Journal of the Korea Institute of Military Science and Technology (한국군사과학기술학회지)

The Korea Institute of Military Science and Technology (한국군사과학기술학회)
권호 표지
DOI QR코드

DOI QR Code

Out-of-Distribution-Aware Contrastive Learning for Trajectory-Based Threat Identification

위협체 식별을 위한 궤적 기반 Out-of-Distribution 고려 대조 학습 기법

  • Su-Jeong Park (Department of Aerospace Engineering, KAIST) ;
  • Daegyeong Roh (Department of Aerospace Engineering, KAIST) ;
  • Minchae Kim (Department of Aerospace Engineering, KAIST) ;
  • Han-Lim Choi (Department of Aerospace Engineering, KAIST) ;
  • Chunghwan Kim (Land Combat System 1 Team, Hanwha Systems Co., Ltd.) ;
  • Mingi Kim (Intelligent Software Team, Hanwha System Co., Ltd.)
  • 박수정 (한국과학기술원 항공우주공학과) ;
  • 노대경 (한국과학기술원 항공우주공학과) ;
  • 김민채 (한국과학기술원 항공우주공학과) ;
  • 최한림 (한국과학기술원 항공우주공학과) ;
  • 김정환 (한화시스템(주) 지상시스템1팀) ;
  • 김민기 (한화시스템(주) 지능형SW팀)
  • Received : 2025.04.21
  • Accepted : 2025.06.10
  • Published : 2025.08.05
Download PDF
⟨ Previous Next ⟩

Abstract

This paper presents a trajectory-based prototype-guided contrastive learning method for threat target classification and out-of-distribution(OOD) detection. To address softmax classifiers' overconfidence and feature dispersion issues, our approach employs a GRU encoder to represent variable-length trajectory data and a projection head to transform these features for contrastive learning. Dual contrastive losses are applied at both the encoder and projection head levels to tighten intra-class representations and separate inter-class features. Furthermore, a nearest prototype classification scheme intrinsically detects OOD samples without external data. Experimental results on simulated radar trajectory datasets demonstrate that our method significantly outperforms conventional softmax-based models, especially in distinguishing challenging OOD samples with dynamics similar to in-distribution data.

Keywords

Acknowledgement

이 논문은 2023년도 정부(방위사업청)의 재원으로 국방기술진흥연구소의 지원을 받아 수행된 연구임(No. KRIT-CT-23-004, 복합형 능동방호 기술).

References

  1. D. Hendrycks and K. Gimpel, "A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks," arXiv preprint arXiv: 1610.02136, 2016.
  2. P. Khosla, et al., "Supervised Contrastive Learning," Advances in Neural Information Processing Systems (NeurIPS), Vol. 33, pp. 18661-18673, 2020.
  3. S. Seifi et al., "OOD-Aware Supervised Contrastive Learning," Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pp. 1956-1966, 2024.
  4. T. Phisannupawong, J. J. Damanik, and H. L. Choi, "Aircraft Trajectory Segmentation-based Contrastive Coding: A Framework for Self-supervised Trajectory Representation," arXiv preprint arXiv:2407.20028, 2024.
  5. S. Liang, Y. Li, and R. Srikant, "Enhancing the Reliability of Out-of-Distribution Image Detection in Neural Networks," arXiv preprint arXiv:1706.02690, 2017.
  6. D. Hendrycks, M. Mazeika, and T. Dietterich, "Deep Anomaly Detection with Outlier Exposure," arXiv preprint arXiv:1812.04606, 2018.
  7. R. Hadsell, S. Chopra and Y. LeCun, "Dimensionality Reduction by Learning an Invariant Mapping," Proc. IEEE Conf. on Computer Vision and Pattern Recognition(CVPR), Vol. 2, pp. 1735-1742, 2006.
  8. K. Q. Weinberger and L. K. Saul, "Distance Metric Learning for Large Margin Nearest Neighbor Classification," Journal of Machine Learning Research, Vol. 10, No. 2, pp. 207-244, 2009.
  9. K. Sohn, "Improved Deep Metric Learning with Multi-Class N-Pair Loss Objective," Advances in Neural Information Processing Systems(NeurIPS), Vol. 29, pp. 1857-1865, 2016.
  10. N. Frosst, N. Papernot and G. Hinton, "Analyzing and Improving Representations with the Soft Nearest Neighbor Loss," Proceedings of the International Conference on Machine Learning(ICML), pp. 2012-2020, 2019.
  11. K. Lee, et al., "A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks," arXiv preprint arXiv:1807.03888, 2018.
  12. R. Müller, S. Kornblith, and G. Hinton, "When Does Label Smoothing Help?," arXiv preprint arXiv: 1906.02629, 2019.
  13. G. Chen, P. Peng, X. Wang and Y. Tian, "Adversarial Reciprocal Points Learning for Open Set Recognition," IEEE Transactions Pattern Anal. Mach. Intell., Vol. 44, No. 11, pp. 8065-8081, 2021.
  14. F. Austin, G. Carbone, M. Falco, H. Hinz and M. Lewis, "Game Theory for Automated Maneuvering during Air-to-Air Combat," Journal of Guidance, Control, and Dynamics, Vol. 13, No. 6, pp. 1143-1149, 1990. https://doi.org/10.2514/3.20590
  15. W. B. Blake, Missile DATCOM User's Manual: 997 FORTRAN 90 Revision, Air Force Research Laboratory, U.S.A., 1998.
  16. R. Yanushevsky, Modern Missile Guidance, CRC Press, Boca Raton, FL, 2018.
  17. W. Jang, et al., "Vision-Based Optimal Landing Guidance Law for VTOL UAVs on a Moving Platform," International Journal of Aeronautical and Space Sciences, Vol. 25, No. 2, pp. 1-24, 2024. https://doi.org/10.1007/s42405-023-00646-9
  18. A. Bendale and T. E. Boult, "Towards Open Set Deep Networks," Proc. IEEE Conf. on Computer Vision and Pattern Recognition(CVPR), pp. 1563-1572, 2016.