Gaussian mixture model-based micro-variation monitoring radar image permanent scatterer selection method and storage medium
Patent Information
- Application Number
- CN202410157022.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-02-01
- Publication Date
- 2026-10-16
- Estimated Expiration
- 2044-02-01
AI Technical Summary
[0002]目前存在的形变监测技术中，传统永久散射体(Permanent Scatterers，PS)点的选取方法往往需要反复设置合适的阈值来选取满足条件的PS点，该操作过于繁琐且不具有适应性，会给PS点的筛选带来了一定的难度
[0038]This disclosure discloses a method for selecting permanent scattering objects (PS) from radar images based on Gaussian mixture models, and a computer-readable storage medium. The method involves at least grouping acquired radar images and performing permanent scattering object (PS) point selection processing; selecting candidate PS points; classifying amplitude deviation based on Gaussian mixtures; classifying coherence coefficient based on Gaussian mixtures; classifying phase stability based on Gaussian mixtures; and selecting target PS points based on sets of highly coherent and highly phase-stable PS points. This is achieved by employing a Gaussian mixture clustering model, which comprehensively considers multiple pixel features to effectively classify the data and select PS points. Firstly, grouping multiple acquired radar images increases the number of PS points within each group and effectively improves the stability of the number of PS points between groups, laying a good foundation for high-precision deformation inversion. Secondly, combining the amplitude and phase information of the radar images with the Gaussian mixture (GMM) clustering idea for PS point selection, and iteratively classifying the candidate PS points based on multiple features, solves the problem of setting the same threshold for selecting PS points in different areas of the same scene.
Smart Images
Figure CN118068327B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of differential interferometry data processing technology for micro-variable monitoring radar, specifically to a method for selecting permanent scatterers in micro-variable monitoring radar images based on a Gaussian mixture model and a computer-readable storage medium. Background Technology
[0002] In current deformation monitoring technologies, the traditional method for selecting permanent scatterer (PS) points often requires repeatedly setting appropriate thresholds to select PS points that meet the criteria. This operation is cumbersome and lacks adaptability, making PS point selection difficult. Furthermore, relying solely on manually setting thresholds frequently results in inaccuracies, reducing the accuracy of deformation inversion results. During long-term observation of a monitored scene, the scattering characteristics of scatterers in micro-deformation monitoring radar change over time, preventing timely updates of PS points and impacting deformation inversion results. When processing multiple radar images obtained from different areas of the same scene, relying solely on traditional fixed threshold methods for PS point selection leads to variations in the quantity and quality of selected PS points over long monitoring periods, introducing significant deformation inversion errors and ultimately affecting the reliability of the deformation inversion results. Summary of the Invention
[0003] This disclosure aims to provide a method for selecting permanent scatterers (PS) in micro-variation monitoring radar images based on a Gaussian mixture model and a computer-readable storage medium. While increasing the number of PS points within a group, it can also improve the stability of the number of PS points between groups, laying a good foundation for high-precision deformation inversion.
[0004] According to one of the solutions disclosed herein, a method for selecting permanent scatterers in micro-variation monitoring radar images based on a Gaussian mixture model is provided, comprising:
[0005] The acquired radar images are grouped and their permanent scatterer points are selected.
[0006] Select candidate permanent scatterer points;
[0007] Amplitude deviation classification based on Gaussian mixture;
[0008] Classification of coherence coefficients based on Gaussian mixtures;
[0009] Phase stability classification based on Gaussian mixtures;
[0010] The target permanent scatterer point is selected based on the set of highly coherent permanent scatterers and the set of highly phase-stable permanent scatterers.
[0011] In some embodiments, the grouping and permanent scatterer selection processing of the acquired radar images includes:
[0012] After monitoring is completed from the first radar image to the N1st radar image, these N1 radar images are used as the first set of radar images for the first permanent scatterer point selection process.
[0013] Each time a new radar image is generated, it is combined with the previous N1-1 radar images to form a new set of radar images for permanent scatterer point selection processing.
[0014] In some embodiments, selecting candidate permanent scatterer points includes:
[0015] Take N1 radar images, use the amplitude information of each pixel to perform preliminary screening, and obtain the amplitude threshold;
[0016] Calculate the minimum amplitude of each pixel extracted from N1 radar images;
[0017] The amplitude threshold of the group is compared with the minimum value. If the amplitude threshold is greater than the minimum value, the pixel is considered a candidate permanent scatterer point.
