InSAR image phase filtering method and device, storage medium and terminal

Abstract

The present invention provides an InSAR image phase filtering method and device, storage medium and terminal, wherein the method comprises obtaining an InSAR image to be filtered, obtaining a phase value set, wherein the phase value set includes a plurality of different preset phase values; based on a probability density function and the InSAR image to be filtered, obtaining a fringe line diagram corresponding to each of the preset phase values; inputting all the fringe line diagrams into a fringe line denoising model corresponding to the fringe line diagram to obtain a corresponding denoised fringe line diagram; synthesizing based on all the denoised fringe line diagrams, obtaining a corresponding synthesized image as a filtered image of the InSAR image to be filtered. The present invention can effectively improve the robustness of InSAR image phase filtering.

Description

Technical Field

[0001] The invention belongs to the field of remote sensing technology and relates to an InSAR image phase filtering method, in particular to an InSAR image phase filtering method and device, storage medium and terminal. Background Art

[0002] Synthetic aperture radar interferometry (InSAR) has developed into a common remote sensing method, which is good at obtaining a variety of important geophysical parameters including terrain mapping and deformation monitoring. Interferometric phase measurements are usually affected by phase noise, which depends on the nature of the observed scattering mechanism, atmospheric conditions, acquisition geometry and sensor parameters. In addition, the temporal and spatial changes between two SLC (single look complex image) acquisitions will cause decorrelation, which will also affect the interferometric phase. Therefore, the coherence diagram of the interferogram is a key indicator that shows the reliability of the interferometric phase. The purpose of phase filtering is to remove phase noise in the interferogram, improve the phase unwrapping accuracy, and improve the terrain and deformation estimation results. InSAR phase filtering is an important step to ensure the correct application of InSAR technology, and choosing a suitable filtering method is crucial to obtain high-precision InSAR products.

[0003] In the prior art, the commonly used methods for InSAR phase filtering include mean filtering, median filtering, adaptive filtering, frequency domain filtering, wavelet domain filtering, and their improved versions. These methods usually construct filtering weights by using the proximity relationship of pixels in space, the size of the signal-to-noise ratio, the size of the coherence, etc., and their accuracy is limited. In addition, deep learning methods also show great potential in InSAR phase filtering. By combining the advantages of convolutional neural networks (CNN) and Transformer networks, the filtering effect can be effectively improved and the integrity of the interference fringe structure can be maintained. However, deep learning methods have the following problems: First, the filter window may cover multiple physical features, resulting in inconsistencies in the filtered phase map on different terrains or boundaries of objects, destroying the phase consistency and affecting the continuity and physical meaning of the phase; second, the design of the filter usually requires the selection of appropriate parameters, such as window size and shape. Under different terrain and noise conditions, the optimal parameters may be different, which makes parameter selection a challenge; third, because the filter smoothes the details in space, small-scale features become less obvious, so the spatial resolution of the image may be reduced.

[0004] Therefore, deep learning models perform well on training data but poorly on unseen data, especially when processing InSAR data with complex terrain or noise patterns, and therefore have poor robustness. At the same time, deep learning models are often considered to be "black boxes" and it is difficult to explain their internal working mechanisms and decision-making processes, which may cause difficulties in scientific analysis and results verification. Summary of the invention

[0005] The object of the present invention is to provide an InSAR image phase filtering method and device, a storage medium and a terminal, which are used to solve the technical problem of low robustness of the InSAR image phase filtering method in the prior art.

[0006] In a first aspect, the present invention provides an InSAR image phase filtering method, comprising:

[0007] Acquire an InSAR image to be filtered and acquire a phase value set, wherein the phase value set includes a plurality of different preset phase values;

[0008] Based on the probability density function and the InSAR image to be filtered, obtaining a fringe line diagram corresponding to each of the preset phase values;

[0009] Inputting all the fringe line images into a fringe line denoising model corresponding to the fringe line image to obtain a corresponding denoised fringe line image;

[0010] Synthesis is performed based on all the denoised fringe line images, and a corresponding synthesized image is obtained as a filtered image of the InSAR image to be filtered.

[0011] In one embodiment of the present invention, the stripe line denoising model is obtained by training a preset neural network model using a stripe line image data set; the stripe line image data set includes multiple groups of stripe line image data, and each group of the stripe line image data uses a noisy stripe line image as a prediction factor and a noise-free stripe line image as a prediction result.

