一种优化的地基SAR时序反演非局部相干点选取方法

An Enhanced Nonlocal Coherent Pixel Selection Method for Time Series Inversion of GBSAR

地基合成孔径雷达(synthetic aperture radar,SAR)时间序列分析主要是从SAR影像中选取具备强散射特性并且相位保持稳定的相干点进行时序反演,然而不同干涉图之间的相干点是可变的,这使得时间序列分析变得复杂,因此选择大量高质量的相干点对确保地基SAR变形监测精度至关重要。为了提高地基SAR时间序列分析的可靠性,提出了一种优化的非局部相干点选取(enhanced nonlocal coherent pixels selection,E-NonPS)方法。从干涉图网络中选取有效的相干点,包括满秩相干点和部分相干点。满秩相干点通过非局部满秩法选取,部分相干点则首先筛选出矩阵的秩大于N/2的单视复数影像像素,然后根据振幅的均值和标准差设置一个信噪比阈值,选取信噪比大于该阈值的相干像素点。通过两组实验验证了E-NonPS方法在不同监测环境(高和低相干区域)中的有效性,并将选取的相干点结果与传统方法进行对比。结果表明,E-NonPS方法能够显著增加相干点的数量,并提高精度,特别是在低相干区域,可以有效解决选取相干点过少的问题,确保在时间序列分析中具有足够的相干点,进而提高地基SAR的监测精度。

Objectives:Ground-based synthetic aperture radar(SAR)time series analysis mainly selects coherent pixels with strong scattering characteristics and stable phase from SAR images for subsequent time series inversion.Coherent pixels between different interferograms are variable,which makes time series analysis complicated.Therefore,it is crucial to select a high number of coherent pixels that are of highquality to ensure precision in ground-based SAR deformation monitoring.Methods:To improve the reliability of time series analysis in ground-based SAR,this paper proposed the enhanced nonlocal coherent pixels selection(E-NonPS)method to select effective coherent pixels from the interferogram network,including full-rank coherent pixels and partially coherent pixels.Full-rank coherent pixels are selected by the nonlocal full-rank method,and partially coherent pixels are selected by setting the signal-to-noise ratio(SNR)threshold for pixels with coherence occurrences of more than N/2 in single look complex images.Results:We used E-NonPS method to test two ground-based SAR datasets to verify the effectiveness in different monitoring environments(e.g.,high and low coherence areas),and compared the selection results of coherent pixels with three traditional methods,including amplitude dispersion index(ADI)method,coherence coefficient(Coh)method and nonlocal full-rank coherent pixel selection(NonPS)method.The experimental results show that E-NonPS method can significantly increase the number of selected coherent pixels,and effectively improve the accuracy,especially for low coherence areas.Conclusions:The propoesed E-NonPS method can effectively address the demands for selecting coherent pixels and ensure sufficient coherent pixels in time series analysis.