Small object detection for traffic signs based on improved YOLOv8 algorithm

doi:10.11772/j.issn.1001-9081.2024030360

Abstract

Abstract:

To address the issues of insufficient precision and high missed detection rate in existing object detection models when dealing with small objects of traffic signs， an improved object detection model based on YOLOv8 algorithm was proposed. Firstly， inspired by Residual Network （ResNet） design concept， residual connection mechanism was introduced in Backbone to integrate multi-layer feature information more efficiently by the model， so as to enhance the recognition ability of small objects. Secondly， a reversal of the Path Aggregation Feature Pyramid Network （PAFPN） structure in Neck was proposed as the I-PAFPN （Inverse PAFPN） structure， which enabled the network to capture the key features of the objects. Thirdly， the original three-level detection was extended to four-level detection， which enabled the model to focus on and extract more detailed features of small objects， so as to improve the sensitivity of the model to small objects. Finally， Wise Intersection over Union （WIoU） loss function was introduced to weaken the influence of low-quality samples on the model， so as to improve the model accuracy. Experimental results on TT100K （Tsinghua-Tencent 100K） dataset after data augmentation show that the improved YOLOv8 model has the mAP₅₀ and mAP_50：95 increased by 17.1 and 12.5 percentage points， respectively， compared to the original YOLOv8 model， verifying the effectiveness and superiority of the improved YOLOv8 model in the detection of small objects of traffic signs.

Key words: traffic sign detection, small object, YOLOv8, residual connection, Path Aggregation Feature Pyramid Network (PAFPN), Wise Intersection over Union (WIoU)

摘要：

为解决现有的目标检测模型在处理小目标交通标志时精度不足以及漏检率较高的问题，提出一种基于YOLOv8算法的改进型目标检测模型。首先，融合残差网络（ResNet）的设计理念，在Backbone中引入残差连接机制使模型更有效地整合多层特征信息，从而增强对小目标的识别能力；其次，逆转Neck部分的路径聚合特征金字塔网络（PAFPN）结构，提出I-PAFPN（Inverse PAFPN）结构，从而使网络更集中地捕捉目标的关键特征；再次，将原先的3级检测扩展为4级检测，使模型关注并更细致地提取小目标的特征，从而提高模型对小目标的敏感度；最后，引入WIoU（Wise Intersection over Union）损失函数弱化低质量样例对模型的影响，提高模型准确率。在数据增强后的TT100K（Tsinghua-Tencent 100K）数据集上的实验结果表明，经过改进的YOLOv8模型的mAP₅₀和mAP_50：95相较于原始的YOLOv8模型分别提高17.1和12.5个百分点，验证了改进YOLOv8模型在小目标交通标志检测方面的有效性和优越性。

关键词: 交通标志检测, 小目标, YOLOv8, 残差连接, 路径聚合特征金字塔网络, WIoU

CLC Number:

TP391.4

Bin WANG, Honghua XU, Doucheng SUN, Yongfan YU. Small object detection for traffic signs based on improved YOLOv8 algorithm[J]. Journal of Computer Applications, 0, (): 274-279.

王斌, 徐洪华, 孙兜成, 俞泳帆. 基于YOLOv8算法改进的小目标交通标志检测[J]. 《计算机应用》唯一官方网站, 0, (): 274-279.

Figures/Tables 12

数据集	交通标志图像数	交通标志实例数
训练集	6 773	16 754
验证集	1 954	4 880
测试集	1 011	2 578

参数	默认值	描述信息
epochs	300	训练轮次
batch	64	训练批次
optimizer	SGD	优化器
fliplr	0.5	图像数据左右旋转的概率
data	tt100k_mt100.yaml	训练所用的数据集配置文件
imgsz	640	输入图像尺寸