[0018] In some embodiments, the amplitude deviation classification based on Gaussian mixture includes:
[0019] The amplitude deviation index of candidate permanent scatterer points is calculated based on the standard deviation of pixel amplitudes in N1 radar images and the mean of pixel amplitudes in N1 radar images.
[0020] Gaussian mixture clustering is performed to obtain a set of low-amplitude deviation permanent scatterers.
[0021] In some embodiments, the classification of coherence coefficients based on Gaussian mixtures includes:
[0022] Based on a set of low-amplitude deviation permanent scatterers, the coherence coefficient of each pixel in the set is calculated and used as input data for the Gaussian mixture clustering algorithm.
[0023] Perform Gaussian mixture clustering iteratively until the iteration terminates to obtain the optimal parameter values in the mixture distribution suitable for the dataset;
[0024] Calculate the probability that each pixel belongs to one of the two clusters using the optimal parameter values, compare the probability values, and select the pixel from the cluster with the higher probability to obtain the high coherence permanent scatterer point set.
[0025] In some embodiments, the phase stability classification based on Gaussian mixture includes:
[0026] Calculate the phase stability information of each pixel in the low-amplitude deviation permanent scatterer point set, and use it as input data for the Gaussian mixture clustering algorithm;
[0027] Perform Gaussian mixture clustering iteratively until the iteration terminates to obtain the optimal parameter values in the mixture distribution suitable for the dataset;
[0028] The probability of each pixel belonging to one of the two clusters is calculated using the optimal parameter values. The pixel values of the cluster with the higher probability are selected to obtain the high phase stability permanent scatterer point set.
[0029] In some embodiments, the method further includes:
[0030] For two clusters G = {G1, G2} in a Gaussian mixture, each cluster has parameters {α} i ,μ i ,∑ i Initialize |i=1,2} and set the initial values for iteration;
[0031] Clustering of data points is performed using various parameters, and the probability of each input data point belonging to a different cluster is calculated.
[0032] In some embodiments, where Gaussian mixture clustering is performed in a loop:
[0033] Each iteration updates the value of each parameter until the parameter values no longer change, at which point the iteration terminates and the loop ends.
[0034] In some embodiments, selecting the target permanent scatterer point based on a set of highly coherent permanent scatterer points and a set of highly phase-stable permanent scatterer points includes:
[0035] The union of the high coherence permanent scatterer point set and the high phase stability permanent scatterer point set is taken as the final target permanent scatterer point.
[0036] According to one of the solutions of this disclosure, a computer-readable storage medium is provided, having stored thereon computer-executable instructions, which, when executed by a processor, implement:
[0037] The above-mentioned method for selecting permanent scatterers in small-variation monitoring radar images based on Gaussian mixture models.
[0038] This disclosure discloses a method for selecting permanent scattering objects (PS) from radar images based on Gaussian mixture models, and a computer-readable storage medium. The method involves at least grouping acquired radar images and performing permanent scattering object (PS) point selection processing; selecting candidate PS points; classifying amplitude deviation based on Gaussian mixtures; classifying coherence coefficient based on Gaussian mixtures; classifying phase stability based on Gaussian mixtures; and selecting target PS points based on sets of highly coherent and highly phase-stable PS points. This is achieved by employing a Gaussian mixture clustering model, which comprehensively considers multiple pixel features to effectively classify the data and select PS points. Firstly, grouping multiple acquired radar images increases the number of PS points within each group and effectively improves the stability of the number of PS points between groups, laying a good foundation for high-precision deformation inversion. Secondly, combining the amplitude and phase information of the radar images with the Gaussian mixture (GMM) clustering idea for PS point selection, and iteratively classifying the candidate PS points based on multiple features, solves the problem of setting the same threshold for selecting PS points in different areas of the same scene.
[0039] It should be understood that the foregoing general description and the following detailed description are exemplary and illustrative only, and are not intended to limit the scope of this disclosure. Attached Figure Description
[0040] In drawings that are not necessarily drawn to scale, similar reference numerals in different views may indicate similar components. Similar reference numerals with letter suffixes or similar reference numerals with different letter suffixes may indicate different instances of similar components. The drawings are generally used to illustrate various embodiments by way of example rather than limitation, and are used together with the specification and claims to explain the disclosed embodiments.
