基于控制线特征的高分辨率光学卫星影像精准几何纠正研究

High precision geometric rectification of high resolution satellite imagery is the basis of image mapping and diverse applications. With the improvements of satellite’s agile and mobile imaging ability, as well as image’s space resolution, the negative influences of image geometric deformation, radiation quality decrease, and local area occlusion, which caused by multi imaging mode and terrain variation, have been more pronounced. To solve the problem of high-cost control point layout and maintenance, identification and matching difficulties in traditional point-based geometric rectification models, this research presents a set of line-based rigorous geometric rectification schemes: First, the vector-feature-based strict affine and rational function models are constructed by improved generalized point strategy; To remove the correlation between parameters, the models are refined based on the difference matrix and the elimination method; For the image under large tilt angle imaging mode, the image vector error compensation is performed to remove the influence caused by imaging attitude and the elevation projection difference, then the resolution normalization processing is also performed; Finally, we provide the process design of model parameter pre-processing, overall adjustment solution and strict geometric correction. In this research, the experimental program which take advantage of a variety of high-resolution optical satellite images and the control line layout program are designed to verify the accuracy and reliability of the proposed methods under multi-imaging mode and complex terrain conditions. It will provide practical and effective technical support for precise and fast geometric rectification of high resolution optical satellite imagery.

高分辨率光学卫星影像的精准几何纠正是影像制图和多场景应用的基础。随着卫星敏捷机动成像能力和影像空间分辨率的提升，多成像模式和地形起伏等因素导致的影像几何形变、辐射质量下降和局部区域遮蔽等影响更加显著。本课题为解决传统点特征几何纠正模型存在的控制点布设和维护成本高、识别和匹配困难等问题，提出了一套基于控制线的严密几何纠正处理方案：首先采用改进的广义点策略构建矢量特征的严格仿射和有理函数模型；为去除参数相关性，基于离差阵和消去变换进行模型参数精化；对于大倾角成像模式的影像，进行像方矢量误差补偿去除成像姿态和高程投影差影响，并进行分辨率归一化处理；最终提供模型参数预处理、整体平差解算和严密几何纠正的完整流程。本课题还设计了多种高分辨率光学卫星影像实验方案和控制线布设方案，以验证本课题方法对多成像模式和复杂地形条件均具有较好的精度和可靠性，可为影像的精准快速几何纠正处理提供实用且有效的技术支撑。

为解决新型高分辨率光学成像卫星，在常规和敏捷机动成像模式下的高精度几何纠正问题，本课题充分发挥线特征在检测识别、语义约束和控制布设等方面的优势，创新性地构建了3种基于线特征的变换模型(LBTM)，并设计了一体化的影像几何纠正解决方案。该方案能够显著提高影像几何纠正的精度、可靠性和工作效率，大幅降低控制成本。本课题的研究内容主要包括6个方面：1）首先，基于线分割算子实现了影像上线特征的自动检测、提取和精确匹配，作为LBTM模型的控制单元；2）基于改进的广义点策略，构建了含卫星倾角和地形起伏改正的严格仿射LBTM模型、有理函数LBTM模型、含像点误差补偿的有理函数LBTM模型，并采用离差阵回归对模型参数进行精化；3）设计了基于控制线特征的影像严密几何纠正方法。针对大区域影像的几何纠正问题，提出了基于变分法的共轭梯度法区域网平差解算方法和基于改进不完全分解的平差预处理方法，提升模型解算的效率和稳定性；4）针对卫星的大倾角多成像模式，设计了含大倾角像方矢量误差补偿的LBTM模型优化方法，以提升模型精度；5）针对卫星平台存在的姿态高频颤振问题，分别提出了基于多光谱配准误差曲线和基于角位移测姿数据的高频颤振检测方法，进一步优化模型精度；6）提出了卫星影像内部和外部精度质量检验方法，用于准确评价影像几何纠正成果。本课题在算法研究的基础上，自主研发了基于线特征的高精度影像几何纠正软件，采用国产资源三号卫星、天绘一号卫星、遥感系列光学卫星，以及国外商用的IKONOS卫星、GeoEye-1卫星进行数据实验和比较分析。结果表明：各类LBTM模型均具有良好的精度和可靠性，严格仿射LBTM模型、有理函数LBTM模型和含像点误差补偿的有理函数LBTM模型的质量依次提升。引入大倾角成像误差补偿和平台高频颤振补偿，能够有效改善影像几何纠正的质量。布设数目充足和均匀分布的线特征，也有助于优化模型的控制水平。