基于轻量化YOLOv8s交通标志的检测

doi:10.11996/JG.j.2095-302X.2024030422

图学学报 ›› 2024, Vol. 45 ›› Issue (3): 422-432.DOI: 10.11996/JG.j.2095-302X.2024030422

• 图像处理与计算机视觉 • 上一篇下一篇

基于轻量化YOLOv8s交通标志的检测

朱强军¹(), 胡斌¹, 汪慧兰², 王杨³()

1.安徽师范大学皖江学院大数据与人工智能系，安徽芜湖 241000
2.安徽师范大学物理与电子信息学院，安徽芜湖 241000
3.安徽师范大学计算机与信息学院，安徽芜湖 241000

收稿日期:2023-10-17 接受日期:2024-01-09 出版日期:2024-06-30 发布日期:2024-06-06
通讯作者:王杨(1972-)，男，教授，博士。主要研究方向为人工智能系统和机器视觉。E-mail：wycap@126.com
第一作者:朱强军(1984-)，男，副教授，硕士。主要研究方向为智能算法、模式识别和数学建模。E-mail：zhuqiangjun@ahnu.edu.cn
基金资助:
安徽省高校自然科学研究重点项目(2023AH052459);安徽师范大学皖江学院重点自然科研项目(WJKYZD-202301);安徽省高等学校省级质量工程项目(2022sx052);安徽师范大学皖江学院教学质量工程项目(WJXGK-202201)

Detection of traffic signs based on lightweight YOLOv8s

ZHU Qiangjun¹(), HU Bin¹, WANG Huilan², WANG Yang³()

1. Department of Big Data and Artificial Intelligence, Wanjiang College of Anhui Normal University, Wuhu Anhui 241000, China
2. School of Physics and Electronic Information, Anhui Normal University, Wuhu Anhui 241000, China
3. School of Computer and Information, Anhui Normal University, Wuhu Anhui 241000, China

Received:2023-10-17 Accepted:2024-01-09 Published:2024-06-30 Online:2024-06-06
First author：ZHU Qiangjun (1984-), associate professor, master. His main research interests cover intelligent algorithm, pattern recognition and mathematical modeling. E-mail：zhuqiangjun@ahnu.edu.cn
Supported by:
Key Natural Science Research Project of Universities in Anhui Province(2023AH052459);Key Natural Science Research Project of Wanjiang College of Anhui Normal University(WJKYZD-202301);Provincial Quality Engineering Project of Higher Education Institutions in Anhui Province(2022sx052);Teaching Quality Project of Wanjiang College of Anhui Normal University(WJXGK-202201)

摘要/Abstract

摘要：

为了提高交通标志检测的实时性和可行性，提出了一种基于YOLOv8s的轻量化交通标志检测模型。首先，用FasterNet中的残差模块FasterNetBlock替换C2f模块中的BottleNeck，降低模型参数量和计算量；其次，用一种小目标检测层去替换大目标检测层，降低Backbone中网络层数，实现大幅度提高检测速度和降低参数量；最后，用Wise-IOU替换原CIOU损失函数，提高速度和精度。在TT100K交通标志数据集上验证，其与YOLOv8s模型比较，mAP50提高了5.16%，参数量降低了76.48%，计算量降低了13.33%，FPS快了35.83%。与其他模型相比，mAP50平均提高了15.11%，参数量平均降低了85.74%，计算量平均下降了46.23%，FPS平均提高了31.49%。该模型具有检测精度高、参数量少、计算量低、速度快等优点，较原算法有很大地提升，且与其他先进的交通标志检测模型比较时表现出了很强的竞争力，在交通标志检测中具有较大优势。

关键词: 轻量化, YOLOv8s, 改进小目标层, 交通标志检测, Wise-IOU, TT100K

Abstract:

To enhance the real-time capability and feasibility of traffic sign detection, a lightweight traffic sign detection model based on YOLOv8s was proposed. Firstly, the BottleNeck in the C2f module was replaced with the residual module FasterNetBlock in FasterNet, reducing the model’s parameter count and computational complexity. Secondly, the large object detection layer was replaced with a small object detection layer, decreasing the number of network layers in Backbone and achieving a significant improvement in detection speed and a reduction in parameter count. Finally, the original complete intersection over union (CIOU) loss function was replaced with the wise intersection over union (Wise-IOU), thereby enhancing both speed and accuracy. Verified on the TT100K traffic sign dataset, compared with the YOLOv8s model, mAP50 increased by 5.16%, parameter count decreased by 76.48%, computational complexity decreased by 13.33%, and frames per second (FPS) improved by 35.83%. In comparison to other models, mAP50 exhibited an average increase of 15.11%, an average decrease of 85.74% in parameter count, an average decrease of 46.23% in computational complexity, and an average increase of 31.49% in FPS. This model achieved the advantages of high detection accuracy, small number of parameters, low computational complexity, and fast speed. It represented a substantial improvement over the original algorithm and demonstrated strong competitiveness when compared to other advanced traffic sign detection models, with great advantages in traffic sign detection.

