基于K-means算法的时序GB-InSAR图像PS实时选择方法

GB-InSAR（Ground-based Interferometric Synthetic Aperture Radar，地基干涉合成孔径雷达）是一种新型的形变测量工具，应用于长期连续监测时，为了保证形变测量的准确性和实时性，通常需要将SAR图像进行分组处理，并基于PS（Permanent Scatterer，永久散射体）技术实现形变反演。对长时间SAR图像序列的分析结果表明，像素点的幅值会随时间发生较大的变化，导致采用固定的幅度离差门限来选择PS点时，各组PS的数量会发生极大改变，严重影响形变测量的精度。本文提出了一种基于K-Means（K均值）聚类的时序GB-InSAR图像PS实时选择方法，首先对时序图像进行分组，然后利用各组图像的像素点幅度和相关系数，基于K-Means算法进行两级聚类。对一处露天矿坑实测数据的分析结果表明，本文方法在保证了各组图像的PS点足够多的同时，对PS点数量的稳定性也有大幅提高。 

Real-time Selection Method of Time Series GB-InSAR Image PS Based on K-means Algorithm

GB-InSAR (Ground-based Interferometric Synthetic Aperture Radar) is a new type of deformation measurement tool. When applied to long-term continuous monitoring, in order to ensure the accuracy and real-time of the deformation measurement, it is usually necessary to group SAR images and realize deformation inversion based on PS (Permanent Scatterer) technology. The analysis results of the long-term SAR image sequence show that the amplitude of the pixel points will change greatly over time, resulting in the use of a fixed amplitude deviation threshold to select PS points, the number of PS points in each group will change greatly and seriously affect the accuracy of deformation measurement. This paper proposes a real-time selection method of time series GB-InSAR image PS based on K-Means (K-means) clustering. First, the time series images are grouped, and then the pixel amplitude and correlation coefficient of each group of images are used, based on K-Means The algorithm performs two-level clustering. The analysis results of the measured data of an open-pit mine show that the method in this paper ensures that the PS points of each group of images are enough, and the stability of the number of PS points is also greatly improved.