基于卷积神经网络的多偏移干涉相位滤波方法

Multi-shift interferometric phase filtering method based on convolutional neural network

为提升干涉信号处理中相位滤波的效果,提出了一种基于卷积神经网络的多偏移干涉相位滤波方法。利用干涉相位噪声模型解释了相位偏移原理,并根据相位偏移原理构建多个卷积神经网络去噪器,利用其分别对不同偏移的干涉相位进行滤波,生成多个去噪相位。利用神经网络计算像素权值,对多个去噪结果进行融合,进而获得质量更好的相位滤波结果。仿真的数据和真实的数据试验表明,相较于传统方法,所提方法具有更好的细节保持能力,并且所得结果的均方根误差和留数点数量更低。

To improve the performance of phase filtering in interferometric signal processing,a multi-shift interferometric phase filtering method based on convolutional neural networks is proposed.First,the phase shift principle is explained using the interferometric phase noise model.Then multiple convolutional neural network denoisers are built based on the phase shift principle and used to filter the interferometric phases with different shifts.Subsequently,a number of denoisers for convolutional neural networks are constructed using the notion of phase shift and employed to filter the interferometric phases with various shifts.Finally,multiple denoised phases are generated.The neural network is then used to calculate the pixel weights and fuse the multiple denoising results,resulting in a higher-quality result.The plenty denoising outputs are then fused and the pixel weights are calculated using the neural network to provide a higher-quality output.The simulated and real data experiments show that the proposed method retains more detail and has a lower root-mean-square error and the number of residues than the traditional methods.Experiments using both simulated and real data demonstrate that the suggested approach outperforms the conventional methods in terms of detail retention,root-mean-square error,and residue count.