InSAR phase filtering method based on complex interferograms

Abstract

The invention discloses an InSAR (Interferometric Synthetic Aperture Radar) phase filtering method based on a plurality of interferograms, which comprises the following steps of: preprocessing acquired main and auxiliary SAR images to obtain the plurality of InSAR interferograms; dividing the plurality of InSAR interferograms into square interferogram patches along the azimuth direction and the distance direction; converting each interferogram patch from a spatial domain into a frequency domain signal, and constructing a corresponding frequency spectrum amplitude diagram; using a K-means clustering method to cluster and extract an initial low-pass filter in the spectrum amplitude diagram, and constructing a phase filter together with a Kaiser window function; the frequency domain signals of the interferogram plaques pass through a phase filter, the signal frequency band power is reserved, the noise at the high-frequency position is filtered out, meanwhile, the noise at the low-frequency position is filtered out through smoothing operation, and the filtered frequency domain signals are obtained; and the filtered frequency domain signals are converted to a spatial domain, filtered interferogram patches are obtained and spliced, and a filtered InSAR interferogram is obtained. The method is beneficial to popularization and promotion of the development of the current time sequence InSAR interferometry technology and the application in engineering.

Description

Technical Field

[0001] The present invention relates to the technical field of time-series data processing of synthetic aperture radar interferometry technology, and in particular to an InSAR phase filtering method based on complex interferograms. Background Art

[0002] Interferometric synthetic aperture radar (InSAR), an active microwave sensor, acquires both surface elevation and ground scattering information. The cloud-and-fog penetration capability of microwaves has enabled significant achievements and contributions in remote sensing and Earth observation. In particular, its interferometric phase measurement is considered a crucial feature in current topographic mapping and deformation monitoring. InSAR measurements are the difference between the phases of the same object observed from two slightly different locations, which inevitably introduces noise. Phase noise significantly impacts unwrapping accuracy, making phase denoising a necessary step in interferometric measurement processing. This noise can be broadly categorized into three components: 1) system noise, such as thermal noise and speckle noise; 2) noise caused by the observation scattering mechanism, including baseline, temporal, atmospheric, and volume decorrelation; and 3) noise caused by signal processing, primarily coregistration error. Phase noise in the interferogram is the result of the combined effects of one or more of these factors.

[0003] Since InSAR interferometric phase was first used to extract earthquake deformation fields in 1993, phase denoising has been a hot topic in InSAR processing research. The goal of phase denoising has always been to remove as much interferometric phase noise as possible while maintaining spatial resolution. These methods can be roughly divided into four categories: traditional spatial domain filters, non-local (NL) filters, transform domain filters, and filters based on new methods.

[0004] In general, Fourier transform-based frequency-domain filtering algorithms are simple in principle and easy to implement. However, existing research is limited to the Goldstein filtering algorithm framework and does not discuss the principles and functions behind the selection of the filter shape S{H(u,v)} in the Goldstein method. Instead, the goal is to improve filtering efficiency by modeling the filter factor. In practical applications, it is impossible to fully compensate for the problems caused by filter shape imperfections simply by selecting the filter factor, and this can reduce filtering efficiency. Ultimately, this manifests as some unremovable noise in the interferogram.

[0005] Therefore, it is necessary to design a new filter shape and filtering method based on the spectral difference between noise and signal in the frequency domain, and expand the Goldstein filter into a two-dimensional ideal low-pass filter; interferograms with strong noise signals no longer require multiple iterations to obtain better phase filtering results, thereby optimizing the filtering results while also simplifying the calculation process of the existing interferogram phase filtering method. Summary of the Invention

[0006] In view of this, the present invention analyzes the difference in spectral amplitude between noise and signal after Fourier transform, and explores a frequency-domain, computationally convenient, and universal complex interferogram-based InSAR phase filtering method that is not restricted by external conditions, thereby improving the deformation monitoring capability of time-series InSAR technology.

[0007] In order to achieve the above tasks, the present invention adopts the following technical solutions:

[0008] InSAR phase filtering methods based on complex interferograms include:

[0009] The acquired primary and secondary SAR images are preprocessed to obtain complex InSAR interferograms; the complex InSAR interferograms are divided into square interferogram patches along the azimuth and range directions;

[0010] Convert each interference pattern patch from the spatial domain to the frequency domain signal and construct the corresponding spectrum amplitude map;

[0011] The K-means clustering method is used to cluster and extract the initial low-pass filter in the spectrum amplitude map, and the phase filter is constructed together with the Kaiser window function;

[0012] The frequency domain signal of the interference pattern patch is passed through a phase filter. The signal frequency band power will be retained, the noise at the high frequency position will be filtered out, and the noise at the low frequency position will be filtered out by smoothing operation to obtain the filtered frequency domain signal.

