Electric vehicle license plate detection is of great significance in non-motor vehicle management and urban road supervision. Aiming at the problems of large changes in shooting Angle, complex background, small target size and serious occlusion of license plate images in actual scenes, an improved YOLO11 license plate detection algorithm integrating lightweight convolution structure and feature enhancement mechanism, called YOLO11-GCE, was proposed. Firstly, the C3k2_GhostConv module was constructed by fusing the C3k2 structure and Ghost convolution, which enhances the feature extraction capability of the backbone network while reducing computational redundancy. Secondly, EMA (efficient multi-scale attention) module was introduced at the end of the network, which stabilizes deep semantic information through a cross-spatial learning mechanism and improves the model's ability to discriminate small targets and blurred license plates in complex backgrounds. Finally, the lightweight GSConv structure was used to replace some standard convolutions, which optimizes the computational efficiency of the feature fusion network and enhances the detection performance for multi-pose targets. Through the synergistic effect of these improvements, YOLO11-GCE achieves significant enhancements in both detection accuracy and inference efficiency. Experiments conducted on a self-built electric vehicle license plate dataset show that the mAP@50:95 of YOLO11-GCE reaches 84.4%, an improvement of 1.7 percentage points over the original YOLO11, with a parameter reduction of about 19.8%, achieving better inference efficiency while ensuring detection accuracy.
HE Xin-yu, LIN Yu-long, HOU Shu-yu, SHI Bo, TONG Jun-wei, WANG Jia, ZHOU An-liang.
Electric Vehicle License Plate Detection Algorithm Based on Improved YOLO11[J].
Science Technology and Engineering, 2026, 26(22): 9596-9605 DOI:10.12404/j.issn.1671-1815.2506919
(3)平均精度为目标检测任务中最常用的综合评估指标,主要衡量模型在不同置信度阈值下的检测性能。采用两种常用的mAP计算方式:mAP@50表示以IoU阈值为0.5时的平均精度(AP);mAP@50:95表示IoU从0.5到0.95,每隔0.05取一个值,共10个IoU阈值下的平均AP。其中,IoU(intersection over union)为重叠区域的面积比并集区域的面积,表示预测框与真实框的重叠程度。
ZhangTeng. Research on electric bicycle license plate recognition method based on deep learning[D]. Hefei: Hefei University of Technology, 2023.
[3]
JiangP, ErguD, LiuF, et al. A review of YOLO algorithm deve-lopments[J]. Procedia computer science, 2022, 199: 1066-1073.
[4]
AliML, ZhangZ. The YOLO framework: a comprehensive review of evolution, applications, and benchmarks in object detection[J]. Computers, 2024, 13(12). DOI: 10.3390/computers13120336.
[5]
WangX, LiH, YueX, et al. A comprehensive survey on object detection YOLO[C]// International Symposium on Advanced Technologies and Applications in the Internet of Things (ATAIT). Kusatsu: ATAIT, 2023: 261125293.
ZhangDi, ZhouAnliang, WenMeng, et al. Quality defect detection of crystal drawing process based on improved YOLOv8n[J]. Science Technology and Engineering, 2025, 25(3): 969-976.
[12]
LinT Y, GoyalP, GirshickR, et al. Focal loss for dense object detection[C]// Proceedings of the IEEE International Conference on Computer Vision. Venice: IEEE, 2017: 2980-2988.
[13]
Ultralytics. YOLOv12 official documentation[EB/OL]. [2025-09-01] https://docs.ultralytics.com/models/yolov12/.
[14]
HuangZ, WangH, WangQ. Research on real-time target identification method for slope slip risk based on YOLOv8-Tiny[C]// 4th International Conference on Cloud Computing, Big Data Application and Software Engineering (CBASE). New York: IEEE, 2025: 243-247.
[15]
XuY, DuW, DengL, et al. Ship target detection in SAR images based on SimAM attention YOLOv8[J]. IET Communications, 2024, 18(19): 1428-1436.
[16]
LiH, LiJ, WeiH, et al. Slim-neck by GSConv: a lightweight-design for real-time detector architectures[J]. Journal of Real-Time Image Processing, 2024, 21(3). DOI: 10.1007/s11554-024-01436-6.
[17]
ZhengX, QiuY, ZhangG, et al. ESL-YOLO: small object betection with effective feature enhancement and spatial-context-guided fusion network for remote sensing[J]. Remote Sensing, 2024, 16.DOI: 10.3390/rs16234374.
[18]
JiangP, YangX, WanY, et al. DRBD-YOLOv8: a lightweight and efficient anti-UAV detection model[J]. Sensors, 2024, 24.DOI: 10.3390/s24227148.
[19]
LiJ, WuJ, ShaoY. FSNB-YOLOV8: improvement of object detection model for surface defects inspection in online industrial systems[J]. Applied Sciences, 2024, 14.DOI: 10.3390/app14177913.
[20]
TangS, ZhangS, FangY. HIC-YOLOv5: improved YOLOv5 for small object detection[C]// IEEE International Conference on Robotics and Automation (ICRA). Yokohama: IEEE, 2024: 6614-6619.
[21]
KhanamR, HussainM. YOLO11: an overview of the key architectural enhancements[J]. arXiv Preprint arXiv: 2410.17725, 2024.
[22]
OuyangD, HeS, ZhangG, et al. Efficient multi-scale attention module with cross-spatial learning[C]// IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Rhodes Island: IEEE, 2023: 1-5.