基于深度学习的高频地波雷达特定目标跟踪方法研究

High-frequency surface wave radar (HFSWR) is an important means for continuous and active detection of moving targets within wide range at sea. Early research has shown that it is difficult to obtain accurate and reliable track only based on the HFSWR data, while prior knowledge can significantly improve the target tracking performance. And knowledge-based methods are becoming the development trend for HFSWR target tracking methods. This project focuses on the application scenario of specified target tracking using HFSWR, based on the idea of deep learning, the target track information obtained by synchronous assisted detection methods will be investigated deeply to establish the relationships with the track information provided by HFSWR, and reliable priori knowledge useful for assisting HFSWR independent tracking will be learned. And then adaptive motion model recognition methods as well as key parameters estimation methods in tracking algorithms will be developed with deep learning and optimization methods employed. The problems such as the motion model recognition during state prediction, parameters estimation both during data association and state filtering procedures will be well addressed for achieving continuous and accurate tracking of specified targets. Detection and tracking experiments of a cooperative target will be conducted integrating multiple tracking methods to validate the effectiveness and applicability of the proposed schemes. The specified target tracking methods proposed in this project will have important significance and reference value both for the research of tracking methods and for the applications of HFSWR target tracking systems.

高频地波雷达是海上运动目标大范围连续、主动探测的一种重要手段。前期研究表明，仅根据地波雷达自身的探测数据难以得到准确、稳定的目标航迹，而引入目标的先验知识可明显改善目标跟踪性能,成为地波雷达目标跟踪方法的发展趋势。本项目聚焦特定目标跟踪的具体应用，研究如何有效利用同步观测获取的目标信息提高地波雷达独立跟踪探测性能的方法。拟基于深度学习的思想，重点研究地波雷达与同步观测目标信息的特征及其关联关系，提出目标运动模型的自适应判别方法，并开展跟踪算法中相关参数的优化方法研究，解决状态预测时运动模型的判别以及数据关联与状态滤波过程中相关参数的估计等问题，实现特定目标的连续、准确跟踪。开展多手段合作目标跟踪探测实验，对方法的有效性及适用性进行验证。本项目提出的地波雷达特定目标跟踪方法，对目标跟踪方法的理论研究及地波雷达目标跟踪系统的业务化应用均具有重要意义及参考价值。

高频地波雷达作为对海上运动目标进行监视监测的重要手段，具有超视距、大范围、可全天候工作、成本低等优点，但同时存在探测精度低、检测概率低、数据率低、虚警率高等不足，给目标检测与跟踪带来很大挑战。目标航迹跟踪的性能由点迹质量与多目标跟踪方法的性能共同决定。基于此，本项目提出利用雷达自身以及多手段同步观测信息来改善目标点迹质量、提升目标跟踪方法性能，以提升地波雷达独立跟踪探测能力的思路。在点迹质量改善方面，提出了一种基于频率分集特性的双频地波雷达海上目标跟踪方法，解决了目标因海杂波遮挡导致的航迹断裂问题；建立了地波雷达与同步AIS航迹信息之间的关联关系，提出了一种基于AIS的地波雷达海上目标航迹校正方法，提高了地波雷达目标定位的准确性。在跟踪方法性能提升方面，提出了一种基于AIS的地波雷达目标跟踪中噪声方差估计方法，提高了滤波跟踪的精度；提出了一种基于航速航向估计的状态预测方法，能够在低方位探测精度下对目标下一时刻的状态进行准确预测，提高了数据正确关联的概率。同时，对SAR及船载光学目标探测手段进行了研究，提出了基于时频分析的ISAR成像方法、采用斜视滑动聚束模式的多角度SAR成像方法以及基于帧间差分的海天线检测方法。参与了多手段目标探测实验，验证了提出方法的有效性。项目研究成果已应用于研制的小阵列地波雷达目标探测系统中，开展了示范应用；发表论文9篇，其中SCI收录1篇，EI收录3篇，授权发明专利1项，申请发明专利1项，获得山东省科技进步一等奖1项，培养研究生3人。项目研究成果对目标跟踪方法的理论研究及地波雷达目标跟踪系统的业务化应用均具有重要意义及参考价值。