CGT-YOLO-Based Algorithm for Small-Target Traffic Sign Recognition

doi:10.12141/j.issn.1000-565X.250092

Abstract

Abstract:

To address the degradation in recognition accuracy caused by false and missed detections of small target traffic signs, this study proposes a small traffic sign recognition algorithm based on CGT-YOLO. First, a context-aware enhancement module (CAM) is introduced to replace the spatial pyramid pooling fast (SPPF) module in the YOLOv5s network. By employing parallel dilated convolutions with different dilation rates, the CAM enhances multiscale feature representation and contextual information of small traffic signs without reducing spatial resolution. Second, a global attention mechanism (GAM) is inserted after the concatenation operation in the backbone network of YOLOv5s. The GAM extracts features enhanced by the CAM and strengthens global interaction between channel and spatial dimensions through 3D permutation, multi-layer perceptron, and convolutional spatial attention, thereby highlighting the features of small traffic signs and mitigating the negative effects of complex backgrounds and long distances. Finally, a task-specific context (TSC) decoupled head is utilized to separate features for classification and localization tasks. Through the semantic context encoder (SCE) and detail preservation encoder (DPE) modules, the head generates semantically rich low-resolution feature maps for classification and high-resolution feature maps containing boundary information for localization, respectively. This disentangles classification and localization tasks at the feature source, resolving feature conflicts between the two tasks for small target traffic signs. Experimental results on a dataset constructed by integrating TT100K and CCTSDB show that the improved model achieves enhanced performance across all metrics: the missed detection rate and false detection rate are reduced by 12.1 and 11.6 percentage points, respectively, while mAP(0.50∶0.95) increases by 0.026 0. Compared to models such as YOLOv8s, NanoDet-Plus, and RT-DETR-Nano, CGT-YOLO demonstrates superior performance across multiple metrics. While maintaining a high inference speed (72.5 FPS), it effectively reduces false and missed detections, significantly improving the detection accuracy and robustness of small target traffic signs in complex scenarios.

Key words: small target recognition, traffic sign recognition, dilated convolution, attention mechanism, decoupled head

CLC Number:

TP391.41

XING Yan, GUO Sihao, ZHANG Zhen, PAN Xiaodong, AN Dong. CGT-YOLO-Based Algorithm for Small-Target Traffic Sign Recognition[J]. Journal of South China University of Technology(Natural Science Edition), 2026, 54(3): 65-78.