[0012] In one embodiment of the present invention, the method for acquiring the fringe line diagram dataset includes:

[0013] Acquire a preset elevation data set, wherein the preset elevation data set includes a plurality of elevation data, and for each elevation data, generate a plurality of absolute phase maps corresponding to the elevation data based on a plurality of different elevation ambiguities;

[0014] Performing a twisting operation on each of the absolute phase images to obtain a corresponding interference image without noise;

[0015] Adding Gaussian noise to the interference pattern without noise to obtain a corresponding interference pattern with noise;

[0016] Acquire a plurality of noise-free fringe line graphs corresponding to each of the noise-free interference patterns based on a plurality of phase values, and acquire a plurality of noisy fringe line graphs corresponding to each of the noisy interference patterns based on a plurality of phase values;

[0017] A fringe line image without noise and a fringe line image with noise corresponding to the same phase value are used as a piece of training data to obtain a fringe line image data set.

[0018] In one embodiment of the present invention, the preset neural network model is Res-Unet.

[0019] In one embodiment of the present invention, based on a probability density function, obtaining a fringe line pattern corresponding to the preset phase value includes:

[0020] Substituting the preset phase value into a preset probability density function formula to obtain a corresponding fringe line diagram;

[0021] The preset probability density function formula is:

[0022]

[0023] in, represents the magnitude of the complex coherence coefficient, represents the phase value of the InSAR image to be filtered, Indicates the preset phase value.

[0024] In one embodiment of the present invention, synthesizing based on all the denoised fringe line images to obtain a corresponding synthesized image as a filtered image of the InSAR image to be filtered includes:

[0025] Obtaining the maximum likelihood estimation value of each pixel in all the stripe denoising results as the filtering result corresponding to the pixel;

[0026] The filtering results of all the pixels are collected to obtain a corresponding synthesized image as the filtered image of the InSAR image to be filtered.

[0027] In one embodiment of the present invention, all phase values ​​in the phase value set form an arithmetic progression.

[0028] In a second aspect, the present invention further provides an InSAR image phase filtering device, characterized in that it comprises:

[0029] A data acquisition module, used to acquire the InSAR image to be filtered and acquire a phase value set, wherein the phase value set includes a plurality of different preset phase values;

[0030] A phase decomposition module, used for acquiring a fringe line diagram corresponding to each of the preset phase values ​​based on a probability density function and the InSAR image to be filtered;

[0031] A fringe denoising module, used for inputting all the fringe line images into a fringe line denoising model corresponding to the fringe line image, so as to obtain a corresponding denoised fringe line image;

[0032] The filtering synthesis module is used to obtain corresponding images based on all the denoised fringe line images as filtered images of the InSAR image to be filtered.

[0033] In a third aspect, the present invention further provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the InSAR image phase filtering method as described above.

[0034] In a fourth aspect, the present invention further provides a terminal, comprising a processor and a memory, wherein the memory is communicatively connected to the processor;

[0035] The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes the InSAR image phase filtering method as described above.

[0036] As described above, the InSAR image phase filtering method and device, storage medium and terminal of the present invention have the following beneficial effects:

[0037] 1. This application changes the learning process from the conventional learning strategy of inputting the InSAR image to be filtered and outputting the filtered image to inputting the fringe line pattern (corresponding to the probability density function) and outputting the denoised fringe line pattern. From the perspective of computer vision and image processing, the fringe line features are more obvious, and the image recognition method based on deep learning is better at processing and extracting this type of data, thereby improving the accuracy.

[0038] 2. This application uses fringe line patterns with different preset phase values ​​as the object of denoising, and the biggest feature of InSAR images is the fringe feature. Therefore, a denoising model is constructed based on the essence of InSAR images, which enhances the interpretability of the model.

[0039] 3. This application uses a stripe line denoising model to remove the noise in the corresponding stripe line image in a personalized and targeted manner. Each stripe line denoising model serves the denoising process of a stripe line image, ensuring the fineness of the output stripe line image and effectively improving the robustness of the InSAR image phase filtering. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 The figure shows a flow chart of the InSAR image phase filtering method according to an embodiment of the present invention.

[0041] Figure 2 A schematic diagram of generating a fringe line diagram in the InSAR image phase filtering method according to an embodiment of the present invention is shown.