连接方式	P	R	mAP₅₀	mAP_50：95
无残差连接	76.8	63.4	73.1	55.8
直接残差连接	74.9	64.7	73.3	55.7
Sigmoid残差连接	74.0	65.4	72.7	54.7
1×1卷积残差连接	79.6	65.4	75.2	57.1

损失函数	P	R	mAP₅₀	mAP_50：95
CIoU	76.8	63.4	73.1	55.8
WIoU v1	79.7	60.2	72.0	54.5
WIoU v2	77.8	63.1	73.4	55.5
WIoU v3（ $α = 2.5$ ， $δ = 2$ ）	76.2	63.7	72.6	54.9
WIoU v3（ $α = 1.9$ ， $δ = 3$ ）	75.5	67.1	74.5	56.1
WIoU v3（ $α = 1.6$ ， $δ = 4$ ）	82.1	64.7	75.1	56.7
WIoU v3（ $α = 1.4$ ， $δ = 5$ ）	79.4	64.6	75.3	56.8
WIoU v3（ $α = 1.2$ ， $δ = 6$ ）	81.1	65.3	75.9	56.2
WIoU v3（ $α = 1.0$ ， $δ = 7$ ）	77.8	68.3	75.7	55.2

损失函数	P	R	mAP₅₀	mAP_50：95
CIoU	76.8	63.4	73.1	55.8
WIoU v1	79.7	60.2	72.0	54.5
WIoU v2	77.8	63.1	73.4	55.5
WIoU v3（ $α = 2.5$ ， $δ = 2$ ）	76.2	63.7	72.6	54.9
WIoU v3（ $α = 1.9$ ， $δ = 3$ ）	75.5	67.1	74.5	56.1
WIoU v3（ $α = 1.6$ ， $δ = 4$ ）	82.1	64.7	75.1	56.7
WIoU v3（ $α = 1.4$ ， $δ = 5$ ）	79.4	64.6	75.3	56.8
WIoU v3（ $α = 1.2$ ， $δ = 6$ ）	81.1	65.3	75.9	56.2
WIoU v3（ $α = 1.0$ ， $δ = 7$ ）	77.8	68.3	75.7	55.2

训练集增强	特征残差连接	增加检测层P2	I-PAFPN	WIoU v3	fliplr=0.0	params/10⁶	计算成本/GFLOPs	mAP₅₀/%	mAP_50：95/%
—	—	—	—	—	—	3.0	8.2	73.1	55.8
√	—	—	—	—	—	3.0	8.2	76.4	57.9
—	√	—	—	—	—	3.1	8.5	75.2	57.1
—	—	√	—	—	—	3.0	13.5	77.4	59.0
—	—	—	√	—	—	5.5	20.2	74.8	56.4
—	—	—	—	√	—	3.0	8.2	75.3	56.8
—	—	—	—	—	√	3.0	8.2	76.9	58.6
√	√	—	—	—	—	3.1	8.5	76.1	58.3
√	√	√	—	—	—	3.1	13.7	81.3	61.8
√	√	√	√	—	—	6.4	59.1	87.4	66.1
√	√	√	√	√	—	6.4	59.1	88.3	66.5
√	√	√	√	√	√	6.4	59.1	90.2	68.3

模型	params/10⁶	计算成本/GFLOPs	mAP₅₀/%	mAP_50：95/%
YOLOv8s	11.2	28.7	83.6	64.2
YOLOv8l	25.9	79.2	86.1	67.0
本文模型	6.4	59.1	90.2	68.3

