容积卡尔曼滤波相位解缠方法相关问题研究

发布时间：2018-03-02 00:26

本文关键词： 相位解缠质量图多视处理容积卡尔曼滤波多参考点　出处：《中国矿业大学》2016年博士论文　论文类型：学位论文

【摘要】：合成孔径雷达干涉测量(In SAR)具有全天候、全天时、高效率获取目标区域数字高程模型(DEM)的能力,已广泛应用在地形测绘、地表形变监测、目标探测等领域。作为In SAR技术中的一个关键环节之一,相位解缠的成功与否直接关系到提取的目标高程信息的准确性,因此一直是In SAR技术应用研究的热点和难点。本文研究了适用于条纹密集、噪声水平高的干涉图的相位解缠方法及其影响该类算法解缠结果的关键问题,包括卡尔曼滤波相位解缠的非线性模型、噪声去除以及质量图选取对相位解缠结果的影响、质量不连续干涉图的解缠方法及一些改进技术和策略等。取得的主要研究成果如下:(1)对基于卡尔曼滤波的相位解缠方法进行了较为深入的研究。对解缠模型中的重要参数—相位梯度估计做了较详细的介绍并研究了影响相位梯度估计精度的因素,指出了梯度估计过程中较合理的参数设置方法。针对相位解缠模型中状态方程呈线性、观测方程呈非线性的特点,提出了一种简化的容积卡尔曼滤波(Cubature Kalman Filtering,CKF)相位解缠算法。与无迹卡尔曼滤波(Unscented Kalman Filter,UKF)算法相比,减少了状态预测步骤的计算量,简化了粒子选取时的参数设置。模拟和实测数据表明基于卡尔曼滤波的相位解缠方法在对条纹密度大噪声水平高的干涉图进行相位展开时具有明显优势。对扩展卡尔曼滤波(Extended Kalman Filter,EKF)、UKF、CKF相位解缠模型中非线性函数的估计精度进行了理论分析,指出当?(28)0时,UKF及CKF相位解缠精度相同;在相位质量较好的区域,EKF和CKF相位解缠的精度相当;在相位质量较差的区域,两者的精度表现出了一定的差异。实验结果符合理论分析结果。(2)研究了多视和预滤波处理对CKF相位解缠结果的影响。对于大噪声密条纹干涉图,采用多视和预滤波可以去除噪声提高信噪比,但容易损失条纹细节信息,而基于CKF的相位解缠方法本身具有去噪功能。基于这些特征,提出采用尽量小的视数,轻微的预滤波技术等策略来取得既保留条纹细节信息又可有效去除噪声的效果。实验表明,视数采用可以成功形成干涉图的视数,预滤波的程度采用远小于传统解缠方法要求的预滤波程度即可取得比较满意的解缠结果。(3)研究了质量图选取对CKF相位解缠结果的影响。通过对已有的五种指导相位解缠路径跟踪的质量图指标进行分析、比较、实验,得出同时考虑像元的相干信息和空间相似信息的Fisher Distance(FD)指标整体性能比较鲁棒,尤其对于噪声较高,地物覆盖类型复杂的实测数据形成的干涉图,FD指标引导的解缠结果具有明显优势。针对原有FD信息只考虑当前像元与周围八个像元的相似程度而忽略了周围相邻像元之间的相似性这一特点,提出了一种改进的FD质量图指标。该指标综合了当前像元与周围八个像元的所有相邻像元的相似性信息,理论上信息更全面。模拟数据和实测数据实验结果均表明改进的质量图可以更有效地避免解缠行为过早穿过噪声严重区域导致的过多像元相位展开精度下降。(4)提出了一种针对不连续质量图的多参考点相位解缠策略,建立了一整套多参考点选取方法,解决了参考点不足导致的高质量区却得到低精度解缠结果这一问题。该策略的解缠行为不是在各区域同时且独立地进行,而是从几个解缠区域的所有邻接像元中寻找质量最高的像元,依次只解缠一个像元。对于提出的参考点选取方法,首先依据区域划分理论把质量图划分为质量高和低两种等级,依据较少参考点的解缠顺序图,结合质量高低图,获得孤立的高质量区,然后在孤立的高质量区增选参考点,得到新的解缠顺序图,依此迭代,直到较大面积孤立的高质量区消失。模拟和实测数据均表明所提出的方法可行有效。
[Abstract]:Interferometric synthetic aperture radar (In SAR) has the advantages of all-weather, all day long, efficient access to regional digital elevation model (DEM) target ability, has been widely used in topographic surveying, deformation monitoring, field target detection. As one of the key links of In SAR technology, phase unwrapping and success it is directly related to the extraction of target elevation information accuracy, so it has been a hot and difficult research on Application of In SAR technology. In this paper is suitable for dense fringes, the interferogram noise level of the solution of the key problem of high unwrapping method and its influence the algorithm of phase unwrapping results, including nonlinear model Calman filtering phase unwrapping the noise removal and quality map selection effect on the wrapped phase, the quality of continuous interference phase unwrapping method of graphs and some improved techniques and strategies. The main research achievements are as follows: (1) The phase unwrapping method based on Calman filter is studied. The unwrapping is an important parameter in the model of phase gradient estimation are introduced in detail and studied the factors affecting accuracy estimation of phase gradient, pointed out that the gradient estimation parameters in the process of setting a reasonable method for phase unwrapping state. A linear equation model, the observation equation is nonlinear characteristics, this paper proposes a simplified volume filter (Cubature Kalman Calman Filtering, CKF) phase unwrapping algorithm and unscented filtering. Calman (Unscented Kalman Filter UKF) algorithm, reduces the computation state prediction step, simplifies the parameter selection of particles set up. Simulation and measured data show that the phase unwrapping based on Calman filter method in phase unwrapping of the interferogram