| 摘要: |
| 针对复杂航迹场景下的航迹关联问题，提出了一种基于卷积神经网络（Convolutional Neural Network,CNN）和Transformer网络混合模型的端到端航迹关联算法，旨在提高航迹关联的准确性和效率。首先，将两个雷达上报的航迹集按目标分组，堆叠为特征对；随后，将标准化的数据输入至CNN-Transformer模型中，经由CNN层和Transformer编码器层提取空间特征并处理全局信息，之后进行目标关联预测；最后，根据预测概率确定最佳匹配目标输出关联结果。实验结果表明，无论是在存在干扰航迹环境还是在高密度航迹环境下，所提算法相比传统航迹关联算法在航迹关联精度上提高了15%~20%，并且在多目标场景中其鲁棒性和准确性大大提升。 |
| 关键词: 端到端航迹关联 深度学习 卷积神经网络 Transformer |
| DOI：10.20079/j.issn.1001-893x.241125004 |
|
| 基金项目:国防科技重点实验室基金项目（2023-JCJQ-LB-016） |
|
| End-to-End Track Association Based on CNN-Transformer in Complex Environments |
| PENG Ruihui,HE Jixian,SUN Dianxing,YANG Xueting |
| (1.Harbin Engineering University Qingdao Innovation and Development Base,Qingdao 266000,China;2.Insitute of Information Fusion,Naval Aeronautical University,Yantai 264001,China) |
| Abstract: |
| To address the challenges of track association in complex flight scenarios,an end-to-end track association algorithm based on a hybrid convolutional neural network(CNN) and Transformer network model is proposed to enhance the accuracy and efficiency of track association.Firstly,the track sets reported by two radars are grouped by target and stacked into feature pairs.Then,the normalized data is fed into the CNN-Transformer model.Spatial features are extracted and global information is processed through the CNN layers and Transformer encoder layers,followed by target association prediction.Finally,the best matching target is determined based on the prediction probabilities to output the association results.The experimental results show that the proposed algorithm improves the accuracy of track association by 15%~20% compared with traditional track association algorithms,whether in the presence of interference trajectory environments or high-density trajectory environments.Moreover,its robustness and accuracy are greatly improved in multi-target scenarios. |
| Key words: end-to-end track association deep learning convolutional neural network Transformer |
