| 摘要: |
| 雷达对海面军民目标的精确分类识别，是海面态势信息生成的坚实基础，对维护海上安全具有重要意义。当前研究多基于航迹特征识别军民船，而现有方法或依赖单帧静态特征，难以应对军船伪装行为；或采用单一深度时序模型，易受噪声干扰。为解决上述问题，本文提出一种基于航迹特征原型选择的军民船双分支融合识别方法。本方法采用特征数据原型选择方法缓解类别不均衡问题，并构建静态分支与时空分支，分别处理瞬时特征与时序模式，其中时空分支设计一种自适应膨胀卷积时序卷积网络，能根据航迹局部平滑度动态调整感受野，高效捕获战术机动模式；最后采用自适应置信加权机制，结合证据融合机制，实现上下文感知的鲁棒融合。所提方法在真实航迹数据集上军船识别准确率达到95.2%。 |
| 关键词: 航迹特征 原型选择 军民船识别 时序卷积网络 双分支融合 |
| DOI： |
| 分类号:TN925.1 |
| 基金项目: |
|
| Dual-Branch Trajectory Feature Fusion for Target Identification of Military and Civilian Vessels |
|
|
|
|
| Abstract: |
| Accurate classification and recognition of maritime military and civilian targets through radar plays a foundational role in generating situational awareness of sea surfaces and holds significant importance for maritime security. Current research predominantly relies on trajectory features for ship classification, yet existing methods either depend on single-frame static features which struggle with military vessel camouflage behaviors, or employ single deep temporal models that are susceptible to noise interference. To address these challenges, this paper proposes a dual-branch fusion recognition framework based on trajectory feature prototype selection for military-civilian vessel discrimination. The method first employs a feature data prototype selection strategy to mitigate category imbalance issues, followed by constructing two parallel branches: a static branch for instantaneous feature processing and a spatiotemporal branch for temporal pattern analysis. Notably, the spatiotemporal branch incorporates an adaptive dilated convolutional temporal convolutional network that dynamically adjusts its receptive field according to local trajectory smoothness, enabling efficient capture of tactical maneuver patterns. Finally, an adaptive confidence weighting mechanism combined with Dempster-Shafer evidence theory is implemented to achieve context-aware robust fusion. Experimental results on real-world trajectory datasets demonstrate that the proposed approach achieves 95.2% accuracy in J-ship identification. |
| Key words: trajectory features prototype selection military-civilian vessel recognition temporal convolutional network dual-branch fusion |