Key words: lightweight, YOLOv8s, improved small target layer, traffic sign detection, Wise-IOU, TT100K

中图分类号:

朱强军, 胡斌, 汪慧兰, 王杨. 基于轻量化YOLOv8s交通标志的检测[J]. 图学学报, 2024, 45(3): 422-432.

ZHU Qiangjun, HU Bin, WANG Huilan, WANG Yang. Detection of traffic signs based on lightweight YOLOv8s[J]. Journal of Graphics, 2024, 45(3): 422-432.

图/表 17

参考文献 34

[1]	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// 2014 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2014: 580-587.
[2]	REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. DOI PMID
[3]	潘卫国, 刘博, 陈英昊, 等. 基于YOLO v3的交通标志牌检测识别[J]. 传感器与微系统, 2019, 38(11): 147-150.
	PAN W G, LIU B, CHEN Y H, et al. Traffic sign detection and recognition based on YOLO v3[J]. Transducer and Microsystem Technologies, 2019, 38(11): 147-150 (in Chinese).
[4]	周苏, 支雪磊, 刘懂, 等. 基于卷积神经网络的小目标交通标志检测算法[J]. 同济大学学报: 自然科学版, 2019, 47(11): 1626-1632.
	ZHOU S, ZHI X L, LIU D, et al. A convolutional neural network-based method for small traffic sign detection[J]. Journal of Tongji University: Natural Science, 2019, 47(11): 1626-1632 (in Chinese).
[5]	吕禾丰, 陆华才. 基于YOLOv5算法的交通标志识别技术研究[J]. 电子测量与仪器学报, 2021, 35(10): 137-144.
	LV H F, LU H C. Research on traffic sign recognition technology based on YOLOv5 algorithm[J]. Journal of Electronic Measurement and Instrumentation, 2021, 35(10): 137-144 (in Chinese).
[6]	韦强, 胡晓阳, 赵虹鑫. 改进YOLOv5的交通标志检测方法[J]. 计算机工程与应用, 2023, 59(13): 229-237. DOI
	WEI Q, HU X Y, ZHAO H X. Improved traffic sign detection method for YOLOv5[J]. Computer Engineering and Applications, 2023, 59(13): 229-237 (in Chinese). DOI
[7]	袁红春, 陶磊. 基于改进的Yolov8商业渔船电子监控数据中鱼类的检测与识别[J]. 大连海洋大学学报, 2023, 38(3): 533-542.
	YUAN H C, TAO L. Detection and identification of fish in electronic monitoring data of commercial fishing vessels based on improved Yolov8[J]. Journal of Dalian Ocean University, 2023, 38(3): 533-542 (in Chinese).
[8]	魏陈浩, 杨睿, 刘振丙, 等. 具有双层路由注意力的YOLOv8道路场景目标检测方法[J]. 图学学报, 2023, 44(6): 1104-1111. DOI
	WEI C H, YANG R, LIU Z B, et al. YOLOv8 with bi-level routing attention for road scene object detection[J]. Journal of Graphics, 2023, 44(6): 1104-1111 (in Chinese).
[9]	熊恩杰, 张荣芬, 刘宇红, 等. 面向交通标志的Ghost- YOLOv8检测算法[J]. 计算机工程与应用, 2023, 59(20): 200-207. DOI
	XIONG E J, ZHANG R F, LIU Y H, et al. Ghost-YOLOv8 detection algorithm for traffic signs[J]. Computer Engineering and Applications, 2023, 59(20): 200-207 (in Chinese). DOI
[10]	高昂, 梁兴柱, 夏晨星, 等. 一种改进YOLOv8的密集行人检测算法[J]. 图学学报, 2023, 44(5): 890-898. DOI
	GAO A, LIANG X Z, XIA C X, et al. A dense pedestrian detection algorithm with improved YOLOv8[J]. Journal of Graphics, 2023, 44(5): 890-898 (in Chinese). DOI
[11]	李晓佩, 张寅宝, 李严培, 等. 一种基于YOLOv5s的红外图像目标检测改进算法[J]. 激光与红外, 2023, 53(7): 1043-1051.
	LI X P, ZHANG Y B, LI Y P, et al. An improved method of infrared image target detection based on YOLOv5s[J]. Laser & Infrared, 2023, 53(7): 1043-1051 (in Chinese).
[12]	赵璐璐, 王学营, 张翼, 等. 基于YOLOv5s融合SENet的车辆目标检测技术研究[J]. 图学学报, 2022, 43(5): 776-782.
	ZHAO L L, WANG X Y, ZHANG Y, et al. Vehicle target detection based on YOLOv5s fusion SENet[J]. Journal of Graphics, 2022, 43(5): 776-782 (in Chinese).
[13]	WU W T, LIU H, LI L L, et al. Application of local fully convolutional neural network combined with YOLO v5 algorithm in small target detection of remote sensing image[J]. PLoS One, 2021, 16(10): e0259283.
[14]	GE Z, LIU S T, WANG F, et al. YOLOX: exceeding YOLO series in 2021[EB/OL]. (2021-08-06) [2023-04-09]. http://arxiv.org/abs/2107.08430.pdf.