[0013] Convert the filtered frequency domain signal to the spatial domain to obtain the filtered interference pattern patch;

[0014] The interference pattern patches are spliced ​​together to obtain the filtered InSAR interferogram.

[0015] Furthermore, the complex InSAR interferogram is slid in the azimuth and range directions with a preset step size to divide the interferogram into patches; there is a preset overlap between adjacent interferogram patches.

[0016] Furthermore, the K-means clustering method divides the spectrum amplitude graph into two categories, specifically:

[0017] First, the number of clusters is determined to be 2, and then two data points in the spectrum amplitude map are randomly selected as cluster centers;

[0018] All data points in the spectrum graph are associated and divided into clusters based on the center of the cluster closest to them; the cluster center is moved to the center of the current divided cluster containing all data points; this process is repeated until the number of iterations or the change in cluster center is less than the preset value, and the spectrum amplitude graph can finally be divided into two categories; the category with larger amplitude value is assigned 1, and the category with smaller amplitude value is assigned 0.

[0019] Furthermore, based on the clustering results of the spectrum amplitude graph, the geometric graph theory and statistical principles are combined to extract the shape with the largest area and closed geometry in the category assigned a value of 1. The geometric shape is used as the low-pass filter shape, and the pixel value within the shape is 1, indicating that the signal passes; the remaining pixels are assigned a value of 0, indicating that the signal is filtered out. The obtained geometric shape is used as the initial low-pass filter for adaptive extraction.

[0020] Furthermore, the Kaise window function is superimposed on the two-dimensional low-pass filter to obtain the final phase filter:

[0021] S{rect(|Z(u,v)|)}·K(n)

[0022] Where rect() represents the initial low-pass filter, S{·} represents Gaussian smoothing, Z(u,v) represents the frequency domain signal, u and v represent the horizontal and vertical frequency components during the two-dimensional Fourier transform, respectively, and K(n) represents the value of the two-dimensional Kaiser window function at time n.

[0023] Furthermore, in the Kaise window function, the control parameter β of the window function is set to 10; the size of the two-dimensional Kaiser window function is consistent with the size of the divided interference pattern patches.

[0024] Furthermore, the filtered frequency domain signal is expressed as:

[0025] H(u,v)=S{rect(|Z(u,v)|)}·K(n)·Z(u,v)

[0026] Where H(u,v) represents the frequency domain signal after filtering.

[0027] Furthermore, the filtered frequency domain signal is converted into the spatial domain using inverse fast Fourier transform;

[0028] The edge of each interference pattern patch obtained by the transformation in the spatial domain is discarded with a preset overlap, and only the remaining part is retained, which is then spliced ​​along the azimuth and range directions to obtain the filtered InSAR interferogram.

[0029] A terminal device comprises a processor, a memory and a computer program stored in the memory; when the processor executes the computer program, the InSAR phase filtering method based on complex interferogram is implemented.

[0030] A computer-readable storage medium stores a computer program; when the computer program is executed by a processor, the InSAR phase filtering method based on complex interferogram is implemented.

[0031] Compared with the prior art, the present invention has the following technical features:

[0032] 1. The complex interferogram-based InSAR phase filtering method of the present invention does not rely on any external data. Instead, it fully utilizes the differences in noise and signal power and frequency in the interferogram patch spectrum after Fourier transform to design a new filter. The key difference from the original Goldstein filter method and its improved series of methods lies in the design of the filter shape. The new filter changes the original frequency domain filter from being limited to amplifying or attenuating signals and noise to retaining signals or completely filtering out noise, resulting in better filtering effects. The K-means clustering method is used to extract the filter shape, which can obtain the adaptive filter shape and optimal filtering effect for each patch; the Kaiser window function can solve the Gibbs phenomenon in the Fourier transform.

