天基红外预警系统作为导弹防御系统的核心环节之一，其任务是通过对导弹目标的精准定位和轨迹预测，为拦截决策提供实时数据支持。然而，在多目标观测场景中，目标运动的高速性、非线性和复杂性，以及多目标交叉与遮挡现象，显著增加了目标定位和轨迹跟踪的难度。传统方法在复杂多目标环境中的定位精度和跟踪稳定性不足，难以满足当前天基预警系统的需求。针对上述问题，本文设计了一种基于LSTM时序特征提取与GNN轨迹关联的多目标跟踪算法，并结合导弹运动建模和多视几何定位，实现了对弹道导弹目标的高精度定位与轨迹关联跟踪。主要研究内容与创新点如下：

（1）针对多星观测弹道导弹主动段的轨迹数据需求，本文研究了卫星观测场景仿真理论，结合动力学与运动学规律，研究了导弹不同飞行阶段的受力特点及约束条件，构建了分段式的导弹飞行程序模型，充分模拟了实际观测中的多目标复杂运动场景，并通过仿真生成了三维主动段轨迹数据集，为后续研究提供了可靠的数据输入。

（2）针对多星协同观测的定位精度和运动特征提取难题，提出了一种集成多视几何定位算法与LSTM时序特征提取方法的创新性框架。在定位方面，通过粒子滤波校正观测值，并分析目标与卫星观测视线的空间几何关系，优化了多视几何前方交会算法，实现了定位误差控制在1km以内。在特征提取方面，基于LSTM构建了高效的运动特征提取网络，通过对弹道导弹主动段运动数据的时序性分析，该方法能够高效挖掘轨迹中的时序动态特征，捕捉目标运动状态的时间变化规律。实验表明，该时序特征提取模型能够精确捕捉导弹目标的运动特征（速度矢量和关机点），为轨迹关联与跟踪提供了高精度的特征，增强模型的鲁棒性和精确性。

（3）针对多目标运动过程中的复杂场景下传统关联算法精度不高的问题，提出了结合LSTM时序特征与GNN空间特征的时空融合轨迹关联方法。通过GNN的空间特征聚合能力与运动时序特征的动态调整，设计了改进的轨迹滤波跟踪算法，实现了对轨迹落点的高精度预测。该方法有效降低了复杂场景下的错误关联率，均方根误差（RMSE）降低了80%，同时提升了轨迹跟踪的实时性和预测精度。实验结果显示，该算法在多目标高动态交叉场景中表现优异，显著提升了多目标跟踪的稳定性与准确性。

As a critical component of missile defense systems, space-based infrared early warning systems are tasked with providing real-time data support for interception decisions through accurate missile target localization and trajectory prediction. However, in multi-target observation scenarios, the challenges posed by the high-speed, nonlinear, and complex motion of targets, coupled with target crossing and occlusion phenomena, significantly increase the difficulty of localization and trajectory tracking. Traditional methods often struggle to meet the requirements of current space-based early warning systems due to insufficient localization accuracy and tracking stability in complex multi-target environments. To address these issues, this thesis proposes a multi-target tracking algorithm based on LSTM temporal feature extraction and GNN trajectory association, combined with missile motion modeling and multi-view geometric localization, to achieve high-precision localization and trajectory association tracking of ballistic missile targets. The main contributions and innovations are as follows:

（1）To meet the trajectory data requirements for multi-satellite observation of ballistic missile boost phases, this study explores satellite observation scenario simulation theories. By integrating dynamic and kinematic principles, the force characteristics and constraints across different flight stages are analyzed, resulting in a segmented missile flight program model. This model effectively simulates complex multi-target motion scenarios in realistic observations and generates a 3D trajectory dataset for the boost phase, providing reliable input data for subsequent studies.