[0042] Figure 3A schematic diagram of the process of obtaining a fringe line diagram data set in the InSAR image phase filtering method according to an embodiment of the present invention is shown.

[0043] Figure 4 A comparison diagram of experimental effects of the InSAR image phase filtering method described in an embodiment of the present invention is shown.

[0044] Figure 5 The figure shows a schematic structural diagram of an InSAR image phase filtering device according to an embodiment of the present invention.

[0045] Figure 6 A schematic diagram of the structure of a terminal according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0046] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0047] The principles and implementation methods of the InSAR image phase filtering method and device, storage medium and terminal of this embodiment will be described in detail below, so that those skilled in the art can understand the InSAR image phase filtering method, device, storage medium and terminal of this embodiment without creative work.

[0048] In order to solve the above technical problems existing in the prior art, an embodiment of the present invention provides an InSAR image phase filtering method.

[0049] Figure 1 FIG. 1 shows a flow chart of the InSAR image phase filtering method according to an embodiment of the present invention, referring to FIG. Figure 1 As shown, the InSAR image phase filtering method according to the embodiment of the present invention mainly includes steps S100 to S400.

[0050] Step S100: obtaining an InSAR image to be filtered and a phase value set, wherein the phase value set includes a plurality of different preset phase values.

[0051] The InSAR image to be filtered is an interference image (interference phase map) generated from synthetic aperture radar (SAR) data, in which each pixel contains height information and noise information. Since InSAR images often contain atmospheric disturbances, scattering and phase noise, filtering is required to extract clearer terrain information. The InSAR image to be filtered is the InSAR image to be filtered. In the InSAR image phase filtering method described in the embodiment of the present invention, it is necessary to process based on multiple phase values, so a phase value set is obtained in advance, and the phase value set includes multiple different preset phase values, and the preset phase value serves as one of the data basis for subsequent processing.

[0052] Optionally, all preset phase values ​​in the phase value set form an arithmetic progression. The phase value set consists of a series of equally spaced preset phase values, and the difference between each preset phase value is fixed. This arithmetic progression can systematically cover the possible phase range. The preset phase value range is arrive , which can be divided into multiple phase values, such as ,in is the difference between adjacent preset phase values. Preferably, the difference is 0.0249, which means that the difference will be 252 discrete values ​​are divided equally as preset phase values ​​to form a phase value set. Optionally, the difference between adjacent preset phase values ​​is used as a hyperparameter for model training. The difference between adjacent preset phase values ​​directly determines the number of fringe line patterns that the model needs to process. Smaller differences require more fringe line patterns to be generated, increasing the diversity of training samples, thereby improving the generalization ability of the model, but also increasing the computational burden of the model. During the training process, by adjusting the difference between adjacent preset phase values, the model can better learn the noise characteristics of different fringe patterns, making the denoising effect more stable. In the model training, the difference between adjacent preset phase values ​​is used as a hyperparameter, and is adjusted through experiments to select the difference between adjacent preset phase values ​​that is optimal in a specific InSAR scene.

[0053] Step S200: Based on the probability density function and the InSAR image to be filtered, a fringe line pattern corresponding to each preset phase value is obtained.

[0054] The probability density function (PDF) is used to describe the noise distribution of the InSAR image to be filtered in order to identify the features and patterns at different phase values. In the prior art, the InSAR image to be filtered is usually directly used as the input data of the neural network for training, but the defects of this method are that, first, there is little complex terrain data in the training data set. Therefore, although the model performs well on the training data, it is difficult to process the InSAR image to be filtered with complex terrain or noise patterns; second, the terrain conditions of multiple noises in the InSAR image to be filtered are inconsistent. Using a neural network model for training may cause the filtered results to be inconsistent on different terrains or boundaries of objects, destroying the phase consistency and affecting the continuity and physical meaning of the phase. The embodiment of the present application obtains the fringe line diagram corresponding to the probability density function at each preset phase value through multiple preset phase values ​​as the feature to be trained and learned by the model. Each fringe line diagram represents the noise distribution of the InSAR image to be filtered at the corresponding preset phase value. Figure 2 FIG. 4 is a schematic diagram showing a fringe line diagram generated in the InSAR image phase filtering method according to an embodiment of the present invention, referring to FIG. Figure 2 As shown, the noisy interferogram (i.e., the InSAR image to be filtered) generates a corresponding fringe line diagram based on the preset phase value 0, and its corresponding fringe line diagram can also be obtained based on other preset phase values. These fringe line diagrams serve as the data basis for subsequent processing.