模型	params/10⁶	计算成本/GFLOPs	推理时间/ms	帧率/（frame·s^-1）	P/%	R/%	mAP₅₀/%	mAP_50：95/%
YOLOv5n	1.80	4.40	6.2	161	69.8	44.4	50.0	36.1
YOLOv6n	4.70	11.40	8.0	125	50.2	72.9	65.9	50.2
YOLOv7-tiny	6.10	13.60	4.6	218	56.4	47.3	43.7	31.5
YOLOv8n	3.00	8.20	6.8	147	76.8	63.4	73.1	55.8
GOLD-YOLOv8n	5.00	11.20	18.1	55	71.4	60.7	68.9	52.8
DAMO-YOLO-T	8.34	17.67	3.4	294	—	—	64.1	47.2
YOLOv9-C	51.10	239.40	20.0	50	88.2	80.3	88.7	69.1
GELAN-C	25.40	103.40	14.5	69	89.7	78.8	88.4	69.0
本文模型	6.40	59.10	8.4	119	90.5	83.2	90.2	68.3

References 23

1	REN S， HE K， 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.
2	REDMON J， DIVVALA S， GIRSHICK R， et al. You only look once： unified， real-time object detection ［C］// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2016： 779-788.
3	LIU W， ANGUELOV D， ERHAN D， et al. SSD： single shot MultiBox detector ［C］// Proceedings of the 2016 European Conference on Computer Vision， LNCS 9905. Cham： Springer， 2016： 21-37.
4	LAI H， CHEN L， LIU W， et al. STC-YOLO： small object detection network for traffic signs in complex environments ［J］. Sensors， 2023， 23（11）： No.5307.
5	ZHU Z， LIANG D， ZHANG S， et al. Traffic-sign detection and classification in the wild ［C］// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2016： 2110-2118.
6	LI S， WANG S， WANG P. A small object detection algorithm for traffic signs based on improved YOLOv7 ［J］. Sensors， 2023， 23（16）： No.7145.
7	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［C］// Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2023： 7464-7475.
8	胡均平，王鸿树，戴小标，等. 改进YOLOv5的小目标交通标志实时检测算法［J］. 计算机工程与应用， 2023， 59（2）：185-193.
9	嵇文，刘全金，黄崇文，等. 基于Swin-Transformer的YOLOX交通标志检测［J］. 无线电通信技术， 2023， 49（3）：547-555.
10	LIU Z， LIN Y， CAO Y， et al. Swin Transformer： hierarchical vision Transformer using shifted windows ［C］// Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision. Piscataway： IEEE， 2021： 9992-10002.
11	GE Z， LIU S， WANG F， et al. YOLOX： exceeding YOLO series in 2021 ［EB/OL］. ［2023-11-12］. .
12	LIU S， QI L， QIN H， et al. Path aggregation network for instance segmentation ［C］// Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2018： 8759-8768.
13	HE K， ZHANG X， REN S， et al. Deep residual learning for image recognition ［C］// Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2016： 770-778.
14	FENG C， ZHONG Y， GAO Y， et al. TOOD： task-aligned one-stage object detection ［C］// Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision. Piscataway： IEEE， 2021： 3490-3499.
15	LI X， WANG W， WU L， et al. Generalized focal loss： learning qualified and distributed bounding boxes for dense object detection［C］// Proceedings of the 34th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2020： 21002-21012.
16	LIN T Y， DOLLÁR P， GIRSHICK R， et al. Feature pyramid networks for object detection ［C］// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2017： 936-944.
17	TONG Z， CHEN Y， XU Z， et al. Wise-IoU： bounding box regression loss with dynamic focusing mechanism ［EB/OL］. ［2023-11-07］..
18	LIN T Y， GOYAL P， GIRSHICK R， et al. Focal loss for dense object detection ［C］// Proceedings of the 2017 IEEE International Conference on Computer Vision. Piscataway： IEEE， 2017： 2999-3007.
19	YUN S， HAN D， CHUN S， et al. CutMix： regularization strategy to train strong classifiers with localizable features ［C］// Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision. Piscataway： IEEE， 2019： 6022-6031.
20	LI C， LI L， JIANG H， et al. YOLOv6： a single-stage object detection framework for industrial applications ［EB/OL］. ［2023-11-25］. .
21	WANG C， HE W， NIE Y， et al. Gold-YOLO： efficient object detector via gather-and-distribute mechanism ［C］// Proceedings of the 37th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2023： 51094-51112.
22	XU X， JIANG Y， CHEN W， et al. DAMO-YOLO： a report on real-time object detection design ［EB/OL］. ［2023-12-12］. .
23	WANG C Y， YEH I H， LIAO H Y M. YOLOv9： learning what you want to learn using programmable gradient information ［EB/OL］. ［2023-12-14］. .

