顾及序列非平稳性的InSAR形变序列噪声分离与精化

Noise separation and refinement for InSAR deformation time series by considering sequence non-stationarity

地表形变序列提取是多时相合成孔径雷达干涉测量技术(Multi-temporal Interferometric Synthetic Aperture Radar, MT-InSAR)的最主要目的之一.受到随机噪声误差的影响, MT-InSAR获取的形变时间序列需要进一步精化.由于形变类型多种多样, 传统序列滤波处理方法容易出现单一模型不适用、滤波窗口难以确定、提取效果难以评估等问题, 难以实现海量监测目标的形变序列批量处理.针对这些问题, 本文提出了一种顾及序列非平稳性的广域InSAR形变序列噪声分离与精化方法.新方法构建了大范围区域的形变拟合模型库, 随后通过综合考虑拟合优度和残差平稳性, 对序列进行两次拟合, 充分分离出形变序列中的趋势性变化信号, 实现序列由非平稳向平稳的转化.最后对残余序列进行大窗口滤波, 实现对残余形变信号的提取, 最终完成含噪序列中的形变时间序列信号高精度分离与精化.本文利用传统三角滤波方法和新方法对规则形变序列和不规则形变序列进行了形变分量滤波提取对比实验, 实验结果表明, 相较于传统方法, 新方法对规则形变序列的形变分量提取精度较传统方法提高了40.69%, 不规则形变序列中提高了27.54%, 同时, 新方法也表现出了较为准确的模型识别能力.另外, 论文对新方法在滤波窗口选择、抗噪能力以及优化效果评价等方面的优越性进行了深入的分析和讨论.

Ground deformation time series monitoring is one of the most important tasks of multi-temporal interferometric synthetic aperture radar technique (MT-InSAR). The deformation time series obtained by MT-InSAR technique need to be further refined because of the effect of noise error. Due to the variety of deformation types, the traditional filtering methods are prone to the problems of the inapplicability of single model, difficult to determine the filtering window, difficult to evaluate filter accuracy, etc. In this paper, we propose a new method to separate noise and refine deformation time series in large-scale areas by considering sequence non-stationarity. We build a model library for large-scale area which contain common ground deformation model and then twice fitting process are performed by considering goodness of fit and time series stationarity to realize the transformation of sequence from non-stationary to stationary. Finally, a large-window filtering is used to extract residual deformation signal and the whole deformation time series can be obtained precisely. Comparison experiments for both regular deformation time series and irregular deformation time series were performed by using traditional triangular filter and new method. Experimental results show that there is an improvement of 40.69% for separation accuracy in regular deformation time series and an improvement of 27.54% in irregular deformation time series. Furthermore, the new method has more superiority in filter window selection, noise handling ability and accuracy assessment.