重庆邮电大学学报（自然科学版）2026，Vol.38Issue(1)：30-38,9.DOI:10.3979/j.issn.1673-825X.202501160023
联邦学习隐私保护的图像语义通信研究
Federated learning-based privacy-preserving image semantic communication
摘要
Abstract
Traditional semantic communication systems typically rely on centralized data processing and model training.However,with the growing demand for privacy protection,such an approach is increasingly inadequate for complex commu-nication scenarios.To address this challenge,this paper proposes a federated learning-based image semantic communication architecture that enhances communication system performance while preserving user privacy.The architecture employs a variational autoencoder(VAE)for image semantic encoding and decoding,and distributes model training across user edge devices.Each user independently trains the model using local data and uploads only the updated model parameters to a central server,thereby preventing raw data leakage.The server aggregates the parameters from multiple users to optimize a global model,improving image semantic reconstruction capability.Compared with purely local training,federated learning effectively integrates information from diverse data distributions,thereby enhancing model generalization and communication efficiency.Simulation results demonstrate that the proposed method can significantly improve image reconstruction quality while ensuring privacy protection.关键词
语义通信/机器学习/联邦学习/隐私保护/变分自编码器(VAE)Key words
semantic communication/machine learning/federated learning/privacy protection/variational autoencoder(VAE)分类
信息技术与安全科学引用本文复制引用
余琦,李云,夏士超,姚枝秀..联邦学习隐私保护的图像语义通信研究[J].重庆邮电大学学报（自然科学版）,2026,38(1):30-38,9.基金项目
国家自然科学基金项目(62071077,62301099) （62071077,62301099）
重庆市自然科学基金项目(2022NSCQ-LZX0191) National Natural Science Foundation of China(62071077,62301099) （2022NSCQ-LZX0191）
Natural Science Foundation of Chongqing(2022NSCQ-LZX0191) （2022NSCQ-LZX0191）
