一种基于深度学习的多基线InSAR高程反演方法  

A multi-baseline InSAR elevation reconstruction method based on deep learning

作　　者：周宇翀 谢先明 ZHOU Yuchong;XIE Xianming(School of Automation,Guangxi University of Science and Technology,Liuzhou 545616,China;School of Electronic Engineering,Guangxi University of Science and Technology,Liuzhou 545006,China)

机构地区：[1]广西科技大学自动化学院,广西柳州545616 [2]广西科技大学电子工程学院,广西柳州545006

出　　处：《广西科技大学学报》2024年第2期65-77,共13页Journal of Guangxi University of Science and Technology

基　　金：国家自然科学基金项目(62161003,41661092);广西自然科学基金项目(2018GXNSFAA281196)资助。

摘　　要：提出一种深度学习与聚类分析算法结合的多基线合成孔径雷达干涉(interferometric synthetic aperture radar,InSAR)高程反演方法。该方法利用深度学习神经网络对干涉图截距信息进行分类,作为干涉像元类别属性判断的依据,精确获取干涉像元聚类中心,在此基础上利用聚类分析算法获得观测场景的高程信息。主要步骤如下:首先,获取干涉图截距信息,随后利用深度学习神经网络对干涉图截距信息进行分类,获得干涉像元类别属性。其次,对网络预测的同一类别像元截距取其平均作为该类别像元的聚类中心,有效避免传统算法因对噪声敏感造成的错误分类。最后,利用聚类分析算法对网络预测聚类中心进行后处理得到观测场景高程信息。模拟和实测实验数据表明,在不同信噪比的情况下,该方法对不同地形高程反演的均方根误差比传统CA算法更小,重建精度更高。A multi-baseline interferometric synthetic aperture radar(InSAR)elevation reconstruction method based on deep learning and cluster analysis is proposed.This method uses deep learning neural network to classify the intercept information of the interferogram as the basis for judging the category attributes of the pixels,and accurately obtains the clustering center of the pixels,and then cluster analysis technique is used to obtain the elevation information of the observed scenes.The main steps are as follows:firstly,the intercept information of interferograms is obtained,and then the deep learning neural network is used to classify the intercept information of the interferograms.Secondly,the intercepts of the pixels classified into the same category predicted by the network are averaged as the clustering center of this category,which effectively avoids the misclassification caused by the traditional technique due to poor noise robustness.Finally,the cluster analysis technique is used to obtain the elevation information of the observed scenes.The simulation and measured experiment results show that the root-mean-square error of the proposed method is smaller and the reconstruction accuracy is higher than that of the traditional CA algorithm under different SNR.

关 键 词：多基线INSAR 深度学习 聚类分析 

分 类 号：P224[天文地球—大地测量学与测量工程] V448.2[天文地球—测绘科学与技术]