FedRD: Privacy-preserving adaptive Federated learning framework for intelligent hazardous Road Damage detection and warning

Abstract

Road damages have caused numerous fatalities. Therefore, the study of road damage detection, especially hazardous road damage detection and warning, is critical in improving traffic safety. Existing road damage detection systems mainly process data on clouds, however, they are not able to warn users timely due to the long latency. Recent edge-computing techniques mitigate this problem while users can only receive warnings of hazardous road damages within a small area due to the limited communication range of edges. Besides, untrusted edges might misuse users’ sensitive information. In this paper, we propose FedRD: a novel privacy-preserving edge-cloud and Federated learning-based framework for intelligent hazardous Road Damage detection and warning. In FedRD, a new hazardous road damage detection model is developed leveraging the advantages of hierarchical feature fusion. A novel adaptive federated learning strategy is designed for robust model learning from different edges with limited and unequally-sized datasets. A new individualized differential privacy approach with pixelization is proposed to protect users’ privacy before sharing data. Simulation results demonstrate that FedRD achieves a high detection performance and provides fast responses with accurate warning information covering a wider area while preserving users’ privacy, even when some edges have limited data.