| 引用本文: | 刘 丛， 鲁彦希， 刘高正， 谭龙龙， 李 芳， 杨 磊. 雷达高度表延时多普勒匹配定位网络[J]. 雷达科学与技术, 2024, 22(4): 454-463.[点击复制] |
| LIU Cong, LU Yanxi, LIU Gaozheng, TAN Longlong, LI Fang, YANG Lei. Radar Altimeter Delay Doppler Map Matching and Positioning Network[J]. Radar Science and Technology, 2024, 22(4): 454-463.[点击复制] |
|
| 摘要: |
| 由合成孔径雷达高度表获得的延迟多普勒图像（Delay Doppler Map, DDM）因具有近垂直耦合的特点，易造成相邻地形DDM相似度高，进而导致图像匹配成功率低。针对该问题，本文提出了一种雷达高度表延时多普勒匹配定位网络（Delay Doppler Map Matching and Positioning Network, DDM?MPN），首次通过对DDM实测基准图像对进行匹配，实现飞行器的实时位置定位。首先，该方法采用浅层卷积神经网络架构，在保持较高匹配成功率的基础上降低了模型的复杂度和参数量，有效地节约了资源成本；其次，在传统图像匹配网络的基础上加入了坐标定位模块，使网络在完成图像匹配的同时实现对飞行器当前位置的三维坐标定位；最后，该方法通过对三元组损失函数、中心损失函数、交叉熵损失函数以加权的方式对网络参数进行协同优化，提高了匹配成功率和定位精度。模拟仿真实验的测试结果表明：沿航向水平面平均定位误差在30 m左右，垂直航向平均定位误差在60 m左右，三维总体平均定位误差在70 m左右，匹配成功率达到80%以上，在GPS拒止的情况下，该自主定位精度对大型飞行平台是可用的。 |
| 关键词: 合成孔径雷达高度表 神经网络 图像匹配 坐标定位 |
| DOI：DOI:10.3969/j.issn.1672-2337.2024.04.012 |
| 分类号:TN958；TN957.52 |
| 基金项目:国家自然科学基金（No.62271487） |
|
| Radar Altimeter Delay Doppler Map Matching and Positioning Network |
|
LIU Cong, LU Yanxi, LIU Gaozheng, TAN Longlong, LI Fang, YANG Lei
|
|
1. Tianjin Key Laboratory for Advanced Signal Processing, Civil Aviation University of China, Tianjin 300300, China;2. Institute of Electronic Engineering, China Academy of Engineering Physics, Mianyang 621000, China
|
| Abstract: |
| Due to the characteristics of near vertical coupling, the delay Doppler map （ DDM ） obtained from the synthetic aperture radar altimeter is easy to cause high similarity of DDM between adjacent terrains, which leads to low success rate of image matching. Aiming at this problem, a radar altimeter delay Doppler map matching and positioning network （DDM?MPN） is proposed, which for the first time realizes the real?time positioning of aircraft by matching the DDM measured reference image pair. Firstly, a shallow convolutional neural network architecture is employed, which can reduce the complexity and parameter number of the model while maintaining a high matching success rate, and effectively saves the resource costs. Secondly, a positioning module is added on the basis of the traditional image matching network, so that the network can realize the three?dimensional coordinate positioning of the current position of the aircraft while completing the image matching. Finally, the proposed method optimizes the network parameters in a weighted way by triplet loss function, center loss function and cross?entropy loss function, which improves the matching success rate and positioning accuracy. According to the test results of simulation experiments, the average positioning error in the horizontal plane is about 30 meters, the average positioning error in the vertical plane is about 60 meters, the overall average positioning error in 3D is about 70 meters, and the matching success rate is more than 80%. In the case of GPS rejection, the autonomous positioning accuracy is available for large flight platforms. |
| Key words: synthetic aperture radar altimeter neural network image matching coordinate positioning |
