Related Experiment Video
Updated: Jan 17, 2026
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
5.3K
Research on license plate recognition based on graphically supervised signal-assisted training.
Dianwei Chi1, Zehao Jia1, Lizhen Liu2
1Artificial Intelligence Institute, Yantai Institute of Technology, Yantai, Shandong, China.
Peerj. Computer Science
|September 24, 2025
Summary
This study enhances license plate recognition (LPR) models for embedded systems. By incorporating graphical supervision, the improved LPRNet achieves higher accuracy while remaining lightweight, crucial for complex urban transportation.
Area of Science:
- Computer Vision
- Deep Learning
- Artificial Intelligence
Background:
- Urban transportation complexity necessitates accurate license plate recognition (LPR).
- Limited computational resources on embedded devices pose challenges for deploying LPR models.
- Developing lightweight and accurate LPR models is a critical research problem.
Purpose of the Study:
- To propose an improved license plate recognition algorithm.
- To enhance the accuracy of the License Plate Recognition Network (LPRNet) model.
- To create a lightweight yet highly accurate LPR model suitable for embedded systems.
Main Methods:
- Utilized the License Plate Recognition Network (LPRNet) as the base architecture.
- Incorporated graphically supervised signals for assisted training.
- Added an auxiliary training branch guided by graphical signals to improve feature learning.
Main Results:
- Achieved accuracy improvements across all test sets of the Chinese City Parking Dataset (CCPD).
- Demonstrated an average accuracy increase of 5.86% and a maximum increase of 10.9% compared to the original LPRNet.
- Showcased average character precision improvement of 2.1% and average recall of 6.9%.
Conclusions:
- The proposed method significantly enhances LPR accuracy while maintaining a lightweight model.
- The approach offers a viable solution for LPR in complex scenarios on resource-constrained devices.
- This study provides valuable insights for other deep learning image recognition tasks.
