Related Experiment Video
Updated: Jul 2, 2026
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
991
Improving Visible Light Positioning Accuracy Using Particle Swarm Optimization (PSO) for Deep Learning Hyperparameter
Chun-Ming Chang1, Yuan-Zeng Lin1, Chi-Wai Chow1
1Department of Photonics & Graduate Institute of Electro-Optical Engineering, College of Electrical and Computer Engineering, National Yang Ming Chiao Tung University, Hsinchu 30010, Taiwan.
Sensors (Basel, Switzerland)
|December 11, 2025
Summary
Automated hyperparameter tuning using particle swarm optimization (PSO) with convolutional neural networks (CNNs) significantly improves visible light positioning (VLP) accuracy. This approach enhances indoor positioning reliability in challenging, non-ideal lighting conditions.
Area of Science:
- Electrical Engineering
- Computer Science
- Artificial Intelligence
Background:
- Visible light positioning (VLP) offers accurate and cost-effective indoor navigation but suffers from signal degradation due to environmental factors like reflections and light-deficient zones.
- Non-uniform light distribution from LED luminaires and complex indoor optical channel characteristics challenge VLP accuracy.
- Machine learning (ML), particularly Convolutional Neural Networks (CNNs), shows promise in addressing these challenges, but their performance is sensitive to hyperparameter settings.
Purpose of the Study:
- To develop an automated system for optimizing CNN hyperparameters for VLP applications.
- To enhance the accuracy, robustness, and reliability of VLP systems under non-ideal lighting conditions.
- To investigate the effectiveness of integrating Received Signal Strength (RSS) pre-processing and Particle Swarm Optimization (PSO) with CNNs for VLP.
Main Methods:
- A VLP system incorporating RSS signal pre-processing was developed.
- A CNN model was implemented and compared against Linear Regression (LR) and Artificial Neural Networks (ANN) across different height planes (200, 225, 250 cm).
- Particle Swarm Optimization (PSO) was employed for automated hyperparameter tuning of the CNN model.
Main Results:
- The CNN model with pre-processing reduced mean positioning error from 9.83 cm to 5.72 cm (41.81% improvement) at the 200 cm receiver plane.
- Further optimization using CNN + pre-processing + PSO reduced the mean error to 4.93 cm.
- The proposed integrated approach demonstrated significant enhancements in positioning accuracy and model robustness compared to baseline ML models.
Conclusions:
- Automated hyperparameter tuning via PSO integration with CNNs and RSS pre-processing substantially improves VLP accuracy and reliability.
- This method provides a scalable and effective solution for real-world indoor positioning in smart buildings and IoT environments.
- The findings highlight the potential of adaptive ML techniques for overcoming limitations in optical wireless communication-based positioning systems.
