Related Experiment Video
Updated: Jun 22, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
515
A lightweight network architecture for traffic sign recognition based on enhanced LeNet-5 network
Yuan An1,2, Chunyu Yang1, Shuo Zhang3
1China University of Mining and Technology, Engineering Research Center of Intelligent Control for Underground Space, Ministry of Education, Xuzhou, China.
Frontiers in Neuroscience
|July 4, 2024
Summary
This study presents an improved, lightweight convolutional neural network (CNN) for traffic sign recognition in unmanned driving systems. The enhanced model achieves 97.53% accuracy, offering fast execution and easy deployment on embedded devices.
Area of Science:
- Computer Science
- Artificial Intelligence
- Robotics
Background:
- Traffic sign recognition is crucial for autonomous driving systems.
- Existing methods often struggle with the computational demands of embedded hardware.
- There is a need for efficient and accurate traffic sign recognition models suitable for real-time applications.
Purpose of the Study:
- To develop a lightweight convolutional neural network (CNN) architecture for traffic sign recognition.
- To optimize the CNN for excellent recognition accuracy and fast execution speed on embedded systems.
- To ensure easy deployment for unmanned driving applications.
Main Methods:
- Modified the classical LeNet-5 convolutional neural network (CNN) architecture.
- Implemented image preprocessing techniques.
- Improved spatial pooling, optimized neurons, and refined activation functions.
- Tested the enhanced CNN on the German Traffic Sign Recognition Benchmark (GTSRB) dataset.
Main Results:
- The improved CNN architecture achieved a recognition accuracy of 97.53% on the GTSRB dataset.
- Demonstrated higher recognition accuracy with reduced interference time.
- Showcased significantly reduced algorithm loss during training.
- Outperformed other lightweight network models in recognition performance.
Conclusions:
- The developed lightweight CNN is suitable for embedded applications in unmanned driving systems.
- The optimized architecture balances high accuracy with fast execution speed.
- The model's simplicity and low algorithmic complexity facilitate wider adoption in autonomous systems.