[15]	WANG C Y, MARK LIAO H Y, WU Y H, et al. CSPNet: a new backbone that can enhance learning capability of CNN[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. New York: IEEE Press, 2020: 1571-1580.
[16]	IOFFE S, SZEGEDY C. Batch normalization: accelerating deep network training by reducing internal covariate shift[EB/OL]. (2015-05-02) [2023-04-12]. http://arxiv.org/abs/1502.03167.pdf.
[17]	MAAS A L, HANNUN A Y, NG A Y. Rectifier nonlinearities improve neural network acoustic models[EB/OL]. (2013-05-06) [2023-04-22]. http://ai.stanford.edu/-amaas/papers/relu_hybrid_icml2013_final.pdf.
[18]	GLOROT X, BORDES A, BENGIO Y. Deep sparse rectifier neural networks[EB/OL]. [2023-04-12]. https://www.docin.com/p-830004906.html.
[19]	ELFWING S, UCHIBE E, DOYA K. Sigmoid-weighted linear units for neural network function approximation in reinforcement learning[J]. Neural Networks, 2018, 107: 3-11. DOI PMID
[20]	HE K M, ZHANG X Y, REN S Q, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9): 1904-1916. DOI PMID
[21]	LIU S, QI L, QIN H F, et al. Path aggregation network for instance segmentation[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 8759-8768.
[22]	FENG C J, ZHONG Y J, GAO Y, et al. TOOD: task-aligned one-stage object detection[C]// 2021 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2021: 3490-3499.
[23]	ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: faster and better learning for bounding box regression[EB/OL]. (2019-11-06) [2023-04-12]. http://arxiv.org/abs/1911.08287.pdf.
[24]	LI X, WANG W H, WU L J, et al. Generalized focal loss: Learning qualified and distributed bounding boxes for dense object detection[EB/OL]. (2020-06-08) [2023-04-02]. https://www.arxiv.org/abs/2006.04388v1.pdf.
[25]	CHEN J R, KAO S H, HE H, et al. Run, don’t walk: chasing higher FLOPS for faster neural networks[C]// 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2023: 12021-12031.
[26]	TONG Z J, CHEN Y H, XU Z W, et al. Wise-IoU: bounding box regression loss with dynamic focusing mechanism[EB/OL]. (2023-08-08) [2023-04-12]. http://arxiv.org/abs/2301.10051.pdf.
[27]	REZATOFIGHI H, TSOI N, GWAK J, et al. Generalized intersection over union: a metric and a loss for bounding box regression[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2019: 658-666.
[28]	GEVORGYAN Z. SIoU loss: more powerful learning for bounding box regression[EB/OL]. (2022-05-25) [2023-04-12]. http://arxiv.org/abs/2205.12740.pdf.
[29]	MA S L, XU Y. MPDIoU: a loss for efficient and accurate bounding box regression[EB/OL]. (2023-07-14) [2023-03-12]. http://arxiv.org/abs/2307.07662.pdf.
[30]	LI C Y, LI L L, JIANG H L, et al. YOLOv6: a single-stage object detection framework for industrial applications[EB/OL]. (2022-09-07) [2023-04-12]. http://arxiv.org/abs/2209.02976.pdf.
[31]	WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[EB/OL]. (2022-07-06) [2023-03-12]. http://arxiv.org/abs/2207.02696.pdf.
[32]	TANG S Y, ZHANG S, FANG Y N. HIC-YOLOv5: improved YOLOv5 for small object detection[EB/OL]. (2023-09-28) [2023-10-01]. http://arxiv.org/abs/2309.16393.pdf.
[33]	赵宏, 冯宇博. 基于CGS-Ghost YOLO的交通标志检测研究[J]. 计算机工程, 2023, 49(12): 194-204. DOI
	ZHAO H, FENG Y B. Research on traffic sign detection based on CGS-Ghost YOLO[J]. Computer Engineering, 2023, 49(12): 194-204 (in Chinese). DOI
[34]	HONG H Q, ZHOU Y, SHU X Y, et al. CCSPNet-Joint: efficient joint training method for traffic sign detection under extreme conditions[EB/OL]. (2023-09-25) [2023-10-01]. http://arxiv.org/abs/2309.06902.pdf.

