摘要
多基线InSAR是在单基线InSAR的基础上发展一种获取高精度DEM的手段。相较于传统InSAR,多基线InSAR有许多技术上的优势。目前,基于马尔可夫场的最大后验(MRF-MAP)高程反演算法是众多多基线InSAR高程反演算法中精度较高、发展较为成熟的一种。此算法的关键步骤是模型超参数的估计,受制于模型的不完全性和问题的高维特性对超参数的估计只能采用expectation-maximization(EM)算法结合了蒙特卡洛计算方法来进行,因而抽样算法的收敛速度和鲁棒性就成为了影响MRF-MAP精度的关键因素。采用multiple-trial Metropolized independence sampler(MTMIS)抽样代替原始的Metropolis-Hastings抽样。实验结果表明这种改进使样本的收敛速度加快,样本多样性提高,因而能够改进最终获取的DEM的精度。此外,还推导出了超参数估计方差的表达式。
Multi-baseline InSAR is an effective extension of the traditional single baseline InSAR, which can provide high- precision DEM. Among all the reconstruction algorithms, Markov random fields based maximum a posteriori estimation （MRF-MAP） is a robust and precise method. The hyperparameter estimation is a key step of this method which will af- fect the final DEM accuracy. In practice, the estimation of the hyperparameter is achieved by EM algorithm combined with Monte Carlo methods. In this paper, we utilize the multiple-trial Metropolized independence sampler （MTMIS） in- stead of Metropolis-Hastings to realize Monte Carlo calculation. The experiment result validate that this improvement can improve the DEM accuracy finally. Moreover, we deduce the estimation variance of the EM steo.
出处
《国外电子测量技术》
2015年第7期55-61,共7页
Foreign Electronic Measurement Technology