[0055] Optionally, based on the probability density function, obtaining a fringe line pattern corresponding to a preset phase value includes:

[0056] Substituting the preset phase value into the preset probability density function formula to obtain the corresponding fringe line diagram;

[0057] The default probability density function formula is:

[0058]

[0059] in, represents the magnitude of the complex coherence coefficient, represents the phase value of the InSAR image to be filtered, Represents a preset phase value. Specifically, the preset phase value is substituted into the above formula to obtain the probability value corresponding to the preset phase value, which represents the probability of the image under the preset phase value, and obtain the probability map corresponding to the probability value, that is, the fringe line map. It should be noted that the process of obtaining the fringe line map based on the above formula in the present application is to obtain the probability of the phase value of the pixel being the preset phase value based on each pixel point, and then plot the probability of the pixel points to form a fringe line map. For example, refer to Figure 2As shown, the fringe line diagram with a preset phase value of 0 is a diagram formed by the probability value when the phase value of each pixel point in the noisy interference diagram is 0. One pixel point corresponds to one probability value. The probability diagram of the probability values ​​of all pixel points is used as the fringe line diagram in this application. Preferably, the amplitude of the complex coherence coefficient is set in this application is 0.9.

[0060] Step S300: inputting all fringe line images into a fringe line denoising model corresponding to the fringe line image to obtain a corresponding denoised fringe line image.

[0061] Specifically, the embodiment of the present invention uses each stripe line graph as a unit for denoising processing, and inputs each stripe line graph into a corresponding stripe line denoising model. After the stripe line denoising model denoises the corresponding stripe line graph, it outputs the corresponding denoised stripe line graph. It should be noted that the preset phase value, the stripe line graph and the stripe line denoising model correspond to each other. Each specific preset phase value corresponds to a specific stripe line graph. To process the specific stripe line graph, it is necessary to use the stripe line denoising model corresponding to the preset phase value, that is, different stripe line graphs correspond to different stripe denoising models, and different stripe line graphs correspond to different preset phase values. Therefore, through the stripe line denoising model, the noise existing in the corresponding stripe line graph can be removed in a personalized and targeted manner. Each stripe line denoising model is applied to the denoising process of a stripe line graph, thereby ensuring the fineness of the output stripe line graph. Optionally, in order to reduce the training burden caused by multiple stripe line denoising models, in the embodiment of the present application, the loss function of the model training is the conventional mean absolute error (MAE) loss, the optimizer selects the Adam optimizer, the batch size is set to 32, the learning rate is set to 0.001, and 5% is randomly selected from the training data set as the verification data set during the training phase. 200 rounds of training are performed, and the output model finally determined corresponds to the round with the lowest verification MAE loss. Those skilled in the art can select other appropriate training parameters according to the actual hardware conditions to control the training burden within a reasonable numerical range.

[0062] Optionally, Figure 3 FIG. 1 shows a schematic diagram of a process for obtaining a fringe line diagram data set in the InSAR image phase filtering method according to an embodiment of the present invention, referring to FIG. Figure 3As shown, the stripe line denoising model is obtained by training the preset neural network model using the stripe line graph data set; the stripe line graph data set includes multiple groups of stripe line graph data, each group of stripe line graph data uses the noisy stripe line graph as a prediction factor and the stripe line graph without noise as a prediction result. Specifically, the stripe line denoising model is a neural network model, and its acquisition method is to train the preset neural network model based on the stripe line graph data set, the stripe line graph data set includes multiple groups of stripe line graph data, each group of stripe line graph data uses the noisy stripe line graph as a prediction factor and the stripe line graph without noise as a prediction result, and the stripe line graph without noise is the denoising result corresponding to the stripe line graph with noise, that is, the stripe line graph with noise is used as the input of the preset neural network model, and the stripe line graph without noise is used as the output of the preset neural network model. It should be noted that based on each specific preset phase value, the stripe line graph generated based on the preset phase value needs to be processed using the corresponding stripe line denoising model. For example, there are 252 preset phase values, and there are also 252 fringe line denoising models, and the two correspond one to one. The fringe line denoising model trained based on the preset neural network has the ability of automatic denoising, and can process fringe line images under different noise conditions, while maintaining the clarity and details of the fringe image, and improving the accuracy of fringe information.

