融合LRT的Bootstrap区间估计同质点识别方法研究

Method of homogeneous pixels selection combining likelihood ratio test and Bootstrap interval estimation

针对传统同质点探测方法检验效率低以及在地物区分度不明显地区探测结果包含过多异质点的问题,提出融合似然比检验(LRT)与Bootstrap区间估计的同质点探测方法,通过模拟和实验对该方法进行验证,并对鄂尔多斯某煤矿地表形变进行探测和验证。研究结果表明:相比于传统方法,融合似然比检验(LRT)与Bootstrap区间估计的同质点探测方法具有较高的拒绝率和稳定的拒绝率标准差,比较适合地物区分度不明显的野外矿区同质点探测;在实验矿区发现1处明显下沉,最大累计沉降为-128.9 mm,通过与实测水准比较,发现实测水准结果与基于本文方法的DS-InSAR结果基本一致,2者均方根误差和平均绝对误差分别为14.4,12.0 mm。研究结果可实现对矿区的非接触、大范围、长时间监测,为DS-InSAR中同质点选取提供新思路。

Aiming at the problems of low testing efficiency of traditional homogeneous pixel selection methods and too many heterogeneous pixels in the detection results on the areas where the ground object differentiation is not obvious,a homogeneous pixels selection method combining the likelihood ratio test(LRT)and bootstrap interval estimation was proposed.The method was verified by the simulation and real experiments,and the surface deformation in a coal mine of Ordos was detected and verified.The results showed that compared with the traditional methods,the homogeneous pixels selection method combining the likelihood ratio test(LRT)and bootstrap interval estimation had high rejection rate and stable standard deviation of rejection rate,which made it suitable for the homogeneous pixels selection in the outdoor mining areas where the ground object differentiation was not obvious.An obvious subsidence was found in the experimental mining area,and the maximum cumulative subsidence was-128.9 mm.Compared with the measured level,it was found that the results of measured level were basically consistent with the results of DS-InSAR based on the proposed method in this paper,and the root mean square error and average absolute error of them were 14.4 mm and 12.0 mm,respectively.The results can realize the non-contact,large-scale and long-term monitoring of the mining area,and provide a new idea for the selection of homogeneous pixels in DS-InSAR.