轨道误差对InSAR相位影响的BP神经网络去除法

针对用二次多项式法去除轨道误差对InSAR相位影响时,须对干涉相位其他项分布性质作假设,且自身存在一定的模型缺陷的问题,该文提出用BP神经网络去除轨道误差对InSAR相位影响的方法。研究表明:BP神经网络法在使用时无须对干涉相位其他项分布性质作假设,模型更优。模拟实验中,轨道误差相位拟合残差更小;真实数据实验中,纠正后非形变区相位集中在零值附近,且波动趋势更为平稳。该方法一定程度上降低了传统二次多项式法的应用局限性。

BP neural network method of removing orbital errors in InSAR interferogram

Aiming at the problem that when quadratic polynomial function is employed in removal of orbital errors which remained in differential interferogram,assumptions about the properties of the rest interferometric phase items must be assumed,meanwhile,there are defects exist in the mathematical model itself,back propagation(BP)neural network method was proposed in this paper to eliminate the effect caused by orbital errors.The validation test using both synthetic data and a real case study was pres ented.Research has shown that back propagation neural network could be used to eliminate the effect induced by orbital errors without hypothesis of distribution properties of interferometric phase items.It also had a better stability,which resulted in smaller fitting residual error.For the case study,the value of interferometric phase of non-deformation area fluctuates around zero after corrected,and the fluctuation trend was more stable.To some extent,back propagation neural network method improved the application limitation of the quadratic polynomial method when removing the effect caused by orbital errors.