（2）To address challenges in localization accuracy and feature extraction for multi-satellite cooperative observation, an innovative framework combining multi-view geometric localization and LSTM-based temporal feature extraction is proposed. For localization, particle filters are employed to correct observations, while the spatial geometric relationships between the target and satellite observation lines are analyzed to optimize the multi-view forward intersection algorithm, achieving localization errors below 1 km. For feature extraction, an efficient motion feature extraction network is constructed based on LSTM, leveraging the temporal analysis of ballistic missile boost-phase data to capture dynamic temporal features in the trajectory. This model accurately extracts critical motion characteristics (e.g., velocity vectors and burnout points), providing high-precision features for trajectory association and tracking. Experimental results demonstrate enhanced model robustness and precision.

（3）To overcome the limitations of traditional association algorithms in complex multi-target scenarios, a spatiotemporal trajectory association method is proposed by combining LSTM temporal features with GNN spatial features. By leveraging the spatial feature aggregation capability of GNN and dynamically adjusting motion temporal features, an improved trajectory filtering and tracking algorithm is designed, enabling high-precision impact point predictions. This approach effectively reduces false association rates in complex scenarios, achieving an 80% reduction in RMSE and significantly improving real-time trajectory tracking and prediction accuracy. Experimental results confirm the superior performance of this algorithm in highly dynamic multi-target intersection scenarios, significantly enhancing the stability and accuracy of multi-target tracking.

[1] 王云萍. 美国天基红外导弹预警技术分析[J]. 光电技术应用, 2019, 34（3）: 1-5.

[2] 姜云彬, 赵华, 郝刚刚, 等. "天基红外系统"高轨卫星构型分析与启示[J]. 太空探索, 2022（4）: 104-110.

[3] 雷萍, 邢晖, 王娟锋, 等. 天基红外预警系统扫描相机预警探测能力研究[J]. 红外与激光工程, 2021, 50（9）: 977-985.

[4] 郭海韵. 基于Agent的导弹防御系统建模与火力分配算法研究[D]. 江苏:南京理工大学, 2020.

[5] 陶新龙, 金阳辉, 陆小科. 地球同步轨道卫星对低轨卫星的目标引导方法[J].指挥控制与仿真, 2024, 46（03）: 153-160.

[6] 杜广洋, 郑学合. 弹道导弹过关机点状态估计方法研究[J]. 现代防御技术, 2018, 46（06）: 24-29.

[7] 聂凯, 曾科军, 左玉东. 基于关机点状态的航天器落点计算及精度分析[J]. 航天控制, 2018, 36（04）: 34-38+45. DOI: 10.16804/j.cnki.issn1006-3242.2018.04.007.

[8] 陈航, 吴建波. 信息化联合作战侦察监视装备发展建设思考[J]. 科技创新与应用, 2021, 11（25）: 73-75.

[9] 王洁, 黄伟, 张剑勇, 等. 航天验证器运动参数摄影测量方法[J]. 光学学报, 2021, 41（03）: 78-88.

[10] 黄良鹏. 基于多视觉传感器信息融合的位姿估计[D]. 浙江工业大学, 2019. DOI: 10.27463/d.cnki.gzgyu.2019.000171.

[11] 周爱国, 于江洋, 蒲家坤, 等. 基于加权Levenberg-Marquardt的多目视觉同名物点定位算法[J]. 激光与光电子学进展, 2021, 58（24）: 351-359.

[12] 倪晓东. 基于空间几何原理的空间前方交会改进算法[J]. 测绘工程, 2016, 25（06）: 33-36. DOI: 10.19349/j.cnki.issn1006-7949.2016.06.008.

[13] Guo Y B, Chen G, Ye D. Researches on binocular vision pose measurement with selected feature points [J]. Optik,2016,127（19）:8090-8095.

[14] He Q, Hu C, Liu W, et al. The fast method of point positioning in multi-vision system[C]∥2010IEEE International Conference on Robotics and Biomimetics, December 14-18, 2010, Tianjin, China. New York：IEEEPress,2010:1463-1467.

[15] He Q, Liu W, Wei N, et al. A methodfor3D-point reconstruction in multi-vision based on perpendicular foot[C]∥20119th World Congress on Intelligent Control and Automation, June21-25,2011,Taipei, Taiwan, China. New York：IEEEPress,2011:147- 151.