[0063] Optionally, the method for acquiring the fringe line image data set includes the following steps:

[0064] Step S301: obtaining a preset elevation data set, wherein the preset elevation data set includes a plurality of elevation data, and for each elevation data, generating a plurality of absolute phase maps corresponding to the elevation data based on a plurality of different elevation ambiguities.

[0065] The preset elevation data set includes multiple different elevation data. For any elevation data, multiple absolute phase maps corresponding to this elevation data are generated based on multiple different elevation ambiguities (such as 50 meters, 100 meters, etc.), and each absolute phase map corresponds to an elevation ambiguity. The setting of different elevation ambiguities can simulate the noise conditions in the real environment, and the absolute phase map is directly calculated from the elevation data and is an unwrapped phase map. Optionally, the preset elevation data set is a global 30-meter SRTM elevation data set.

[0066] Step S302: performing a winding operation on each absolute phase image to obtain a corresponding interference image without noise.

[0067] Each absolute phase image is twisted to generate a noise-free interferogram. The noise-free interferogram is closer to the interferometric phase data actually observed by InSAR and is an important basis for the subsequent fringe line image generation.

[0068] Step S303: Adding Gaussian noise to the interference pattern without noise to obtain a corresponding interference pattern with noise.

[0069] In order to simulate the actual noise conditions, Gaussian noise is added to the interference pattern without noise to generate the corresponding interference pattern with noise. Different noise intensities can generate different noise scenes and enhance the diversity of data sets. The interference pattern with noise is used to simulate the noise effect of InSAR images in actual observations and provide noisy samples for model training. Optionally, the range of Gaussian noise is -5db to 3db. The lower the dB value, the greater the noise and the more obvious the impact on the image; the higher the dB value, the smaller the noise and the higher the image fidelity. The present invention sets the noise range of -5dB to 3dB to reflect the obvious noise characteristics in the image and retain the stripe structure in the image, ensuring that the model can identify stripe information in the noise, and ultimately improve the robustness of the stripe line denoising model.

[0070] Step S304: acquiring a plurality of noise-free fringe line graphs corresponding to each noise-free interference pattern based on a plurality of phase values, and acquiring a plurality of noisy fringe line graphs corresponding to each noisy interference pattern based on a plurality of phase values.

[0071] Specifically, for any interference pattern without noise, a corresponding noise-free fringe line graph can be generated using a phase value, so based on multiple phase values, multiple noise-free fringe line graphs corresponding to the interference pattern without noise are obtained; for any interference pattern with noise, a corresponding noise fringe line graph can be generated using a phase value, so based on multiple phase values, multiple noise fringe line graphs corresponding to the interference pattern with noise are obtained. For example, if there are x noise-free interference patterns, then the number of noisy interference patterns is also x, and based on y different phase values, a total of xy noise-free fringe line graphs and xy noise fringe line graphs can be obtained.

[0072] Step S305: using a fringe line pattern without noise and a fringe line pattern with noise corresponding to the same phase value as a piece of training data to obtain a fringe line pattern data set.

[0073] Pair the stripe line graph without noise and the stripe line graph with noise corresponding to the same phase value, and use them as a training data for model training. All paired data are combined into a stripe line graph data set, and each data in the data set contains a pair of samples, namely, a stripe line graph with noise (input) and a stripe line graph without noise (target output). The stripe line graph data set of the embodiment of the present invention is produced by a combination of different elevation ambiguities, phase values ​​and Gaussian noises. The data set can cover a variety of noise scenes, which helps to enhance the robustness of the stripe line denoising model. Through paired noisy stripe line graphs and stripe line graphs without noise, the stripe line denoising model can learn the noise pattern and stripe structure more accurately, providing a reliable data basis for subsequent denoising model training. It should be noted that, since there is a corresponding relationship between the preset phase value, the stripe line graph and the stripe line denoising model, the stripe line graph data set used for the stripe line denoising model training corresponding to the stripe line graph must meet the requirement that the data in the stripe line graph data set are all data produced based on the preset phase value (the preset phase value corresponding to the stripe line graph).

