基于LSTM联合卷积网络的低信噪比干涉相位解缠方法  

Low Signal-to-noise Ratio Interferometric Phase Unwrapping Method Based on Long Short-term Memory Convolution Network

作　　者：黄柏圣 刘婷[1] 杨金鹏 孙喆 吴雅琦 HUANG Bai-sheng;LIU Ting;YANG Jin-peng;SUN Zhe;WU Ya-qi(School of Electronics and Information Engineering,Nanjing University of Information Science and Technology,Nanjing 210044 China)

机构地区：[1]南京信息工程大学电子与信息工程学院,江苏南京210044

出　　处：《中国电子科学研究院学报》2024年第5期403-409,共7页Journal of China Academy of Electronics and Information Technology

基　　金：南京信息工程大学引进人才资助基金项目(21r036)。

摘　　要：针对合成孔径雷达(Synthetic Aperture Radar,SAR)干涉测量处理技术中的缠绕相位解缠的问题,提出了一种基于LSTM联合卷积网络的低信噪比干涉相位解缠方法。该方法采用复合损失函数对网络进行训练,并利用空间四向长短时记忆(Spatial Quadrature-Difference Long Short-Term Memory,SQD-LSTM)网络模块将相位解缠分解为一个回归问题,以避免典型卷积神经网络(Convolutional Neural Networks,CNN)难以学习局部空间特征参数的困难。实验结果表明,所提方法在低信噪比的情况下性能指标优于现有相位解缠方法,同时计算速度快,不需大规模训练数据集,相位解缠精度高,在SNR=0 dB时,归一化均方根误差达到了1.3%,提升了干涉测量的精准性。In response to the problem of phase unwrapping in synthetic aperture radar(SAR)interferometric measurement processing,a low signal-to-noise ratio(SNR)phase unwrapping method based on long short-term memory(LSTM)convolution network architecture is proposed.This method utilizes a composite loss function for network training and decomposes the phase unwrapping into a regression problem using the Spatial Quadrature-Difference Long Short-Term Memory(SQD-LSTM)network module.This approach avoids the difficulty of typical Convolutional Neural Networks(CNN)in learning local spatial feature parameters.The experimental results show that the proposed method outperforms existing phase unwrapping methods in low signal-to-noise ratio,with fast calculation speed,no need for largescale training datasets,and high phase unwrapping accuracy.When SNR=0 dB,the normalized root mean square error reaches 1.3%,improving the accuracy of interference measurement.

关 键 词：干涉相位图 相位解缠 低信噪比 长短时记忆网络 卷积神经网络 

分 类 号：TN957[电子电信—信号与信息处理]