| 摘要: |
| 针对密集场景下航迹关联中无法有效区分相似航迹的问题，提出了一种基于深度三元组网络的异步航迹关联算法。该算法是一种二阶段方法：在第一阶段，在利用长短时记忆网络提取航迹时序特征的基础上，采用基于三元损失的深度度量学习将航迹映射到度量空间，有效拉开正负样本特征距离，进而显著提升对相似目标的微小航迹特征的区分能力；在第二阶段，将关联问题转化为二分类问题，利用二分类器将航迹特征进行有效分类，实现异步航迹关联。仿真实验表明，在密集目标环境下，相较于传统方法，该方法平均正确关联率达到91.6%，比传统深度学习方法高6.9%。 |
| 关键词: 航迹关联 三元组网络 长短时记忆网络 深度度量学习 |
| DOI：10.20079/j.issn.1001-893x.250213004 |
|
| 基金项目: |
|
| An Asynchronous Track-to-Track Association(T2TA) Algorithm Based on Deep Triplet Network |
| WANG Xianyuan |
| (1.Southwest China Institute of Electronic Technology，Chengdu 610036，China; 2.National Key Laboratory of Complex Aviation System Simulation，Chengdu 610036，China) |
| Abstract: |
| To solve the problem that current track-to-track association(T2TA) algorithms cannot distinguish similar tracks in dense scenarios，an asynchronous T2TA method based on deep triplet network is proposed.This algorithm is a two-stage method.In the first stage，on the basis of extracting temporal features by long short-term memory(LSTM)，this algorithm utilizes deep metric learning with triplet loss to map tracks into a metric space.It can effectively increase the feature distance between positive and negative samples，thereby distinguishing the subtle feature of similar target.In the second stage，the association problem is taken as a binary classification task，and the track features are effectively classified by a binary classifier.Simulation experiments show that in dense target environment，the proposed method achieves an average correct association rate of 91.6%，which is 6.9% higher than that of traditional deep learning method. |
| Key words: track-to-track association triplet network long short-term memory deep metric learning |
《电讯技术》编辑部 