[16] 乔玉晶, 高胜彪, 皮彦超. 多目立体视觉测量网络节点的位姿优化方法[J]. 激光与光电子学进展, 2019, 56（05）: 186-193.

[17] 吴贤权, 尹仕斌, 任永杰, 等.多目视觉定向天线位姿测量[J]. 自动化与仪器仪表, 2019, （05）:1-6. DOI: 10.14016/j.cnki.1001-9227.2019.05.001.

[18] Liu L, Xi Z H, Sun Q. Multi-vision tracking and collaboration based on spatial particle filter[J]. Journal of Visual Communication and Image Representation,2019,59:316-326.

[19] 李瑾. 多目立体视觉的飞机零件三维重建和三维分割研究与应用[D].北京工业大学, 2019. DOI: 10.26935/d.cnki.gbjgu.2019.000358.

[20] 盛卫东, 黄源, 马超. 基于可行性矩阵遍历的重叠多目标角轨迹关联算法[J]. 信号处理, 2022, 38（08）: 1758-1765. DOI: 10.16798/j.issn.1003-0530.2022.08.021.

[21] 杨树亮, 毕硕本, 黄铜, 等. 一种出租车载客轨迹空间聚类方法[J]. 计算机工程与应用, 2018, 54（14）:249-255.

[22] Gong X R, Huang Z, Wang Y L, et al. High-performance spatiotemporal trajectory matching across heterogeneous data sources[J]. Future Generation Computer Systems, 2020, 105: 148-161.

[23] 周开来, 孟庆磊, 冯鑫伟. 轨迹相似性度量方法研究新进展[J]. 现代信息科技, 2023, 7（23）: 99-105. DOI: 10.19850/j.cnki.2096-4706.2023.23.021.

[24] 魏迪, 曾海彬, 洪锋, 等. 基于LSTM网络和特征融合的通信干扰识别[J]. 电讯技术, 2022,62（4）:450-456.

[25] Devlin J, Chang M W, Lee K, et al. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding[C]//Proceedings of NAACL-HLT Minneapolis, Minnesota: Association for Computational Linguistics,2019: 4171-4186.

[26] Liu Y H, OTT M, Goyal N, et al. Roberta: A robustly optimized bert pretraining approach[C]//International Conference on Learning Representations. Addis Ababa, Ethiopia: OpenReview,2020:317-333.

[27] Lan Z Z, Chen M D, Goodman S, et al. ALBERT: A Lite BERT for Self-supervised Learning of Language Representations[C]//International Conference on Learning Representations. Addis Ababa, Ethiopia Open Review ,2020:301-318.

[28] 游思思. 基于时空信息嵌入的多目标跟踪方法研究[D]. 合肥工业大学, 2022. DOI: 10.27101/d.cnki.ghfgu.2022.000077.

[29] 彭宝. 基于机器视觉辅助驾驶系统中行人实时检测跟踪研究[D]. 东华大学, 2016.

[30] Wang L, Pham N T, Ng T T, et al. Learning deep features for multiple object tracking by using a multi-task learning strategy[C]//Proceedings of the IEEE International Conference on Image Processing. IEEE, 2014: 838-842.

[31] Yu F, Li W, Li Q, et al. POI: multiple object tracking with high performance detection and appearance feature[C]//Proceedings of the European conference on computer vision Workshop: volume 9914. IEEE, 2016: 36-42.

[32] Chen J, Sheng H, Zhang Y, et al. Enhancing detection model for multiple hypothesis tracking [C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. IEEE, 2017: 18-27.

[33] Girshick R, Donahue J, Darrell T, et al. Region-based convolutional networks for accurate object detection and segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 38（1）: 142-158.

[34] Wojke N, Bewley A, Paulus D. Simple online and real time tracking with a deep association metric[C]//Proceedings of the IEEE International Conference on Image Processing. IEEE, 2017: 3645-3649.

[35] Bewley A, Ge Z, Ott L, et al. Simple online and real time tracking[C]//Proceedings of the IEEE International Conference on Image Processing. IEEE, 2016: 3464-3468.

