The development of effective and safe machine learning systems for vehicle
number plate recognition (VNPR) is now necessary due to the emergence of the
Internet of Vehicles (IoV). However, traditional centralised techniques run into issues
with data privacy, communication overhead, and centralised data access restrictions.
This study explores the possibilities of decentralized federated learning for VNPR in
the IoV to solve these constraints. Decentralised federated learning, which overcomes
the centralization barrier, allows local model training on the edge devices of
participating cars, protecting data privacy and cutting down on communication
overhead. The ramifications of this paradigm change are examined in this research,
including improved data privacy and security, shared intelligence, and resilience
against errors and assaults. It also looks at the trade-off between performance and
decentralisation while emphasising the balance attained via improved model
aggregation and resource use. Additionally covered is the difficulty of consensus
algorithms and blockchain-based networks, highlighting the need for further
investigation and development. Decentralised federated learning has been identified as
a possible strategy for overcoming the centralization barrier in VNPR systems, opening
the door to the implementation of efficient, secure, and private machine learning in the
IoV.
Keywords: Blockchain technology, Decentralization, Internet of Vehicles (IoV), Vehicle number plate recognition (VNPR).
