Spatial-temporal federated transfer learning with multi-sensor data fusion for cooperative positioning
Xiaokang Zhou,Qiuyue Yang,Qiang Liu,Wei Liang,Kevin Wang,Zhi Liu,Jianhua Ma,Qun Jin
DOI: https://doi.org/10.1016/j.inffus.2023.102182
IF: 18.6
2023-12-06
Information Fusion
Abstract:With the development of advanced embedded and communication systems, location information has become a crucial factor in supporting context-aware or location-aware intelligent services. Among these services, intelligent transportation systems have the strictest requirements for real-time, accurate, and private location data. In this study, a Spatial-Temporal Federated Transfer Learning (ST-FTL) model is designed and introduced to achieve more accurate cooperative positioning by effectively utilizing spatial–temporal data in intelligent transportation systems (ITS). Specifically, a three-layer FTL based framework is constructed to enhance location prediction accuracy while ensuring the privacy of acquired location data. The proposed federated learning(FL) approach considers, selects, and aggregates both spatial and temporal data attributes to improve prediction accuracy and achieve faster convergence. A multi-attribute based spatial–temporal clustering algorithm, along with convolutional gated units, is developed for efficient global model initialization and weight aggregation based on a newly designed transfer model that selects data from different regions in the spatial dimension. Additionally, a deep fusion based local training model is built to integrate data from various sensors and improve prediction accuracy. A lightweight Siamese network is employed to enhance data augmentation for insufficient GPS data in the temporal dimension, leading to improved prediction error. Experimental evaluations are conducted using the KITTI dataset, and the results demonstrate that the proposed method achieves superior prediction accuracy and faster convergence compared to other state-of-the-art positioning methods.
computer science, artificial intelligence, theory & methods