[1]	Quan WANG, Xinyu CAO, Qidong CHEN. Roadside traffic object detection model and deployment for vehicle-road collaboration [J]. Journal of Computer Applications, 2025, 45(3): 1016-1024.
[2]	Qiurun HE, Jie HU, Bo PENG, Tianyuan LI. Fabric defect detection algorithm based on context information and multi-scale feature fusion [J]. Journal of Computer Applications, 2025, 45(2): 640-646.
[3]	Songsen YU, Zhifan LIN, Guopeng XUE, Jianyu XU. Lightweight large-format tile defect detection algorithm based on improved YOLOv8 [J]. Journal of Computer Applications, 2025, 45(2): 647-654.
[4]	Jiayang GUI, Shunji WANG, Zhengkang ZHOU, Jiashan TANG. Tunnel foreign object detection algorithm based on improved YOLOv8n [J]. Journal of Computer Applications, 2025, 45(2): 655-661.
[5]	Boran YANG, Suzhen LIN, Dawei LI, Xiaofei LU, Chenhui CUI. Infrared small target detection method based on information compensation [J]. Journal of Computer Applications, 2025, 45(1): 284-291.
[6]	Shang LIU, Yuwei ZHOU, Rao DAI, Linfang DONG, Meng LIU. Small target detection algorithm in remote sensing images integrating attention and contextual information [J]. Journal of Computer Applications, 2025, 45(1): 292-300.
[7]	Yexin PAN, Zhe YANG. Optimization model for small object detection based on multi-level feature bidirectional fusion [J]. Journal of Computer Applications, 2024, 44(9): 2871-2877.
[8]	Yeheng LI, Guangsheng LUO, Qianmin SU. Logo detection algorithm based on improved YOLOv5 [J]. Journal of Computer Applications, 2024, 44(8): 2580-2587.
[9]	Wudan LONG, Bo PENG, Jie HU, Ying SHEN, Danni DING. Road damage detection algorithm based on enhanced feature extraction [J]. Journal of Computer Applications, 2024, 44(7): 2264-2270.
[10]	Zhangjian JI, Na DU. Tiny target detection based on improved VariFocalNet [J]. Journal of Computer Applications, 2024, 44(7): 2200-2207.
[11]	Yue LIU, Fang LIU, Aoyun WU, Qiuyue CHAI, Tianxiao WANG. 3D object detection network based on self-attention mechanism and graph convolution [J]. Journal of Computer Applications, 2024, 44(6): 1972-1977.
[12]	Huantong GENG, Zhenyu LIU, Jun JIANG, Zichen FAN, Jiaxing LI. Embedded road crack detection algorithm based on improved YOLOv8 [J]. Journal of Computer Applications, 2024, 44(5): 1613-1618.
[13]	Chenhui CUI, Suzhen LIN, Dawei LI, Xiaofei LU, Jie WU. Infrared dim small target tracking method based on Siamese network and Transformer [J]. Journal of Computer Applications, 2024, 44(2): 563-571.
[14]	Yudong PANG, Zhixing LI, Weijie LIU, Tianhao LI, Ningning WANG. Small target detection model in overlooking scenes on tower cranes based on improved real-time detection Transformer [J]. Journal of Computer Applications, 2024, 44(12): 3922-3929.
[15]	Dahai LI, Bingtao LI, Zhendong WANG. Underwater target detection algorithm based on improved YOLOv8 [J]. Journal of Computer Applications, 2024, 44(11): 3610-3616.