[36] Zagoruyko S, Komodakis N. Wide residual networks[J]. arXiv preprint arXiv:1605.07146, 2016.

[37] Mahmoudi N, Ahadi S M, Rahmati M. Multi-target tracking using cnn-based features: Cnnmtt [J]. Multimedia Tools and Applications, 2019, 78（6）: 7077-7096.

[38] Bae S H, Yoon K J. Confidence-based data association and discriminative deep appearance learning for robust online multi-object tracking[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 40（3）: 595-610.

[39] Kim M, Alletto S, Rigazio L. Similarity mapping with enhanced siamese network for multi-object tracking[J]. arXiv preprint arXiv:1609.09156, 2016.

[40] Wang F, Zuo W, Liang L, et al. Joint learning of single-image and cross-image representations for person re-identification[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2016: 1288-1296.

[41] Zhang S, Gong Y, Huang J B, et al. Tracking persons-of-interest via adaptive discriminative features[C]//Proceedings of the European Conference on Computer Vision. Springer, 2016: 415-433.

[42] Son J, Baek M, Cho M, et al. Multi-object tracking with quadruplet convolutional neural networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017: 5620-5629.

[43] Lee S, Kim E. Multiple object tracking via feature pyramid siamese networks[J]. IEEE access, 2019, 7: 8181-8194.

[44] Kalman R E. A new approach to linear filtering and prediction problems[J]. ASME Journal of Basic Engineering, 1960, 82（1）: 35-45.

[45] Bergmann P, Meinhardt T, Leal-Taixe L. Tracking without bells and whistles[C]//Proceedings of the IEEE International Conference on Computer Vision. IEEE, 2019: 941-951.

[46] Yang B, Nevatia R．Multi-Target Tracking by Online Learning of Non-linear Motion Patterns and Robust Appearance Models[C]//IEEE Conference on Computer Vision and Pattern Recognition.2012

[47] Dicle C, Camps O I, Sznaier M. The way they move: Tracking multiple targets with similar appearance[C]//Proceedings of the IEEE International Conference on Computer Vision. IEEE, 2013: 2304-2311.

[48] Sadeghian A, Alahi A, Savarese S. Tracking the untrackable: Learning to track multiple cues with long-term dependencies[C]//Proceedings of the IEEE International Conference on Computer Vision. IEEE, 2017: 300-311.

[49] 刘予怀, 黄文德, 张晓飞. 基于几何分析法的低轨星座设计[J]. 大地测量与地球动力学, 2023, 43（10）: 1003-1007. DOI: 10.14075/j.jgg.2023.10.003.

[50] 肖广瀚, 胡泽岩, 刘军虎, 等. 空间目标偶数重连续覆盖星座设计方法[J]. 航空学报, 2024, 45（14）: 290-304.

[51] 牛立强, 谢拥军, 张春刚, 等. 对抗环境下宙斯盾系统探测弹道导弹的仿真[J]. 系统工程与电子技术, 2019, 41（06）:1195-1201.

[52] 赵蒙, 王明宇, 王健, 等. 基于运动方程的弹道导弹建模仿真方法[J]. 兵器装备工程学报, 2022,43（12）:118-124.

[53] 陈出新. 弹道导弹跟踪方法和算法研究[D]. 西北工业大学, 2014.

[54] 杨丽, 吴雨茜, 王俊丽, 等. 循环神经网络研究综述[J]. 计算机应用, 2018, 38(S2):1-6+26.

[55] 程冬梅. LSTM 研究现状综述[J]. 信息系统工程, 2022, 4（01）:149-152.

[56] 熊晗. 图神经网络的开发与应用研究[J]. 电视技术, 2022（004）:046.

[57] Kipf T N , Welling M .Semi-Supervised Classification with Graph Convolutional Networks[J]. 2016.DOI:10.48550/arXiv.1609.02907.

[58] 李铿, 邱欢, 张维娟, 等. 给定点连通度的图的补图的无符号拉普拉斯谱半径[J]. 新疆师范大学学报（自然科学版）, 2024, 43（03）: 64-68. DOI: 10.14100/j.cnki.1008-9659.20240408.001.