[0033] 2. This invention not only effectively addresses one of the major obstacles currently limiting the accuracy of time-series InSAR technology, but also designs a filtering method from a frequency-domain perspective that better reflects the nature of noise generation in synthetic aperture radar imaging. Furthermore, the algorithm is simple and computationally efficient, requiring only the complex interferograms obtained after interference. The method is easy to implement and apply, contributing to the advancement and advancement of current time-series InSAR interferometry technology and its application in engineering. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic flow chart of the InSAR phase filtering method for complex interferograms provided by the present invention;

[0035] Figure 2 (a) is the spectrum signal after Fourier transform of the simulated interference pattern patch, including signal component and noise component; (b) is the shape of the designed low-pass filter, as shown by the blue solid line, through which only the signal s can be retained. x part;

[0036] Figure 3 Schematic diagram of the process of K-means clustering into two categories;

[0037] Figure 4 Schematic diagram of extracting low-pass filter shape on a two-dimensional spectrum image;

[0038] Figure 4 (a) is the two-dimensional spectrum image after Fourier transformation; (b) is the clustering result obtained after using K-means clustering on the two-dimensional spectrum image, and the image is divided into two categories; (c) is the low-pass filter shape extracted according to geometric principles after classification, and red represents the signal passing;

[0039] Figure 5 Kaiser window function used in the filter of the present invention;

[0040] Figure 5 (a) is a two-dimensional Kaiser window function; (b) is a one-dimensional Kaiser window function, which is the cross section at the center frequency of (a);

[0041] Figure 6 (a) is the simulated noise-free interferogram phase; (b) is the simulated noise-containing interferogram phase;

[0042] Figure 7 (a) is the low-pass filter shape extracted from the frequency spectrum of 16 interferogram patches uniformly selected from the simulated interferogram data along the azimuth and range directions; (b) is the InSAR phase map after filtering of 16 interferogram patches uniformly selected from the simulated interferogram data along the azimuth and range directions; (c) is the InSAR phase map after filtering of the simulated interferogram data; (d) is the scatter density map between the InSAR phase after filtering of the simulated interferogram data and the noise-free InSAR phase. DETAILED DESCRIPTION

[0043] The present invention provides an InSAR phase filtering method based on complex interferograms, which can be applied to the fields of InSAR height measurement, time-series deformation monitoring, etc.

[0044] See also Figure 1 , the present invention provides a flow chart of an InSAR phase filtering method based on complex interferogram; the method comprises the following steps:

[0045] Step 1: Preprocess the acquired primary and secondary SAR images, including registration, resampling, interferometry, removal of flat-earth phase, and phase correction of orbit errors caused by inaccurate orbit state data, to obtain complex InSAR interferograms of the primary and secondary SAR images.

[0046] The complex InSAR interferogram is divided into square interferogram patches with a step size of 8 pixels along the azimuth and range directions. Each interferogram patch contains N×N pixels, where N is 32. It is ensured that there is at least 75% overlap between each adjacent interferogram patch.

[0047] Step 2: Use two-dimensional fast Fourier transform to convert each interference pattern patch from the spatial domain to the frequency domain signal, and construct the corresponding spectrum amplitude diagram by taking the absolute value of the frequency domain signal.

[0048] like Figure 2 The orange solid line in (a) shows that the signal component and the noise component can be represented by s x and n means:

[0049] s=s x +n (1)

[0050] Where s is the frequency domain signal representation of the interference pattern patch signal.

[0051] There are two main differences between the signal component and the noise component in the spectrum amplitude diagram: 1) the amplitude of the signal component is generally greater than the amplitude of the noise component; 2) in the spectrum diagram, the signal component is always continuously located in the low-frequency area, that is, near the center frequency, while the noise component is located in the high-frequency area.

[0052] According to these two characteristics, the present invention designs an ideal low-pass filter shape rect(), such as Figure 2 As shown in (b), only the signal component s x ; In addition, since it is a two-dimensional image, the filter is also two-dimensional. Extracting the filter from the noise based on the spectrum can be transformed into a two-dimensional image classification problem.

[0053] Step 3, using the K-means clustering method to cluster and extract the initial low-pass filter in the spectrum amplitude diagram, and together with the Kaiser window function, forming the phase filter proposed by the present invention; specifically including:

[0054] Step 31, based on the spectrum amplitude image obtained by fast Fourier transform of each interference pattern patch dimension, in order to extract the boundary between the signal component and the noise component in the spectrum amplitude image and thus obtain the low-pass filter shape rect(), the K-means clustering method is used to cluster in the spectrum amplitude image, divide the image into two categories, and extract the low-pass filter.