[0074] Optionally, the preset neural network model is Res-Unet. Res-Unet has a symmetrical encoding-decoding structure. The encoder extracts image features, while the decoder gradually restores the resolution of the image, and finally generates a high-quality denoised image. Through residual connections and jump connections, it not only improves the denoising effect, but also effectively retains the details of the fringe pattern. Therefore, in the denoising process, Res-Unet as a preset neural network model can reconstruct the fringe structure to the greatest extent, ensure the fringe clarity and phase consistency of the fringe line image, and provide high-quality denoising for subsequent InSAR filtering.

[0075] Step S400: synthesizing based on all the denoised fringe line images, and acquiring a corresponding synthesized image as a filtered image of the InSAR image to be filtered.

[0076] Specifically, after all the fringe line images are processed by the denoising model, corresponding denoised fringe line images are generated. These denoised fringe line images contain the denoised fringe structure information, reduce noise interference, and retain the details and phase characteristics of the fringe. At the same time, each denoised fringe line image corresponds to a specific phase value, so the collection of all denoised fringe line images can cover the entire phase range of the InSAR image. Synthesis is performed based on all the denoised fringe line images, and the synthesized image obtained after the synthesis is completed is the final filtered image. This image is the optimized processing result of the denoised fringe line image, which can minimize noise interference, retain the core fringe and phase information of InSAR, and has high accuracy and stability.

[0077] Optionally, the synthesis method is based on maximum likelihood estimation. Specifically, synthesis is performed based on all denoised fringe line images, and obtaining the corresponding synthesized image as the filtered image of the InSAR image to be filtered includes: obtaining the maximum likelihood estimation value of each pixel in all fringe denoising results as the filtering result corresponding to the pixel; and collecting the filtering results of all pixels to obtain the corresponding synthesized image as the filtered image of the InSAR image to be filtered. Specifically, the InSAR image to be filtered is composed of multiple pixels. For each pixel, all pixel values ​​from different denoised fringe line images are collected to form a to-be-selected set for the pixel, and the filtering result of each pixel is obtained from the to-be-selected set by the maximum likelihood estimation method. For any pixel, the to-be-selected set corresponding to the pixel is represented as a probability model, and the maximum likelihood estimation method is used based on the probability model to obtain the optimal value in the to-be-selected set as the filtering result corresponding to the pixel. Specifically, taking any pixel as an example, the probability corresponding to multiple preset phase values ​​of the pixel in the noisy interference map after the fringe line denoising model is obtained (that is, the denoised fringe line map). Each probability value represents the probability of a possible phase value of the pixel, and the preset phase value corresponding to the maximum probability value is obtained as the phase value of the pixel after filtering, that is, the filtering result corresponding to the maximum probability value is selected as the filtering result of the pixel. By calculating the maximum likelihood estimate of the pixel value and selecting the most likely phase value, the filtered image can be made closer to the real terrain phase. The filtered image of the InSAR image retains the clear structure of the fringe line map, making the stripes smoother and less noisy.

[0078] In addition, other appropriate synthesis methods can be selected according to actual needs. For example, based on the pixel-by-pixel averaging method, that is, for each pixel point, the pixel values ​​of the corresponding position in all denoised fringe line images are averaged. This method can integrate the fringe information at different phases and reduce the residual noise; based on the phase weighted synthesis method, that is, weighting the fringe information according to the phase value of each fringe line image, enhancing the stability of the fringe signal, so that the synthesized image is closer to the real terrain phase.

[0079] Figure 4 The experimental effect comparison diagram of the InSAR image phase filtering method according to the embodiment of the present invention is shown. Figure 4 As shown, Figure 4 The noise-free simulated interferogram in the mountainous area near Isolation Peak in (a) is used as an example for comparison. Figure 4 (b) is a noisy interferogram with a coherence coefficient of 0.7. Figure 4 (c) is to use the Goldstein filter in the prior art Figure 4 (b) The result of filtering, Figure 4(d) is to use the pi-Net filter in the prior art Figure 4 (b) The result of filtering, Figure 4 (e) using the method of the present invention to Figure 4 (b) Filtering results. As can be seen from the figure, the filtering results obtained using the method of the present invention have almost no phase discontinuity; and in terms of the mean square error (MSE) between the filtering results of different filtering methods and the noise-free interference pattern, the filtering result of the present invention has an MSE of 1.1, while the filtering result of the Goldstein filter has an MSE of 1.5, and the filtering result of the pi-Net filter has an MSE of 1.3, which shows that the filtering method of the present invention is significantly better than the other two filtering methods. Therefore, the filtering method of the present invention learns the stripe lines of the interference pattern, and does not require any window operation during the filtering process, so the stripe consistency of the obtained filtering result is better.