样本真实情况	预测结果
样本真实情况	正样本	负样本
正样本	TP	FN
负样本	FP	TN

样本真实情况	预测结果
样本真实情况	正样本	负样本
正样本	TP	FN
负样本	FP	TN

模型	P/%	R/%	mAP50/%	参数量/M	GFLOPs	FPS
YOLOv8s	85.34	78.39	87.18	11.14	28.5	32.15
YOLOv8s+小目标检测层	88.02	85.46	92.18	10.86	37.9	26.67
YOLOv8s+改进小目标检测层	87.39	84.71	91.71	3.53	31.0	37.88

模型	P/%	R/%	mAP50/%	参数量/M	GFLOPs	FPS
YOLOv8s	85.34	78.39	87.18	11.14	28.5	32.15
YOLOv8s+小目标检测层	88.02	85.46	92.18	10.86	37.9	26.67
YOLOv8s+改进小目标检测层	87.39	84.71	91.71	3.53	31.0	37.88

模型	小目标		中等目标		大目标
模型	AP/%	AR/%	AP/%	AR/%	AP/%	AR/%
YOLOv8s	47.20	62.24	73.36	82.27	85.23	88.55
YOLOv8s+小目标检测层	58.62	71.27	75.88	82.74	82.89	89.41
YOLOv8s+改进小目标检测层	55.19	77.18	75.30	68.71	84.77	89.02

基于轻量化YOLOv8s交通标志的检测

Detection of traffic signs based on lightweight YOLOv8s

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 17

参考文献 34

相关文章 15

编辑推荐

Metrics

本文评价

模型	P/%	R/%	mAP50/%	FPS
YOLOv8s+GIOU	86.50	77.09	86.15	32.79
YOLOv8s+CIOU	85.34	78.39	87.18	32.15
YOLOv8s+DIOU	87.04	78.5	87.59	30.40
YOLOv8s+SIOU	85.77	78.95	87.31	26.60
YOLOv8s+MPDIOU	85.80	78.02	86.96	33.00
YOLOv8s+Wise-IOU	89.61	77.16	87.70	33.22

模型	mAP50/%	参数量/M	GFLOPs	FPS
YOLOv6s^[30]	84.50	18.51	45.21	30.30
YOLOv7^[31]	56.33	36.62	103.60	38.91
YOLOv8s	87.18	11.14	28.50	32.15
HIC-YOLOv5^[32]	91.62	9.38	31.40	30.12
CGS-Ghost-YOLO^[33]	66.04	6.25	17.00	27.25
CCSPNet-Joint^[34]	92.59	36.81	65.80	42.02
Ghost-YOLOv8^[9]	86.78	2.69	12.60	30.67
Ours	91.68	2.62	24.70	43.67

评价指标	CCTSDB		RSOD-aircraft		TT100K
评价指标	YOLOv8s	本文模型	YOLOv8s	本文模型	YOLOv8s	本文模型
P/%	81.51	87.56	95.13	98.00	85.34	88.43
R/%	53.72	73.46	84.54	84.67	78.39	82.98
mAP50/%	61.92	80.56	91.74	94.02	87.18	91.68
mAP50~95/%	32.81	50.30	62.66	62.66	66.44	69.57
参数量/M	11.14	2.62	11.14	2.62	11.14	2.62
GFLOPs	28.50	24.70	28.50	24.70	28.50	24.70
FPS	50.00	60.61	113.64	172.41	32.15	43.67

