基于松弛约束的卫星影像序列多特征地面无控协同定位

Due to accuracy limitation of satellite attitude and orbit control system (AOCS), the directly positioning accuracy of existing spaceborne optical sensors cannot fully meet the needs of practical applications, and precise geometric positioning without ground control data is one of the challenges of high-resolution satellite image processing. The project intends to study a co-location method of satellite image sequence without ground without control based on multi-feature and relaxation constraints, and the main contents include: 1) to study the multi-view matching method of multi-source satellite image sequence based on invariant features and phase information, and get multi-view tied points and tied straight lines; 2) to establish a straight line based geometric imaging model as well as a union block adjustment model based on multi-view tied points and tied straight lines, meanwhile to overcome the problem of weak intersection when performing block adjustment using nadir satellite images; 3) using the precise multi-view tied features obtained from satellite imagery sequences and the low-precision control points extracted from low-resolution reference images, to design a reliable block adjustment with relaxation constraints and compensate the errors of the orientation element of spaceborne sensor via multiple geometric constraints. The research of this project is expected to break the technical bottleneck of high-precision positioning of satellite images without ground control, and thereby improving the utilization of satellite images and efficiency of geometric processing, and promote the application of domestic satellites.

由于卫星姿轨控制系统精度的限制，现有星载光学传感器的直接定位精度还不能完全满足实际应用的需求，无地面控制资料的高精度几何定位是高分辨率卫星影像处理面临的难题之一。本项目拟研究一种基于卫星影像序列及多特征松弛约束的地面无控协同定位方法，主要研究内容包括：1）研究基于不变特征及相位信息的多源卫星影像序列多视匹配方法，获取多视同名点和同名直线；2）建立基于直线的几何成像模型及多视同名点和同名直线的联合平差模型，同时克服下视卫星影像区域网平差中的弱交会问题；3）利用从低分辨率的参考影像中提取的低精度控制点及从卫星影像序列中获取的精确多视同名特征，设计可靠的带松弛约束条件的区域网模型，通过多种几何约束补偿星载传感器方位元素的误差。本项目研究有望突破无地面控制条件下卫星影像高精度几何定位的技术瓶颈，从而提高卫星影像利用率与几何处理效率，推动国产卫星的应用水平。

针对国产卫星自主几何定位精度不足，而高精度、高分辨率参考影像难以获取的现状，项目提出在缺少精确地面控制资料的条件下，基于可公开获取的较低分辨率的参考影像以及重复观测的多源卫星遥感影像实现高分辨率卫星影像的高精度几何定位的思想，并建立了一套完整的基于卫星影像序列和中低分辨率参考影像的地面无控几何定位技术方案。项目提出一种新的带松弛约束条件的区域网平差方法，该方法并不直接最小化几何模型和控制点之间的误差，而是将低精度的控制点视为松弛的几何约束，仅保证几何模型能够以先验的精度满足控制点的几何约束，优化目标是将连接点的投影误差最小化。由于控制点是从较低分辨率的参考影像中提取得到，该方案对影像匹配精度的要求非常高，像素级乃至传统亚像素及的影像匹配方法都难以保证所提取控制点的精度。因此，项目以亚像素级的相位相关理论为基础，从提高相位相关平移估计精度、消除影像边缘效应克服大平移量、提高相似变换估计精度等方面展开深入研究，提出了基于高斯混合模型和低秩矩阵分解的遥感影像平移估计、通过周期分解消除影像傅里叶变换边缘效应、基于多极傅里叶变换的遥感影像相似变换估计、联合深度比较网络和相位相关的匹配等方法，实现了多尺度遥感影像的高精度（可达到1/10~1/100像素的精度）自动配准。项目在没有高精度控制点的条件下，借助中分辨率的Landsat-8/Sentinel-2等参考影像，通过多源卫星的重复观测，使国产高分辨率卫星影像达到了2个像素左右的几何定位精度。