[0055] The K-means algorithm sets each cluster to be identified by the cluster center of each class, and formalizes the clustering problem as a problem of minimizing the distance between sample points and cluster centers. It realizes the geometric division of samples through alternating iterations of two stages: sample point allocation and cluster center update. In the sample point allocation stage, the cluster to which the data point belongs is determined based on the distance calculation results from all data points to the cluster center. In the cluster center update stage, the average value of all data points in the class is set as the new cluster center.

[0056] The K-means algorithm flow chart is as follows Figure 3 As shown:

[0057] (1) First, determine the number of clusters K = 2, and then randomly select two data points in the spectrum amplitude graph as cluster centers.

[0058] (2) All data points in the spectrum graph are associated and divided into cluster centers closest to them, and clustering is divided based on this. The classification formula is:

[0059] C i =arg min‖x i - j ‖ 2 (2)

[0060] Where C i Represents the i-th data point x i The category closest to the cluster; arg is a marker symbol indicating the category to which the sample parameter belongs; μ j is the cluster center of the jth cluster.

[0061] (3) Move the cluster center to the center (means) of the current partitioned cluster containing all data points. The cluster center calculation formula of the data class is:

[0062]

[0063] Here, l{·} represents the set of distances and m represents the number of data points.

[0064] Repeat steps (2) and (3) for N=10 times. There are two conditions for terminating the iteration: when the number of iterations is reached or the cluster center no longer changes significantly, the spectrum amplitude diagram can be finally divided into two categories; the category with the larger amplitude value is assigned 1, and the category with the smaller amplitude value is assigned 0.

[0065] Step 32: Based on the classification result of step 31, the shape with the largest area and closed geometry in the category assigned a value of 1 is extracted by combining geometric graph theory and statistical principles; the geometric shape is used as the low-pass filter shape, and the pixel value within the shape is 1, indicating that the signal passes; the remaining pixels are assigned a value of 0, indicating that the signal is filtered out; the obtained geometric shape is used as the initial low-pass filter rect() for adaptive extraction.

[0066] Figure 4 Schematic diagram of the initial low-pass filter rect() extraction; four 32×32 pixel interference pattern patches randomly selected from the real interference pattern data are subjected to two-dimensional Fourier transform to obtain a two-dimensional spectrum diagram as shown in the figure. Figure 4(a) is shown in Figure 2; (b) is the clustering result of the K-means method on the spectrum graph, and the result is divided into two categories; only two values ​​are assigned to the two-dimensional image: 1 for the signal area and 0 for the non-signal area; in addition, it can be seen that for the clustering result, the signal is discontinuous; based on the above situation, the geometric shape of the largest continuous area in the shape classified as the signal component in (b) is extracted again as the initial low-pass filter shape rect(), and the pixels outside the geometric shape are assigned 0, as shown in Figure 2. Figure 4 As shown in (c).

[0067] Step 33: Generate a two-dimensional Kaiser window function.

[0068]

[0069] Where K(n) represents the value of the two-dimensional Kaiser window function at time n; N is the length of the window function, and in this embodiment, n=32 represents the size of the window function; I0 represents the first-class modified Bessel function, β is the parameter that controls the window function, and in this embodiment, β=10; I0[β] represents the value of the first-class modified Bessel function at β. The size of the two-dimensional window function is consistent with the size of the divided interference pattern patch, which is 32×32 pixels in this embodiment. K(n) is as follows Figure 5 shown.

[0070] Step 33: The Kaiser window function can suppress the Gibbs phenomenon caused by signal jumps. The Kaise window function is superimposed on the two-dimensional low-pass filter to obtain the final phase filter:

[0071] S{rect(|Z(u,v)|)}·K(n) (5)

[0072] Where rect() represents the initial low-pass filter, S{·} represents Gaussian smoothing, Z(u,v) represents the frequency domain signal, and u and v represent the horizontal and vertical frequency components during the two-dimensional Fourier transform process, respectively.

[0073] Step 4: The frequency domain signal of the interference pattern patch is passed through a phase filter composed of an initial low-pass filter and a Kaiser window function. The signal frequency band power will be retained, the noise at the high frequency position will be filtered out, and the noise at the low frequency position will be filtered out by smoothing operation:

[0074] H(u,v)=S{rect(|Z(u,v)|)}·K(n)·Z(u,v) (6)

[0075] Where H(u,v) represents the frequency domain signal after filtering.

[0076] Step 5: Perform inverse fast Fourier transform on the filtered frequency domain signal H(u,v) to obtain the interference pattern patch after filtering in the spatial domain.

