InSAR interference image phase unwrapping method based on U-net

Abstract

The invention discloses an InSAR interference image phase unwrapping method based on U-net. The InSAR interference image phase unwrapping method comprises the following steps: S1, creating an InSAR analog data set; s2, creating a quasi-actual measurement data set; s3, putting the two types of data created in the S1 and the S2 into an improved U-net model for training; and S4, putting the to-be-unwound phase image into the trained U-net model to obtain an unwound real phase image. According to the invention, the U-net architecture, the ASPP network and the bottleneck residual error network arecombined, the expansion convolution features with different expansion rates are combined to capture rich context information, the feature receiving field can be expanded while the feature spatial resolution is not sacrificed, the feature information of the wrapping interferogram can be accurately obtained, and the robustness of the phase unwrapping algorithm is improved; the bottleneck residual error unit can prevent network degradation while reducing the parameter calculation amount of the network model, and improves the network training precision and efficiency. Compared with the prior art,the unwrapping precision is relatively high, and the anti-noise performance is relatively high.

Description

technical field [0001] The invention belongs to the field of image phase unwrapping, relates to InSAR interferometric image phase unwrapping, in particular to a U-net-based InSAR interferometric image phase unwrapping method. Background technique [0002] At present, phase unwrapping algorithms include path-tracking algorithms represented by branch-cutting and quality-guided algorithms, least-norm algorithms represented by least-squares algorithms, network planning algorithms represented by network flow methods, and Kalman-based algorithms. State estimation algorithms represented by filtering methods, deep learning algorithms represented by fully convolutional networks, etc. Path tracking algorithms use various strategies to define a suitable path, and integrate along this path to obtain its unwrapped phase, so as to minimize or avoid the error accumulation effect in the phase unwrapping process. The minimum norm algorithm first constructs a cost function of the difference ...

AI Technical Summary

Problems solved by technology

[0003] Path tracking algorithms, minimum norm algorithms, and network planning algorithms are susceptible to interference phase noise, and sometimes it is difficult to effectively unwrap noise interferograms, and both path tracking algorithms and network planning algorithms exist to a certain extent. It is difficult to balance the accuracy and efficiency of phase unwrapping; state estimation algorithms have strong anti-phase noise performance, and can usually effectively deal with the phase unwrapping problem of low signal-to-noise ratio interferograms, but time-consuming and costly; deep learning algorithms It has a certain degree of generalization, but the unwrapping accuracy of such algorithms is limited at present and cannot unwrap the measured terrain data

Images