Research output: Contribution to Journal/Magazine › Journal article › peer-review
Research output: Contribution to Journal/Magazine › Journal article › peer-review
}
TY - JOUR
T1 - Traffic Sign Recognition Using Optimized Federated Learning in Internet of Vehicles
AU - Lian, Zhuotao
AU - Zeng, Qingkui
AU - Wang, Weizheng
AU - Xu, Dequan
AU - Meng, Weizhi
AU - Su, Chunhua
PY - 2024/2/15
Y1 - 2024/2/15
N2 - Traffic sign recognition (TSR) is vital for vehicle safety and navigation, especially in the era of autonomous cars. Internet of Vehicles (IoV) provide a promising infrastructure for vehicular networks due to their agility and interoperability. However, privacy concerns and network restrictions hinder the collection of massive data from distributed automotive sensors in IoV. To address these challenges, this article proposes the application of federated learning (FL) and model sparsification to optimize traffic sign recognition (TSR) in autonomous vehicles. FL enables decentralized learning while preserving data privacy, and model sparsification significantly reduces communication costs. Furthermore, we incorporate the Adam optimizer for local training, ensuring efficient model optimization on each vehicle. Experimental results demonstrate the effectiveness of our approach, with improved TSR performance while mitigating privacy risks and enhancing communication efficiency. This research contributes to the advancement of TSR in IoV by introducing FL, model sparsification, and the use of the Adam optimizer for local training, facilitating efficient and privacy-preserving vehicular network learning.
AB - Traffic sign recognition (TSR) is vital for vehicle safety and navigation, especially in the era of autonomous cars. Internet of Vehicles (IoV) provide a promising infrastructure for vehicular networks due to their agility and interoperability. However, privacy concerns and network restrictions hinder the collection of massive data from distributed automotive sensors in IoV. To address these challenges, this article proposes the application of federated learning (FL) and model sparsification to optimize traffic sign recognition (TSR) in autonomous vehicles. FL enables decentralized learning while preserving data privacy, and model sparsification significantly reduces communication costs. Furthermore, we incorporate the Adam optimizer for local training, ensuring efficient model optimization on each vehicle. Experimental results demonstrate the effectiveness of our approach, with improved TSR performance while mitigating privacy risks and enhancing communication efficiency. This research contributes to the advancement of TSR in IoV by introducing FL, model sparsification, and the use of the Adam optimizer for local training, facilitating efficient and privacy-preserving vehicular network learning.
U2 - 10.1109/JIOT.2023.3312348
DO - 10.1109/JIOT.2023.3312348
M3 - Journal article
VL - 11
SP - 6722
EP - 6729
JO - IEEE Internet of Things Journal
JF - IEEE Internet of Things Journal
SN - 2327-4662
IS - 4
ER -