[0080] The protection scope of the InSAR image phase filtering method of the embodiment of the present invention is not limited to the execution order of the steps listed in this embodiment. All solutions implemented by adding, reducing or replacing steps in the prior art based on the principles of the present invention are included in the protection scope of the present invention.

[0081] The InSAR image phase filtering method of the embodiment of the present invention changes the learning process from the conventional learning strategy of inputting the InSAR image to be filtered and outputting the filtered image to inputting the stripe line diagram (corresponding to the probability density function) and outputting the denoised stripe line diagram. From the perspective of computer vision and image processing, the stripe line features are more obvious, and the image recognition method based on deep learning is better at processing and extracting such data, which improves the accuracy and enhances the model interpretability. In addition, the present application can remove the noise existing in the corresponding stripe line diagram in a personalized and targeted manner through the stripe line denoising model. Each stripe line denoising model serves the denoising process of a stripe line diagram, ensuring the fineness of the output stripe line diagram and effectively improving the robustness of the InSAR image phase filtering.

[0082] In order to solve the above technical problems existing in the prior art, an embodiment of the present invention further provides an InSAR image phase filtering device.

[0083] Figure 5 FIG. 1 shows a schematic diagram of the structure of the InSAR image phase filtering device according to an embodiment of the present invention, referring to FIG. Figure 5 As shown, the InSAR image phase filtering device according to the embodiment of the present invention includes a data acquisition module, a phase decomposition module, a fringe denoising module and a filtering synthesis module.

[0084] A data acquisition module is used to acquire the InSAR image to be filtered and acquire a phase value set, wherein the phase value set includes a plurality of different preset phase values;

[0085] A phase decomposition module is used to obtain a fringe line diagram corresponding to each preset phase value based on a probability density function and the InSAR image to be filtered;

[0086] A stripe denoising module, used for inputting all stripe line images into a stripe line denoising model corresponding to the stripe line image to obtain a corresponding denoised stripe line image;

[0087] The filtering synthesis module is used to obtain corresponding images based on all the denoised fringe line images as filtered images of the InSAR image to be filtered.

[0088] The InSAR image phase filtering device of the embodiment of the present invention changes the learning process from the conventional learning strategy of inputting the InSAR image to be filtered and outputting the filtered image to inputting the stripe line diagram (corresponding to the probability density function) and outputting the denoised stripe line diagram. From the perspective of computer vision and image processing, the stripe line features are more obvious, and the image recognition method based on deep learning is better at processing and extracting such data, which improves the accuracy and enhances the model interpretability. In addition, the present application can remove the noise existing in the corresponding stripe line diagram in a personalized and targeted manner through the stripe line denoising model. Each stripe line denoising model serves the denoising process of a stripe line diagram, ensuring the fineness of the output stripe line diagram and effectively improving the robustness of the InSAR image phase filtering.

[0089] In order to solve the above technical problems existing in the prior art, an embodiment of the present invention further provides a storage medium on which a computer program is stored, characterized in that when the program is executed by a processor, all steps of the InSAR image phase filtering method of the embodiment are implemented.

[0090] The specific steps of the InSAR image phase filtering method and the beneficial effects obtained by applying the readable storage medium provided by the embodiment of the present invention are the same as those in the above embodiment, and will not be described in detail here.

[0091] A person of ordinary skill in the art can understand that all or part of the steps in the method for implementing the above embodiment can be completed by instructing a processor through a program, and the program can be stored in a computer-readable storage medium, and the storage medium is a non-transitory medium, such as a random access memory, a read-only memory, a flash memory, a hard disk, a solid-state hard disk, a magnetic tape, a floppy disk, an optical disc, and any combination thereof. The above storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a digital video disc (DVD)), or a semiconductor medium (e.g., a solid-state disk (SSD)), etc.

