基于分层特征提取的无人艇多目标航迹关联算法OA
A Multi-Object Track Association Algorithm for Unmanned Surface Vehicles Based on Hierarchical Feature Extraction
针对搭载导航雷达和船舶自动识别系统(Automatic Identification System,AIS)的无人艇,感知水面舰船时的跨模态航迹关联问题,将航迹物理规律与神经网络特性相结合,提出基于航迹位置与运动信息的分层特征提取航迹关联方法.使用卷积神经网络(Convolutional Neural Network,CNN)提取位置特征,引入自注意力机制捕捉重要运动特征,在此基础上将两种特征融合后加入长短时记忆(Long Short-Term Memory,LSTM)网络实现航迹跨模态关联.实验结果表明,与传统基于深度学习的关联方法相比,所提方法在公开数据集上表现更优,在不大幅增加计算量的同时有效提高关联准确度.
Aiming at the cross-modal track association problem for unmanned surface vehicles e-quipped with navigation radar and an automatic identification system during surface ship percep-tion,this paper integrates the physical principles of tracks with the characteristics of neural net-works to propose a hierarchical feature extraction-based track association method utilizing both po-sitional and motion information.Specifically,convolutional neural networks are employed to ex-tract positional features,while a self-attention mechanism is introduced to capture critical motion features.The two types of features are then fused and fed into a long short-term memory network to achieve cross-modal track association.Experimental results demonstrate that the proposed method outperforms traditional deep learning-based association approaches on public datasets,ef-fectively improving association accuracy without significantly increasing computational overhead.
怀雪墁;郭一捷;任鹏宇;奔粤阳;李倩
哈尔滨工程大学智能科学与工程学院·哈尔滨·150001太原重工股份有限公司技术中心·太原·030032哈尔滨工程大学智能科学与工程学院·哈尔滨·150001哈尔滨工程大学智能科学与工程学院·哈尔滨·150001哈尔滨工程大学智能科学与工程学院·哈尔滨·150001
信息技术与安全科学
航迹关联雷达自动识别系统卷积神经网络自注意力机制长短时记忆网络
track associationradarAISCNNself-attention mechanismLSTM network
《飞控与探测》 2026 (4)
104-111,8
国家自然科学基金(52371368)
评论