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(Convo-lutional 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=O 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.
Low Signal-to-noise Ratio Interferometric Phase Unwrapping Method Based on Long Short-term Memory Convolution Network
In response to the problem of phase unwrapping in synthetic aperture radar(SAR)interfero-metric 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 prob-lem 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 spa-tial 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 large-scale 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.
NETL
NSTL
万方数据
