A Collaborative Federated Approach for Multi-Client SAR Target Recognition

Synthetic aperture radar (SAR) is widely used due to its capability for all-weather, all-day surveillance. However, the high costs and challenges of data acquisition often lead to a scarcity of data for individual satellites, impeding application expansion. Moreover, the sensitive nature of SAR data necessitates strict privacy requirements when exchanging information among clients. In this context, federated learning (FL) offers a promising solution that provides a distributed learning paradigm that leverages a single client’s data while preserving data privacy. Despite its potential, few studies have applied FL to multi-client SAR target recognition. This paper introduces a novel FL-based framework for collaborative multi-client SAR target recognition, enabling multiple satellites to collaboratively train a model without data sharing, thereby upholding data privacy. Experimental results demonstrate that it effectively preserves model accuracy while protecting privacy. Through iterative training, our framework’s performance converges towards that of centralized training, achieving a validation accuracy of 99.25%, which highlights its effectiveness and practical applicability.