Figures/Tables 18

Fig.1

Fig.2

Fig.3

Fig.4

Fig.5

Fig.6

Fig.7

Fig.8

Table 1

Table 2

Table 3

Fig.9

Table 4

Table 5

Table 6

Table 7

Fig.10

Table 8

References 26

[1]	NANDI D， SAIF A F M S， PAUL P，et al ．Traffic sign detection based on color segmentation of obscure image candidates：a comprehensive study［J］．International Journal of Modern Education and Computer Science，2018，10（6）：35-46.
[2]	GÓMEZ-MOEENO H， MALDONADO-BASCÓN S， GIL-JIMÉNEZ P，et al ．Goal evaluation of segmentation algorithms for traffic sign recognition［J］．IEEE Transactions on Intelligent Transportation Systems，2010，11（4）：917-930.
[3]	WALI S B， HANNAN M A， HUSSAIN A，et al ．An automatic traffic sign detection and recognition system based on colour segmentation，shape matching，and SVM［J］．Mathematical Problems in Engineering，2015，DOI：10.1155/2015/250461 .
[4]	TABERNIK D， SKOČAJ D ．Deep learning for large-scale traffic-sign detection and recognition［J］．IEEE Transactions on Intelligent Transportation Systems，2019，21（4）：1427-1440.
[5]	AYACHI R， AFIF M， SAID Y，et al ．Traffic signs detection for real-world application of an advanced driving assisting system using deep learning［J］．Neural Processing Letters，2020，51（1）：837-851.
[6]	AHMED S， KAMAL U， HASAN M K ．DFR-TSD：a deep learning based framework for robust traffic sign detection under challenging weather conditions［J］．IEEE Transactions on Intelligent Transportation Systems，2021，23（6）：5150-5162.
[7]	马鸽，李洪伟，严梓维，等．基于多注意力的改进YOLOv5s小目标检测算法［J］．工程科学学报，2024，46（9）：1647-1658.
	MA Ge， LI Hongwei， YAN Ziwei，et al ．Improved small target detection algorithm based on multiattention and YOLOv5s for traffic sign recognition［J］．Chinese Journal of Engineering，2024，46（9）：1647-1658.
[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.
	HU Junping， WANG Hongshu， DAI Xiaobiao，et al ．Real-time detection algorithm for small-target traffic signs based on improved YOLOv5［J］．Computer Engineering and Applications，2023，59（2）：185-193.
[9]	罗玉涛，高强．基于通道注意力和特征增强的交通标志检测［J］．华南理工大学学报（自然科学版），2023，51（12）：64-72.
	LUO Yutao， GAO Qiang ．Traffic sign detection based on channel attention and feature enhancement［J］．Journal of South China University of Technology （Natural Science Edition），2023，51（12）：64-72.
[10]	郭君斌，于琳，于传强．改进YOLOv5s算法在交通标志检测识别中的应用［J］．国防科技大学学报，2024，46（6）：123-130.
	GUO Junbin， YU Lin， YU Chuanqiang ．Application of improved YOLOv5s algorithm in traffic sign detection and recognition［J］．Journal of National University of Defense Technology，2024，46（6）：123-130.
[11]	WANG J， CHEN Y， CAO M，et al ．Improved YOLOv5 network for real-time multi-scale traffic sign detection［J］．Neural Computing and Applications，2023，35（10）：7853-7865.
[12]	MAHAUR B， MISHRA K K ．Small-object detection based on YOLOv5 in autonomous driving systems［J］．Pattern Recognition Letters，2023，168：115-122.
[13]	孟勃，史伟大．改进YOLOv7的交通标志识别模型［J］．中国图象图形学报，2024，29（9）：2737-2752.
	MENG Bo， SHI Weida ．Improved traffic sign recognition model for YOLOv7［J］．Journal of Image and Graphics，2024，29（09）：2737-2752.
[14]	田鹏，毛力．改进YOLOv8的道路交通标志目标检测算法［J］．计算机工程与应用，2024，60（8）：202-212.
	TIAN Peng， MAO Li ．Improved YOLOv8 object detection algorithm for traffic sign target［J］．Computer Engineering and Applications，2024，60（8）：202-212.
[15]	MIN W， LIU R， HE D，et al ．Traffic sign recognition based on semantic scene understanding and structural traffic sign location［J］．IEEE Transactions on Intelligent Transportation Systems，2022，23（9）：15794-15807.
[16]	YAO Y， HAN L， DU C，et al ．Traffic sign detection algorithm based on improved YOLOv4-Tiny［J］．Signal Processing：Image Communication，2022，107：116783/1-11.
[17]	高明华，杨璨．基于改进卷积神经网络的交通目标检测方法［J］．吉林大学学报（工学版），2022，52（6）：1353-1361.
	GAO Ming-hua， YANG Can ．Traffic target detection method based on improved convolution neural network［J］．Journal of Jilin University （Engineering and Technology Edition），2022，52（6）：1353-1361.
[18]	崔静雯，马杰，张宇．基于多尺度联合权重分配的目标检测算法［J］．计算机工程与应用，2022，58（17）：101-110.
	CUI Jingwen， MA Jie， ZHANG Yu ．Target detection algorithm based on multi-scale combined weight distribution［J］．Computer Engineering and Applications，2022，58（17）：101-110.
[19]	XIAO J， ZHAO T， YAO Y，et al ．Context augmentation and feature refinement network for tiny object detection［C］∥ Proceedings of International Conference on Learning Representations．［S.l.］：ICLR，2022：25-29.
[20]	LIU Y， SHAO Z， HOFFMANN N ．Global attention mechanism：retain information to enhance channel-spatial interactions［J］．arXiv preprint：2112.05561，2021.
[21]	ZHUANG J， QIN Z， YU H，et al ．Task-specific context decoupling for object detection［J］．arxiv preprint arxiv：，2023.
[22]	翁俊辉，成乐，黄曼莉，等．基于CS-YOLOv5s的无人机航拍图像小目标检测［J］．电子测量技术，2024，47（7）：157-162.
	WENG Junhui， CHENG Le， HUANG Manli，et al ．Small target detection for UAV aerial images based on CS-YOLOv5s［J］．Electronic Measurement Techno-logy，2024，47（7）：157-162.
[23]	祝江，魏汉迪，肖龙飞，等．通航场景下的海上目标检测算法［J］．船舶工程，2024，46（12）：93-100.
	ZHU Jiang， WEI Handi， XIAO Longfei，et al ．Algorithms for maritime target detection and recognition in navigation scenes［J］．Ship Engineering，2024，46（12）：93-100.
[24]	谢竞，邓月明，王润民．改进YOLOv8s的交通标志检测算法［J］．计算机工程，2024，50（11）：338-349.
	XIE Jing， DENG Yueming， WANG Runmin ．Improved traffic sign detection algorithm based on YOLOv8s［J］．Computer Engineering，2024，50（11）：338-349.
[25]	赵会鹏，曹景胜，潘迪敬，等．改进YOLOv8算法的交通标志小目标检测［J］．现代电子技术，2024，47（20）：141-147.
	ZHAO Huipeng， CAO Jingsheng， PAN Dijing，et al ．Traffic sign small target detection based on improved YOLOv8 algorithm［J］．Modern Electronics Technique，2024，47（20）：141-147.
[26]	任安虎，姜子渊，马晨浩．基于改进YOLOv5s的道路裂缝检测算法［J］．激光杂志，2024，45（4）：88-94.
	REN Anhu， JIANG Ziyuan， MA Chenhao ．Road crack detection algorithm based on improved YOLOv5s［J］．Laser Journal，2024，45（4）：88-94.