[59] 黄小平. 粒子滤波原理及应用: MATLAB 仿真[M]. 北京：电子工业出版社，2017.

[60] 张迪, 王延春. 天基红外系统的干扰策略[J]. 科技资讯, 2008（34）:12.

[61] 肖晖, 朱崇瑞, 刘昕圻, 等. 基于时间视线约束的无先验信息依赖非合作空间目标估计算法（英文）[J]. Transactions of Nanjing University of Aeronautics and Astronautics, 2024, 41(04): 526-540. DOI: 10.16356/j.1005-1120.2024.04.009.

[62] 周维. 预警面阵相机的像点定位误差补偿方法研究[D]. 南京航空航天大学, 2020. DOI: 10.27239/d.cnki.gnhhu.2020.000596.

[63] 高宏宇. 基于多片前方交会算法的倾斜无人机影像测图技术探讨[J]. 经纬天地, 2020, （06）: 79-82+86.

[64] Xiao H, Zhu C, Sheng Q, Wang B, Li J, Ling X, Wu F, Wu Z, Yu K. Parameter Precise Estimation Technology of Active Segment of Non-cooperative Targets Based on Long Short-Term Memory. Space Sci Technol. 2024;4:0150. DOI:10.34133/space.0150

[65] 张峰, 田康生. 弹道导弹预警仿真系统中弹道构造方法[J]. 火力与指挥控制, 2012, 37（03）:94-98+102.

[66] 李晓宇, 田康生, 郑玉军, 等. 基于关机点状态的弹道导弹落点估计及误差分析[J]. 舰船电子对抗, 2014, 37（05）: 71-74. DOI: 10.16426/j.cnki.jcdzdk.2014.05.002.

[67] 张腾, 牛青林, 柳云峰等. 弹道导弹助推段喷焰红外辐射特性数值计算[J/OL]. 兵工学报: 1-11[2023-11-02].

[68] 刘华军, 赖少发. 汽车毫米波雷达目标跟踪的快速平方根CKF算法[J]. 南京理工大学学报, 2016（1）: 6.DOI: 10.14177/j.cnki.32-1397n.2016.40.01.009.

[69] 李秋荣. 改进容积卡尔曼滤波及其导航应用研究[D]. 哈尔滨工程大学, 2015.

[70] 舒雨. 基于粒子滤波和相关滤波的目标跟踪算法研究[D]. 重庆大学, 2022.

[71] 任帅. 导弹外测数据高可用性实时处理技术研究与应用[D]. 哈尔滨工业大学, 2018.

[72] Paszke, Gross, Massa, et al.PyTorch:An Imperative Style, High-Performance Deep Learning Library[C]//Advances in Neural Information Processing Systems 32 （NIPS 2019）.

[73] Keyulu X, Wweihua H, Jure L, et al. How Powerful are Graph Neural Networks [C]//International Conference on Learning Representations（ICLR）, Louisiana, USA, 2019: 835-852.

[74] Nachiket D, Mohan M. Convolutional Social Pooling for Vehicle Trajectory Prediction[C]//IEEE Conference on Computer Vision and Pattern Recognition（CVPR）, California, USA, 2018: 1468-1476.

[75] Ce J, Zheng W, Cheng L, et al. Interaction-aware Kalman Neural Networks for Trajectory Prediction[C]//IEEE Intelligent Vehicles Symposium （IV）. Las Vegas, USA, 2020:1793-1800.

[76] Alahi G, Ramanathan, et al. Social LSTM:Human Trajectory Prediction in Crowded Spaces[C]//IEEE Conference on Computer Vision and Pattern Recognition（CVPR）, Las Vegas, USA, 2016: 961-971.

[77] Ziyan G, Zhanbo S. Modeling spatio-temporal interactions for vehicle trajectory prediction based on graph representation learning [C]//IEEE Intelligent Transportation Systems Conference （ITSC）, 2021: 1334-1339.