基于数据生成和深度神经网络的空间非合作目标行为意图识别  

Spatial Non-cooperative Target Behavior Intent Recognition Based on Data Generation and Deep Neural Networks

作　　者：余静 彭晓东 谢文明 覃润楠 王有亮 YU Jing;PENG Xiaodong;XIE Wenming;QIN Runnan;WANG Youliang(School of Fundamental Physics and Mathematical Sciences,Hangzhou Institute for Advanced Study,UCAS,Hangzhou 310024;National Space Science Center,Chinese Academy of Sciences,Beijing 100190)

机构地区：[1]国科大杭州高等研究院基础物理与数学科学学院,杭州310024 [2]中国科学院国家空间科学中心,北京100190

出　　处：《空间科学学报》2024年第6期1134-1146,共13页Chinese Journal of Space Science

摘　　要：在信息化条件下,空间环境变得日益复杂,空间非合作目标数量日益增长,地面操作人员难以迅速准确地根据非合作目标的运动规律识别其意图,因此提出基于堆叠自编码器(SAE)和门控循环网络(GRU)的空间非合作目标行为意图识别模型,用于协助地面操作人员识别非合作目标的意图.该模型利用自编码器对时间序列数据进行压缩,提取其中的关键特征,并采用GRU网络对轨迹进行分类.由于目前尚无公开的非合作目标行为的轨道数据可供使用,仅依靠少量已知数据难以充分训练模型.为解决样本不足导致识别效果不佳的问题,提出一种仿真样本生成方法,通过仿真得到大量目标行为的轨道数据,可用于空间非合作目标行为意图的识别.得到仿真数据后,将仿真数据集作为输入开展实验,结果显示,与仅使用长短期记忆网络(LSTM)、门控循环单元–全卷积网络(GRU-FCN)、堆叠自编码器(SAE)以及反向传播(BP)等单一模型相比,本方法在准确率、损失值性能指标上均有显著提升,准确率达到了97.8%.Under the conditions of informatization,the space environment has become increasingly complex,and the number of non cooperative targets in space is growing.Ground operators find it diffi-cult to quickly and accurately identify the intentions of non cooperative targets based on their motion patterns.Therefore,a spatial non cooperative target behavior intention recognition model based on Stacked Autoencoder(SAE)and Gated Recurrent network Unit(GRU)was proposed to assist ground operators in identifying the intention of non cooperative targets.This model utilizes an autoencoder to compress time series data,extract key features,and uses a GRU network to classify trajectories.At present,there is no publicly available orbit data for non cooperative target behavior,and it is difficult to fully train the model with only a small amount of known data.To solve the problem of poor recognition performance caused by insufficient samples,a simulation sample generation method is proposed,which obtains a large amount of target behavior trajectory data through simulation for the recognition of spa-tial non cooperative target behavior intentions.After the simulation data is obtained,the simulation da-ta set is used as the input.The experimental results show that compared with the single model only us-ing the Long and Short Term Memory network(LSTM),GRU-FCN,SAE,and Back-Propagation(BP),this method has significantly improved the accuracy and loss value performance indicators,reaching 97.8% accuracy.

关 键 词：非合作目标 样本仿真方法 意图识别 深度学习 

分 类 号：V19[航空宇航科学与技术—人机与环境工程]