]. 图学学报, 2025, 46(1): 35-46.
[7]	张文祥, 王夏黎, 王欣仪, 杨宗宝. 一种强化伪造区域关注的深度伪造人脸检测方法[J]. 图学学报, 2025, 46(1): 47-58.
[8]	苑朝, 赵明雪, 张丰羿, 冯晓勇, 李冰, 陈瑞. 基于点云特征增强的复杂室内场景3D目标检测[J]. 图学学报, 2025, 46(1): 59-69.
[9]	王杨, 马唱, 胡明, 孙涛, 饶元, 袁振羽. 基于多尺度特征融合的轻量型野外蝙蝠检测[J]. 图学学报, 2025, 46(1): 70-80.
[10]	孙前来, 林绍杭, 刘东峰, 宋晓阳, 刘佳耀, 刘瑞珍. 基于元学习的小样本指针式仪表检测方法[J]. 图学学报, 2025, 46(1): 81-93.
[11]	卢洋, 陈林慧, 姜晓恒, 徐明亮. SDENet：基于多尺度注意力质量感知的合成缺陷数据评价网络[J]. 图学学报, 2025, 46(1): 94-103.
[12]	李琼, 考月英, 张莹, 徐沛. 面向无人机航拍图像的目标检测研究综述[J]. 图学学报, 2024, 45(6): 1145-1164.
[13]	李珍峰, 符世琛, 徐乐, 孟博, 张昕, 秦建军. 基于MBI-YOLOv8的煤矸石目标检测算法研究[J]. 图学学报, 2024, 45(6): 1301-1312.
[14]	闫建红, 冉同霄. 基于YOLOv8的轻量化无人机图像目标检测算法[J]. 图学学报, 2024, 45(6): 1328-1337.
[15]	胡凤阔, 叶兰, 谭显峰, 张钦展, 胡志新, 方清, 王磊, 满孝锋. 一种基于改进YOLOv8的轻量化路面病害检测算法[J]. 图学学报, 2024, 45(5): 892-900.