[0092] In order to solve the above technical problems existing in the prior art, an embodiment of the present invention further provides a terminal. Figure 6 A schematic diagram of the structure of the terminal according to an embodiment of the present invention is shown, referring to Figure 6 As shown, the terminal of the embodiment of the present invention includes a processor and a memory, and the memory is communicatively connected to the processor; the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes all steps of the InSAR image phase filtering method of the above embodiment.

[0093] The specific steps of the InSAR image phase filtering method and the beneficial effects obtained by applying the terminal provided by the embodiment of the present invention are the same as those in the above embodiment, and will not be described in detail here.

[0094] It should be noted that the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage. Similarly, the processor may also be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0095] Although the embodiments disclosed in the present invention are as above, the contents described are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the protection scope of the present invention shall still be subject to the scope defined in the attached claims.

Claims

1. An InSAR image phase filtering method, comprising: Acquire an InSAR image to be filtered and acquire a phase value set, wherein the phase value set includes a plurality of different preset phase values; Based on the probability density function and the InSAR image to be filtered, obtaining a fringe line diagram corresponding to each of the preset phase values; Inputting all the fringe line images into a fringe line denoising model corresponding to the fringe line image to obtain a corresponding denoised fringe line image; Synthesis is performed based on all the denoised fringe line images, and a corresponding synthesized image is obtained as a filtered image of the InSAR image to be filtered.

2. The method according to claim 1, characterized in that The stripe line denoising model is obtained by training a preset neural network model using a stripe line image data set; the stripe line image data set includes multiple groups of stripe line image data, and each group of the stripe line image data uses a noisy stripe line image as a prediction factor and a non-noise stripe line image as a prediction result.

3. The method according to claim 2, characterized in that The method for obtaining the fringe line diagram data set includes: Acquire a preset elevation data set, wherein the preset elevation data set includes a plurality of elevation data, and for each elevation data, generate a plurality of absolute phase maps corresponding to the elevation data based on a plurality of different elevation ambiguities; Performing a twisting operation on each of the absolute phase images to obtain a corresponding interference image without noise; Adding Gaussian noise to the interference pattern without noise to obtain a corresponding interference pattern with noise; Acquire a plurality of noise-free fringe line graphs corresponding to each of the noise-free interference patterns based on a plurality of phase values, and acquire a plurality of noisy fringe line graphs corresponding to each of the noisy interference patterns based on a plurality of phase values; A fringe line image without noise and a fringe line image with noise corresponding to the same phase value are used as a piece of training data to obtain a fringe line image data set.

4. The method according to claim 2, characterized in that: The preset neural network model is Res-Unet.

5. The method according to claim 1, characterized in that Based on the probability density function, obtaining a fringe line pattern corresponding to the preset phase value includes: Substituting the preset phase value into a preset probability density function formula to obtain a corresponding fringe line diagram; The preset probability density function formula is: in, represents the magnitude of the complex coherence coefficient, represents the phase value of the InSAR image to be filtered, Indicates the preset phase value.

6. The method according to claim 1, characterized in that The step of synthesizing based on all the denoised fringe line images and obtaining a corresponding synthesized image as a filtered image of the InSAR image to be filtered includes: Obtaining the maximum likelihood estimation value of each pixel in all the stripe denoising results as the filtering result corresponding to the pixel; The filtering results of all the pixels are collected to obtain a corresponding synthesized image as the filtered image of the InSAR image to be filtered.

7. The method according to claim 1, characterized in that All preset phase values ​​in the phase value set form an arithmetic progression.

8. An InSAR image phase filtering device, characterized in that: include: A data acquisition module, used to acquire the InSAR image to be filtered and acquire a phase value set, wherein the phase value set includes a plurality of different preset phase values; A phase decomposition module, used for acquiring a fringe line diagram corresponding to each of the preset phase values ​​based on a probability density function and the InSAR image to be filtered; A fringe denoising module, used for inputting all the fringe line images into a fringe line denoising model corresponding to the fringe line image, so as to obtain a corresponding denoised fringe line image; The filtering synthesis module is used to obtain corresponding images based on all the denoised fringe line images as filtered images of the InSAR image to be filtered.

9. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the InSAR image phase filtering method according to any one of claims 1 to 7 is implemented.

10. A terminal, characterized in that: It comprises a processor and a memory, wherein the memory is communicatively connected to the processor; the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes the InSAR image phase filtering method according to any one of claims 1 to 7.

Images