[0077] Step 6: Splice the interference pattern patches to obtain the filtered InSAR interferogram.

[0078] Step 61 : According to the interference pattern patch having 75% overlapping pixels, 75% of the pixels at the edge of the patch are discarded after inverse Fourier transform, and only the center pixels of the patch are retained, that is, a small block of 8×8 pixels in the center of the patch.

[0079] Step 62: Using the retained center pixels of the patches, the patches are spliced ​​in the azimuth and range directions to obtain a filtered InSAR interferogram.

[0080] Example:

[0081] In one embodiment of the present invention, the specific implementation steps based on the simulation data are as follows:

[0082] Step 1: Generate a 512×512 pixel mountain terrain using the peak function in MATLAB. Based on this terrain, restore the InSAR phase strips, as shown in the following example: Figure 6 As shown in (a), the interferogram amplitude is uniformly assigned a value of 1; then a Gaussian noise of 512×512 pixels and a standard deviation of 0.8 (rad) is generated and superimposed on the noise-free interferogram to simulate the noisy observed InSAR phase, as shown in Figure 6 As shown in (b).

[0083] The simulated complex InSAR interferogram is divided into several 32×32 pixel interferogram patches with a step size of 8 pixels in azimuth and range directions.

[0084] Step 2: For each square interference pattern patch, use a two-dimensional fast Fourier transform to convert the interference pattern patch signal in the spatial domain into a frequency domain signal, and take the absolute value of the frequency domain signal to obtain a two-dimensional image with the spectrum amplitude as the value.

[0085] Step 3: For each patch spectrum image, the K-means clustering method is used to cluster the spectrum amplitude map into two categories. Then, the closed classification shape with the largest continuous center area is extracted using geometry, and the pixels inside the closed shape are assigned a value of 1, and the pixels outside the shape are assigned a value of 0, which is used as a low-pass filter, as shown in the following example: Figure 7 (a) shows the filter shape extracted after Fourier transform of the interference pattern patch uniformly selected along the azimuth and range directions in the simulation data.

[0086] A two-dimensional Kaiser window function is generated according to formula (4), and a two-dimensional low-pass filter and the Kaiser window function are combined to form a filter proposed by the present invention according to formula (5);

[0087] In step 4, the spectrum signal after Fourier transformation is passed through the proposed filter based on formula (6) to obtain a filtered spectrum signal; in the filtered spectrum signal, high-frequency noise is filtered out by the low-pass filter, low-frequency noise is filtered out by the smoothing operation, and the Gibbs phenomenon at the signal jump is suppressed by the Kaiser window function.

[0088] Step 5: Perform inverse fast Fourier transform on the frequency domain signal to obtain the complex interferogram patches filtered in the spatial domain; perform phase operation on the complex interferogram to obtain the phase of the InSAR interferogram after filtering of each patch, as shown in the following example: Figure 7 (b) shows the InSAR phase after filtering of the interferogram patches uniformly selected along the azimuth and range directions in the simulated data.

[0089] Step 6: Based on the initial segmentation method of the interferogram patches, the patches are spliced ​​into filtered interferograms. Since 75% of the pixels overlap during segmentation, only a small block of 8×8 pixels in the center of each patch is retained during splicing. The splicing is performed along the azimuth and range directions. The phase operation of the spliced ​​complex interferogram is performed to obtain the filtered InSAR phase, as shown in the following example: Figure 7 As shown in (c), the filtered phase is close to the noise-free InSAR phase, which also shows that the filter method proposed in the present invention can effectively filter out the existence of noise.

[0090] In order to show the reliability of the filtered InSAR phase, it is compared with the noise-free real InSAR phase to generate a scatter density map, such as Figure 7 As shown in (d), it can be seen that R 2 =0.997, indicating that the filtered InSAR phase is consistent with the true InSAR phase distribution; and the RMSE is 0.117 (rad), indicating that the filtering effect is good, which also illustrates the reliability of the InSAR phase filtering method based on complex interferogram proposed in the present invention.

[0091] In summary, the present invention provides a new frequency-domain InSAR phase filtering method for reducing phase noise in InSAR data. The present invention analyzes the differences between the interference pattern signal and the noise in the spectrum diagram, demonstrates the existence and characteristics of various types of noise, and proposes a new filter shape to effectively separate the noise from the signal. Through the filtering process, the new filter shape aims to retain the signal and noise, rather than amplify or attenuate them. Importantly, our new filtering method eliminates the need for iterative filtering, thereby achieving faster operation. Finally, we use a simulated data set to verify the performance of the filtering method of the present invention. It can be seen that the implementation process of this method is simple and effective, and it has great significance for large-scale, multi-data time-series InSAR deformation monitoring and topography measurement.