fringe density at high noise level has significant advantages. The extended Calman filter (Extended Kalman, Filter, EKF, UKF, CKF) phase estimation accuracy bound nonlinear function model in the theoretical analysis, pointed out that when? (28) 0, UKF and CKF phase unwrapping accuracy in the same area; good quality of EKF phase, and CKF phase unwrapping accuracy is in poor quality; phase region, the accuracy of both showed some differences. The experimental results accord with the theoretical analysis results. (2) studied multi view and pre filtering effect on CKF phase. The bound for large dense noise interference fringe, using multi view and pre filtering can remove noise to improve the signal-to-noise ratio, but easy to lose stripe details, and CKF phase unwrapping method has denoising function. Based on these features, using as little as the number of pre filtering technology to achieve the strategy of minor is retained and there are fringe details Effect of noise removal effect. Experimental results show that the number of the interference can be successfully formed as visual map, pre filtering used far less than traditional unwrapping method requires pre filtering degree can obtain unwrapping results satisfactory. (3) studied the quality map selection effect on CKF phase. The unwrapped by five kinds of phase unwrapping guidance on existing index quality map path tracking analysis, comparison, experiment, obtained considering the pixel coherence information and spatial similarity information of Fisher Distance (FD) index of the overall performance is relatively robust, especially for high noise, interference pattern of land cover data type complex formation, winding it has the advantage of solution guided FD index. Based on the original FD information only consider the similarity of the current pixel and around eight pixels while ignoring the adjacent pixel similarity of this feature,. A kind of improved FD index. The index of comprehensive quality map similarity information of all the current pixel and the adjacent pixels around eight pixels, more comprehensive information theory. The simulation data and the experimental results show that the measured data quality map improvement can effectively avoid premature behavior through unwrapping area leads to serious noise the multi pixel phase unwrapping accuracy decreased. (4) proposed a multiple reference points for continuous quality map of phase unwrapping strategy, the establishment of a set of multiple reference point selection method, solve the high quality reference point has been low due to the lack of precision of the unwrapping results. The problem around the behavior of solutions this strategy is not in all regions simultaneously and independently, looking for the best quality of all adjacent pixels from the pixel but few unwrapping in regions that are only unwrapping a pixel. For the selected reference point. The first method, on the basis of the regional division theory of the quality of high and low quality graph was divided into two levels, according to the sequence diagram solution less wrapped around the reference point, combined with the quality of map, get high quality area isolated, and then co opted in isolated high quality reference point, get the unwrapping sequence diagram, this new iteration. High quality large area disappear until isolated. Simulated and measured data show that the proposed method is feasible and effective.

【学位授予单位】：中国矿业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：P237

【相似文献】