首页|期刊导航|天地一体化信息网络|星—机协同联邦学习:重叠感知与多属性博弈融合的海洋船舶检测

星—机协同联邦学习:重叠感知与多属性博弈融合的海洋船舶检测OA

Satellite-UAV collaborative federated learning:marine ship detection integrating overlap awareness and multi-attribute game-theoretic

中文摘要英文摘要

海洋船舶检测是海上交通安全与海洋资源管理的核心支撑,卫星遥感广覆盖与无人机高分辨率的协同应用为提升检测精度提供了可能,但传统集中式方法面临数据隐私泄露、多源数据冗余冲突及通信开销过高的挑战.为此,提出一种重叠感知与多属性博弈融合的卫星—无人机协同联邦学习架构,融合基于Haversine公式的重叠区域处理策略与多属性博弈论客户端筛选机制.通过本地训练与安全聚合保障数据隐私,自适应裁剪减少数据冗余,动态筛选高价值客户端,优化通信效率.实验结果表明,该架构在交并比阈值为0.5时的平均精度(mAP@0.5)较FedAvg提升15.8%,通信开销降低50%,推理速度达28.5 FPS,实现了检测精度、效率与隐私保护的协同优化.

Marine ship detection is a core support for maritime traffic safety and marine resource management.The collaborative appli-cation of satellite remote sensing with wide coverage and UAVs(unmanned aerial vehicles)with high resolution provides the possibility to improve detection accuracy,but traditional centralized methods face challenges such as data privacy leakage,multi-source data redun-dancy and conflict,and excessive communication overhead.To address these issues,an overlap-aware federated learning framework for satellite-UAV collaboration is proposed,integrating an overlapping region processing strategy based on the Haversine formula and a multi-attribute game-theoretic optimization mechanism.Data privacy is ensured through local training and secure aggregation,data re-dundancy is reduced via adaptive cropping,and communication efficiency is optimized by dynamically selecting high-value clients.Ex-perimental results show that compared with FedAvg,the framework improves mAP@0.5 by 15.8%,reduces communication overhead by 50%,and achieves an inference speed of 28.5 FPS,realizing the coordinated optimization of detection accuracy,efficiency,and pri-vacy protection.

迟子杰;徐双;孙昭

太原理工大学人工智能学院,山西 太原 030024太原理工大学人工智能学院,山西 太原 030024太原理工大学人工智能学院,山西 太原 030024

信息技术与安全科学

海洋船舶检测星—机协同联邦学习重叠区域处理多属性博弈

marine ship detectionsatellite-UAV collaborationfederated learningoverlapping region processingmulti-attribute game-theory

《天地一体化信息网络》 2026 (3)

67-77,11

国家自然科学基金资助项目(No.62102280)山西省科技合作交流专项(No.202404041101042)吕梁市高层次科技人才专项(No.2024RC04)山西省重点研发计划(No.202102020101001) The National Natural Science Foundation of China(No.62102280),Shanxi Science and Technology Cooperation and Exchange Special Project(No.202404041101042),Lüliang City High-level Scientific and Technological Talents Special Project(No.2024RC04),Shanxi Key Research and Development Program(No.202102020101001)

10.11959/j.issn.2096-8930.tdyt2026005

评论