面向海上无人系统的边缘模型协同与数据压缩算法

doi:10.12305/j.issn.1001-506X.2025.05.34

摘要/Abstract

摘要：

在海上环境中, 无人系统可以通过边缘人工智能(artificial intelligence, AI)模型协同实现数据采集与边缘处理任务。面向通信环境差、通信带宽有限、通信链路易被干扰等问题, 首先, 从无人机AI模型协同的角度, 提出一种联邦互蒸馏的模型协同训练方法, 以减少模型训练数据传输带宽要求。其次，从数据去冗余压缩的角度出发, 提出一种数据差分动态压缩方法, 以降低数据传输频次。仿真实验表明, 所提出的联邦智能模型互蒸馏训练方法训练的模型性能明显优于分布式训练的模型的性能, 且较之于集中式的模型训练方法节省大量通信带宽, 提出的差分动态数据压缩方法能够大幅减少通信报文的长度与发送频率, 适应于带宽受限的通信弱连接环境。

关键词: 通信弱连接, 边缘模型协同, 联邦互蒸馏, 数据压缩

Abstract:

In sea environment, unmanned system relies on edge artificial intelligence (AI) model collaboration to implement data collection and edge processing tasks. However, facing with problems such as poor communication links, limited communication bandwidth, and sensitive to interferences, this paper firstly proposes a model collaborative training method, federated mutual distillation, to reduce the bandwidth requirements for model training data, from the perspective of AI model collaboration among unmanned aerial vehicle. Secondly, from the perspective of data de-redundancy transmission, a data differential dynamic compression method is proposed to reduce the frequency of data transmission. Simulation results show that the performance of the model, trained with the federated mutual distillation training method, is better than that of the benchmarks, and a cost of communication bandwidth is reduced compared to the centralized training models. The proposed data differential dynamic compression method can greatly reduce the sending length and frequency of communication messages, and adapt to the bandwidth bottleneck in the weak communication connection environment.

Key words: weak communication link, edge model collaboration, federated mutual distillation, data compression

中图分类号:

TP301

姜俊, 张家瑞, 潘吉龙, 孙国林. 面向海上无人系统的边缘模型协同与数据压缩算法[J]. 系统工程与电子技术, 2025, 47(5): 1718-1727.

Jun JIANG, Jiarui ZHANG, Jilong PAN, Guolin SUN. Edge model collaboration and data compression algorithm for unmanned systems in sea[J]. Systems Engineering and Electronics, 2025, 47(5): 1718-1727.

