利用推扫式卫星遥感图像进行像移探测的关键问题研究

Employing remote sensing images to obtain high-frequency image motion is one of the hot topics in the domain of image motion detection for push-broom satellites. Traditional method has two problems, on one hand, it only detect the motion at peak point, and other motions is hard to measure, on the other hand, it’s not allowed to change parameters during imaging for the sensors that provide images for motion detection, which will results in serious sensors’ imaging quality degradation in the case of large satellite attitude angle. This project studies on an algorithm that enable us to detect motions in a wider band with higher frequency resolution, and propose a way to minimize the impact of motion detection on imaging. First of all, we found that any two adjacent TDICCDs’ overlapping area will produce images that contain both low and high frequency motions, based on which we explore an algorithm that allow us to mine image motions from 0Hz to the predicted frequency of 300Hz using the overlapping images. Secondly, we found that the effect of image acquisition on imaging quality lies in satellite attitude and the location of TDICCD, based on the previous research, we study a selection algorithm of TDICCD for image motion measurement to make it clear how to select the TDICCD adaptively according to satellite attitude. Finally, the proposed mechanism will be verified by experiments. The new proposal in this project is expected to provide theoretical basis for improving the performance of image motion detection of push-broom satellites.

采用遥感图像测量像移是推扫式卫星像移探测领域的研究热点，而现有算法存在两个问题：第一，峰值频率点的像移可测、其他频率点的像移难以探测；第二，对于所成图像用于像移探测的成像传感器而言，成像参数的调整受到严格限制，其成像质量受到影响。本项目旨在探索一种能够对较宽频带内、各个频率点处像移全覆盖的像移探测算法，并试图将像移探测对像质的影响降到最低。首先，基于申请人对TDICCD拼接区图像中蕴含丰富像移信息的重要发现，研究基于拼接区图像的像移探测算法，预计能够对[0,300Hz]频带内的像移进行全覆盖探测。其次，申请人发现像移探测对像质的影响，取决于卫星姿态和TDICCD所处位置，提出用于像移探测的TDICCD自适应选择算法，根据卫星姿态变化，自适应地选择最优位置的TDICCD所成图像用于像移探测，将像移探测对像质的影响降到最低；最后，实验验证算法的可行性，为推扫式卫星像移探测性能的提升提供依据。

像移测量是实现像移补偿，进而削弱像移对像质影响的首要前提。然而现有像移测量算法存在两个问题：第一，测量带宽较低，随着空间相机焦距不断加长，中频和高频像移对像质的影响越来越严重，而目前尚缺少有效的中高频像移测量手段；第二，对于所成图像用于像移探测的成像传感器而言，成像参数的调整受到严格限制，其成像质量受到影响。本项目提出了基于成像传感器拼接区图像的推扫式卫星像移探测方法。利用工程参数计算像移初值低频分量。构建基于像移初值低频分量和成像位置差的双重约束函数，判定双重约束函数为凸函数后，利用共轭梯度优化算法，求解双重约束函数的最小值点，并以此作为像移初值最优估计。以若干行重叠图像为一个单元，利用相位相关和曲面拟合相结合的图像配准方法，得到两幅重叠图像中的成像位置差。构建成像位置差与像移之间的关系模型，进而利用像移初值和成像位置差估计像移。用于像移探测的成像传感器需工作在相同的行扫描时间下，导致成像质量下降，尤其是卫星存在较大姿态角的情况下。提出了用于像移探测的CCD自适应选择算法，能够根据卫星姿态变化，自适应地选择最优位置的传感器所成图像用于像移探测，将像移探测对像质的影响降到最低；最后，对算法的可探测的频率范围、时间范围和测量误差进行了理论分析，并通过数值仿真、卫星在轨拍摄遥感图像、地面振动成像实验三种方式进行测试，结果表明：关于可探测的频率范围，像移最高可测频率能够达到CCD行频的1/2。关于可探测的时间范围，算法能够对一次拍摄任务的全程像移进行全覆盖探测，包括无重叠图像的成像初期阶段。关于测量误差，当成像位置差的数据精度为1像元时，像移测量结果的均方根误差在垂轨方向可以达到1.3像元，在沿轨方向可以达到1.4像元。研究结果将为长焦距高分辨率空间相机的宽频带像移测量提供理论依据。