[0092] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. InSAR phase filtering method based on complex interferogram, characterized by: include: The acquired primary and secondary SAR images are preprocessed to obtain complex InSAR interferograms; the complex InSAR interferograms are divided into square interferogram patches along the azimuth and range directions; Convert each interference pattern patch from the spatial domain to the frequency domain signal and construct the corresponding spectrum amplitude map; The K-means clustering method is used to cluster and extract the initial low-pass filter in the spectrum amplitude map, and the phase filter is constructed together with the Kaiser window function; The frequency domain signal of the interference pattern patch is passed through a phase filter. The signal frequency band power will be retained, the noise at the high frequency position will be filtered out, and the noise at the low frequency position will be filtered out by smoothing operation to obtain the filtered frequency domain signal. Convert the filtered frequency domain signal to the spatial domain to obtain the filtered interference pattern patch; The interference pattern patches are spliced ​​together to obtain the filtered InSAR interferogram.

2. The InSAR phase filtering method based on complex interferogram according to claim 1, characterized in that: The complex InSAR interferogram is slid in the azimuth and range directions with a preset step size to divide the interferogram into patches; there is a preset overlap between adjacent interferogram patches.

3. The InSAR phase filtering method based on complex interferogram according to claim 1, characterized in that: The K-means clustering method divides the spectrum amplitude graph into two categories, specifically: First, the number of clusters is determined to be 2, and then two data points in the spectrum amplitude map are randomly selected as cluster centers; All data points in the spectrum graph are associated and divided into clusters based on the center of the cluster closest to them; the cluster center is moved to the center of the current divided cluster containing all data points; this process is repeated until the number of iterations or the change in cluster center is less than the preset value, and the spectrum amplitude graph can finally be divided into two categories; the category with larger amplitude value is assigned 1, and the category with smaller amplitude value is assigned 0.

4. The InSAR phase filtering method based on complex interferogram according to claim 1, characterized in that: Based on the clustering results of the spectrum amplitude graph, the geometric graph theory and statistical principles are combined to extract the shape with the largest area and closed geometry in the category assigned a value of 1. The geometric shape is used as the low-pass filter shape, and the pixel value within the shape is 1, indicating that the signal passes. The remaining pixels are assigned a value of 0, indicating signal filtering; The resulting geometry serves as an initial low-pass filter for adaptive extraction.

5. The InSAR phase filtering method based on complex interferogram according to claim 1, characterized in that: The Kaise window function is superimposed on the two-dimensional low-pass filter to obtain the final phase filter: S{rect(|Z(u,v)|)}·K(n) Where rect() represents the initial low-pass filter, S{·} represents Gaussian smoothing, Z(u,v) represents the frequency domain signal, u and v represent the horizontal and vertical frequency components during the two-dimensional Fourier transform, respectively, and K(n) represents the value of the two-dimensional Kaiser window function at time n.

6. The InSAR phase filtering method based on complex interferogram according to claim 1, characterized in that: In the Kaise window function, the control parameter β of the window function is set to 10; the size of the two-dimensional Kaiser window function is consistent with the size of the divided interference pattern patches.

7. The InSAR phase filtering method based on complex interferogram according to claim 1, characterized in that: The filtered frequency domain signal is expressed as: H(u,v)=S{rect(|Z(u,v)|)}·K(n)·Z(u,v) Where H(u,v) represents the frequency domain signal after filtering.

8. The InSAR phase filtering method based on complex interferogram according to claim 1, characterized in that: The filtered frequency domain signal is converted to the spatial domain using inverse fast Fourier transform; The edge of each interference pattern patch obtained by the transformation in the spatial domain is discarded with a preset overlap, and only the remaining part is retained, which is then spliced ​​along the azimuth and range directions to obtain the filtered InSAR interferogram.

9. A terminal device comprising a processor, a memory, and a computer program stored in the memory; characterized in that: When the processor executes the computer program, the InSAR phase filtering method based on complex interferogram according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium storing a computer program; wherein: When the computer program is executed by a processor, the InSAR phase filtering method based on complex interferogram according to any one of claims 1 to 8 is implemented.

Images