图/表 10

图1

表1

表2

图2

图3

表3

图4

图5

图6

图7

参考文献 30

1	汪燕梅, 徐海晖. 浅析无人机在海事监管中的应用[J]. 中国水运, 2020, 20 (1): 44- 45.
	WANG Y M , XU H H . On the application of UAV in maritime supervision[J]. China Water Transport, 2020, 20 (1): 44- 45.
2	苏文. 智慧海洋通信网络系统建设研究[J]. 中国信息化, 2019 (8): 81- 82.
	SU W . Research on the construction of intelligent ocean commu nication network system[J]. Zhongguo Xinxihua, 2019 (8): 81- 82.
3	HU Y F, XU L, GAO J C, et al. Analysis of marine wireless communication channel under high sea conditions[C]//Proc. of the IEEE 5th International Conference on Electronic Information and Communication Technology, 2022: 274-277.
4	黄柏铭. 卫星数据压缩[M]. 哈尔滨: 哈尔滨工业大学出版社, 2015.
	HUANG B M . Satellite data compression[M]. Harbin: Harbin Institure of Technology Press, 2015.
5	许小华, 陈豹, 王海菁, 等. 基于改进YOLOv4-Tiny的河湖船舶目标检测算法[J]. 人民长江, 2023, 54 (9): 264- 271.
	XU X H , CHEN B , WANG H J , et al. Object detection algorithm for ships on rivers and lakes based on improved YOLOv4-Tiny[J]. Yangtze River, 2023, 54 (9): 264- 271.
6	WU F Y, DONG C, QU Y B, et al. CIOFL: collaborative infer ence-based online federated learning for UAV object detection[C]// Proc. of the IEEE 19th International Conference on Mobile Ad Hoc and Smart Systems, 2022: 258-259.
7	FANG L L, HU H R, BI J Q. Research on UAV target recognition technology based on federated learning[C]//Proc. of the 2nd International Conference on Computer Engineering and In-telligent Control, 2021: 119-122.
8	KHAN F S, KHAN S, MOHD M, et al. Federated learning-based UAVs for the diagnosis of plant diseases[C]//Proc. of the International Conference on Engineering and Emerging Techno-logies, 2022.
9	SUN T K, WANG X Y, UMEHIRA M, et al. Split learning assisted multi-UAV system for image classification task[C]//Proc. of the IEEE 97th Vehicular Technology Conference, 2023.
10	WANG R. Privacy-preserving incentive scheme design for UAV-enabled federated learning[C]//Proc. of the IEEE Wireless Communications and Networking Conference, 2024.
11	HOU X , WANG J , JIANG C , et al. UAV enabled covert fede rated learning[J]. IEEE Trans.on Wireless Communication, 2023, 22 (10): 6793- 6809. doi: 10.1109/TWC.2023.3245621
12	LYU B Y . Analysis and research on common lossless data compression algorithms[J]. Electronics World, 2019, 562 (4): 5- 6.
13	LI L D , MA T H , YOU W B . Analysis of common lossless compression algorithm[J]. Electronic Design Engineering, 2009, 17 (1): 49-50, 53.
14	蒋鹏, 吴建峰, 吴斌, 等. 基于自适应最优消零的无线传感器网络数据压缩算法研究[J]. 通信学报, 2013, 34 (2): 1- 7.
	JIANG P , WU J F , WU B , et al. Data compression method for wireless sensor networks based on adaptive optimal zero suppression[J]. Journal of Communications, 2013, 34 (2): 1- 7.
15	LUO J , JIANG G R , JIANG Y R . A new scatheless compression scheme for oceanological grid data[J]. Marine Forecasts, 2011, 28 (3): 55- 61.
16	LIU C , LI Y F , CHEN H . Improvement and implementation of lossless data compression technology based on LZW[J]. Electronic Design Engineering, 2019, 27 (24): 51- 56.
17	KALAIYARASI D , KALPALATHA R . Design and imple-mentation of least mean square adaptive FIR filter using offset binary coding based distributed arithmetic, microprocessors and microsystems[J]. Microprocessors and Microsystems, 2019, 71, 102884.
18	WU X Y . Research of arithmetic coding in image compression[J]. Computer Digital and Engineering, 2017, 45 (9): 1863- 1865.
19	陈海生, 郭晓云, 王峰, 等. 基于北斗短报文的渔获信息压缩传输方法[J]. 农业工程学报, 2015, 31 (22): 155- 160.
	CHEN H S , GUO X Y , WANG F , et al. Fishery harvesting information compressing and transmitting method based on Beidou short message[J]. Transactions of the Chinese Society of Agricultural Engineering, 2015, 31 (22): 155- 160.
20	欧阳中辉, 樊辉锦, 陈青华, 等. 基于北斗短报文的特种车辆状态信息压缩传输方法研究[J]. 兵器装备工程学报, 2020, 41 (9): 124- 129.
	OUYANG Z H , FAN H J , CHEN Q H , et al. Research on compression transmission method of special vehicle status information based on beidou short message[J]. Journal of Ordnance Equipment Engineering, 2020, 41 (9): 124- 129.
21	HUANG M, WANG B J, WAN J C, et al. Improved blood cell detection method based on YOLOv5 algorithm[C]//Proc. of the IEEE 6th Information Technology Networking Electronic and Automation Control Conference, 2023: 992-996.
22	WU H, ZHAO X L, QIAO J H, et al. An improved YOLOv5 algorithm for elderly fall detection[C]//Proc. of the International Conference on Innovation, Knowledge, and Management, 2023: 83-88.
23	ZHU D, DAI L, DU P. CCE-YOLOv5s: an improved YOLOv5 model for UAV small target detection[C]//Proc. of the IEEE 5th International Conference on Civil Aviation Safety and Information Technology, 2023: 824-829.
24	SU T , ZHANG J , YU Z , et al. STKD: distilling knowledge from synchronous teaching for efficient model compression[J]. IEEE Trans.on Neural Networks and Learning System, 2023, 34 (12): 10051- 10054.
25	PENG Y T , CHENG K H , FANG I S , et al. Single image reflection removal based on knowledge-distilling content disentanglement[J]. IEEE Signal Processing Letters, 2022, 29, 568- 572.
26	CHAMBI S , LEMIRE D , KASER O , et al. Better bit-map performance with roaring bit-maps[J]. Software: Practice and Experience, 2016, 46 (5): 709- 719.
27	LEMIRE D , SI Y , KAI G , et al. Consistently faster and smaller compressed bitmaps with roaring[J]. Software: Practice and Experience, 2016, 46 (11): 1547- 1569.
28	HU L. An improved YOLOv5 algorithm of target recognition[C]//Proc. of the IEEE 2nd International Conference on Electrical Engineering, Big Data and Algorithms, 2023: 1373-1377.
29	XU L D, WU Z B, SUN J, et al. A distributed hyperspectral target detection algorithm based on background reconstruction for cloud platforms[C]//Proc. of the 13th Workshop on Hyperspectral Imaging and Signal Procession: Evolution in Remote Sensing, 2023.
30	ZHANG S Y , LI J , SHI L . Federated learning in intelligent transportation systems: recent applications and open problems[J]. IEEE Trans.on Intelligent Transportation Systems, 2023, 25 (5): 3259- 3285.

长度/B	信息内容
1	无人机编号
4	无人机位置变化量
1	船舶数量变化量
1	船舶编号
1	船舶类别变化量
4	船舶位置变化量
1	船舶速度变化量
1	船舶航向变化量
4	时间戳

信息	类型	阈值
船舶数量	离散状态	变化即传
船舶类型	离散状态	变化即传
船舶位置	连续状态	跨越网格
船舶航速	连续状态	变化值超过5格/h
船舶航向	离散状态	变化即传
无人机位置	连续状态	跨越网格

训练方式	传输数据量
集中式训练/GB	1.2
联邦聚合训练/MB	72
联邦互蒸馏训练/MB	145
分布式训练/MB	0

[1]	谭洪1,2,3, 洪峻2,3, 仇晓兰1,2. 原始数据压缩对简缩极化SAR极化信息的影响[J]. 系统工程与电子技术, 2016, 38(8): 1780-1786.
[2]	赵旦峰, 司佳希, 梁明珅, 段晋珏. 基于算术编码的低冗余LT码及其在安全通信中的应用[J]. 系统工程与电子技术, 2016, 38(2): 409-414.
[3]	谭洪, 仇晓兰, 洪峻. 原始数据压缩对全极化SAR极化信息的影响[J]. 系统工程与电子技术, 2015, 37(9): 2029-2034.
[4]	李达, 胡飞, 郑学合. 数据压缩卡尔曼滤波的多传感器精确配准方法[J]. Journal of Systems Engineering and Electronics, 2011, 33(11): 2363-2367.
[5]	宋鸿梅1,2,王岩飞1,潘志刚1. 基于FFTBAQ的SAR原始数据压缩新算法[J]. Journal of Systems Engineering and Electronics, 2009, 31(11): 2613-2617 .