[1]	曾志超, 徐玥, 王景玉, 叶元龙, 黄志开, 王欢. 基于SOE-YOLO轻量化的水面目标检测算法[J]. 图学学报, 2024, 45(4): 736-744.
[2]	赵磊, 李栋, 房建东, 曹琪. 面向交通标志的改进YOLO目标检测算法[J]. 图学学报, 2024, 45(4): 779-790.
[3]	苑朝, 赵亚冬, 张耀, 王嘉璇, 徐大伟, 翟永杰, 朱松松. 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.
[4]	翟永杰, 赵晓瑜, 王璐瑶, 王亚茹, 宋晓轲, 朱浩硕. IDD-YOLOv7：一种用于输电线路绝缘子多缺陷的轻量化检测方法[J]. 图学学报, 2024, 45(1): 90-101.
[5]	皮骏, 牛厚兴, 高志云. 融合CA-BiFPN的轻量化人体姿态估计算法[J]. 图学学报, 2023, 44(5): 868-878.
[6]	熊举举, 徐杨, 范润泽, 孙少聪. 基于轻量化视觉Transformer的花卉识别[J]. 图学学报, 2023, 44(2): 271-279.
[7]	皮骏, 刘宇恒, 李久昊. 基于YOLOv5s的轻量化森林火灾检测算法研究[J]. 图学学报, 2023, 44(1): 26-32.
[8]	张运波, 易鹏飞, 周东生, 张强, 魏小鹏. 深度可分离卷积和标准卷积相结合的高效行人检测器[J]. 图学学报, 2022, 43(2): 230-238.
[9]	张芳兰, 刘龙吉, 姚宛彤. 面向关键用户需求的踝足矫形器定制化设计方法[J]. 图学学报, 2021, 42(5): 841-848.
[10]	肖文磊, 王志明, 王世平, 赵罡. 基于周期性晶格的 3D 打印模型轻量化方法[J]. 图学学报, 2021, 42(2): 263-270.
[11]	董健 1，董玉德 1，刘福 1，王帅 1，时晓蕾 2，陶高周 2 . 基于复杂装配体特征抑制的轻量化方法[J]. 图学学报, 2019, 40(5): 968-975.
[12]	郭思怡，陈永锋. 建筑运维阶段信息模型的轻量化方法[J]. 图学学报, 2018, 39(1): 123-128.
[13]	王晓斌，宁涛，王可. 3DXML文件格式解析及应用[J]. 图学学报, 2010, 31(2): 33-37.
[14]	谢峰，沈维蕾，林巨广. C型压力机机身的轻量化设计[J]. 图学学报, 2010, 31(1): 13-19.
[15]	赵韩，陈兴玉，董玉德，张坚，余来宏. 参数化有限元分析的液压机机身轻量化设计[J]. 图学学报, 2010, 31(1): 20-25.

模型	C2faster	改进小目标层	Wise-IOU	mAP50/%	参数量/M	GFLOPs	FPS
YOLOv8s	无	无	无	87.18	11.14	28.5	32.15
模型2	有	无	无	86.54	8.32	21.4	31.75
模型3	无	有	无	91.71	3.53	31.0	37.88
模型4	无	无	有	87.70	11.14	28.5	33.22
模型5	有	有	无	91.04	2.62	24.7	43.48
模型6	无	有	有	92.01	3.53	31.0	29.5
模型7	有	无	有	85.67	8.32	21.4	39.68
本文	有	有	有	91.68	2.62	24.7	43.67

模型	C2faster	改进小目标层	Wise-IOU	mAP50/%	参数量/M	GFLOPs	FPS
YOLOv8s	无	无	无	87.18	11.14	28.5	32.15
模型2	有	无	无	86.54	8.32	21.4	31.75
模型3	无	有	无	91.71	3.53	31.0	37.88
模型4	无	无	有	87.70	11.14	28.5	33.22
模型5	有	有	无	91.04	2.62	24.7	43.48
模型6	无	有	有	92.01	3.53	31.0	29.5
模型7	有	无	有	85.67	8.32	21.4	39.68
本文	有	有	有	91.68	2.62	24.7	43.67