TSC-FL: Communication-Efficient Federated Learning Based on Three-Stage Compression Mechanism for Internet of Vehicles

In the Internet of Vehicles (IoV), there is a widespread need for intelligent collaboration between vehicles. However, traditional centralized machine learning methods face numerous challenges in this highly dynamic environment, particularly concerning privacy protection and communication efficiency. To address these challenges, this paper proposes a communication-efficient federated learning method based on a three-stage compression mechanism and applies it to the IoV scenario. The three-stage compression mechanism integrates parameter sparsity, bipolar quantization, and hybrid encoding to achieve efficient model compression. Experimental results based on the KITTI dataset and YOLOv5 model demonstrate that the proposed method significantly reduces communication load while maintaining the desired model accuracy.