类别	标志	英文	缩写
禁令标志	禁止驶入	No Entry	NE
	禁止向左转弯	No Left Turn	NLT
	禁止向右转弯	No Right Turn	NRT
	禁止直行	No Straight Through	NST
	禁止掉头	No U-Turn	NUT
	禁止超车	No Overtaking	NO
	禁止车辆停放	No Parking	NP
	限制标志	Regulatory Sign	RS
指示标志	直行	Straight Ahead	SA
	向左转弯	Turn Left	TL
	向右转弯	Turn Right	TR
	允许掉头	U turn	UT
	分隔带右侧行驶	Keep Right	KR
	分隔带左侧行驶	Keep Left	KL
	人行横道	Pedestrian Crossing	PC
	机动车行驶	Motor Vehicles Only	MVO
	非机动车行驶	Non-Motor Vehicles Only	NMVO
警告标志	交叉口	Intersection	IN
	隧道	Tunnel	TU
	注意行人	Caution： Pedestrians	CP
	注意儿童	Caution： Children	CC
	车道变少	Lane Reduction	LR

参数	数值
epochs	300
batch-size	16
imgsz	640
learning Rate	0.001
Weight_decay	0.000 5
momentum	0.937
depth_multiple	0.33
width_multiple	0.5

YOLOv5s	C	G	T	小目标召回率	ACS	分类准确率/%	收敛时间（300轮）/h
√				0.793	0.45	95.2	1.863
√	√			0.869	0.62	95.4	2.013
√		√		0.831	0.59	95.3	2.084
√			√	0.820	0.46	96.1	1.681
√	√	√		0.880	0.68	95.7	2.131
√	√		√	0.871	0.63	96.5	1.865
√		√	√	0.842	0.60	96.2	1.898
√	√	√	√	0.913	0.72	97.4	1.949

YOLOv5s	C	G	T	mAP @0.50	mAP @0.50∶0.95	精确率	召回率
√				0.971 6	0.761 8	0.965 8	0.946 2
√	√			0.970 5	0.767 3	0.969 5	0.947 3
√		√		0.979 0	0.768 3	0.969 6	0.952 1
√			√	0.980 1	0.778 0	0.968 2	0.951 5
√	√	√		0.966 9	0.767 8	0.972 0	0.952 4
√	√		√	0.980 5	0.781 7	0.972 9	0.953 5
√		√	√	0.981 1	0.784 6	0.974 8	0.953 3
√	√	√	√	0.982 3	0.787 8	0.974 7	0.957 9

模型	mAP@0.50∶0.95	精确率	召回率
Faster R-CNN	0.759 7	0.954 3	0.942 8
YOLOv5s	0.761 8	0.965 8	0.946 2
YOLOv8s	0.773 5	0.969 1	0.949 7
NanoDet-Plus	0.761 2	0.959 1	0.937 6
RT-DETR-Nano	0.778 9	0.969 8	0.951 3
CGT-YOLO	0.787 8	0.974 7	0.957 9