[0041] Figure 1 A general flowchart of the PS point selection method based on GMM according to an embodiment of the present disclosure is shown;
[0042] Figure 2 A flowchart illustrating the specific steps of the GMM-based PS point selection method according to an embodiment of this disclosure is shown.
[0043] Figure 3 The flowchart shows the process of grouping and PS selection of multiple radar images;
[0044] Figure 4 The GMM clustering flowchart is shown;
[0045] Figure 5 A comparison chart of the quantitative results is shown between the amplitude deviation method (left) and the method for selecting PS points based on GMM (right);
[0046] Figure 6A comparison diagram of phase results is shown between the amplitude deviation method (left) and the GMM-based PS point selection method (right);
[0047] Figure 7 A comparison chart showing the number of PS points for each group is presented. Detailed Implementation
[0048] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the described embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.
[0049] Micro-deformation monitoring radar is a technology with advantages such as all-day, all-weather, long-range, and high precision. It utilizes phase differential interferometry to achieve high-precision measurement of surface deformation. Permanent scatterer (PS) interferometry is widely used in micro-deformation monitoring radar to monitor ground deformation and obtain high-precision deformation information. However, in practical applications, extracting stable, highly correlated, and high signal-to-noise ratio PS points from radar images is crucial for obtaining high-precision deformation results. During differential interferometry in micro-deformation monitoring radar, the phase stability of many pixels in the radar image is low due to the influence of temporal and spatial decorrelation. For example, the scattering characteristics of vegetation and water bodies may change significantly in a short period of time, interfering with the accurate measurement of surface deformation. During long-term monitoring, PS points may disappear or reappear. Due to differences in pixel quality between different radar image groups, the accuracy of deformation results decreases, seriously affecting the quality of deformation inversion.
[0050] In conjunction with the preceding background section, this disclosure provides illustrative examples of solutions to address the deficiencies in the prior art, but these are not intended to limit the scope of patent protection claimed in this disclosure.
[0051] As one solution, embodiments of this disclosure provide a method for selecting permanent scatterers in micro-variation monitoring radar images based on a Gaussian mixture model, including:
[0052] The acquired radar images are grouped and their permanent scatterer points are selected.
[0053] Select candidate permanent scatterer points;
[0054] Amplitude deviation classification based on Gaussian mixture;
[0055] Classification of coherence coefficients based on Gaussian mixtures;
[0056] Phase stability classification based on Gaussian mixtures;
[0057] The target permanent scatterer point is selected based on the set of highly coherent permanent scatterers and the set of highly phase-stable permanent scatterers.
[0058] In response to the aforementioned issues, such as Figure 1 The flowchart shown is an overall representation of the Gaussian Mixture Model (GMM)-based PS point selection method of this disclosure. Each embodiment aims to employ a Gaussian Mixture Clustering (GMM) model to effectively classify data by comprehensively considering multiple features of pixels, thereby selecting PS points. This method first groups time-series radar images and continuously updates the PS points for each group. By analyzing pixel information, the amplitude and phase information of the radar images are combined with the GMM clustering algorithm to jointly identify the permanent scattering object (PS) points. The entire PS point selection process uses automatic classification instead of manually setting a fixed threshold. Compared to traditional threshold methods, this method increases the number of PS points within a group while improving the stability of the number of PS points between groups, laying a solid foundation for high-precision deformation inversion.
[0059] For example, the following steps S1 to S6 further illustrate the method for selecting permanent scatterers in micro-variation monitoring radar images based on Gaussian mixture models.
[0060] In some implementations, the method of this disclosure may include: grouping acquired radar images and performing permanent scatterer selection processing, including:
[0061] After monitoring is completed from the first radar image to the N1st radar image, these N1 radar images are used as the first set of radar images for the first permanent scatterer point selection process.
[0062] Each time a new radar image is generated, it is combined with the previous N1-1 radar images to form a new set of radar images for permanent scatterer point selection processing.
[0063] Combination Figure 2 The flowchart shown is a detailed step diagram of the PS point selection method based on GMM in this embodiment. Step S1 in this embodiment may further include steps S11 to S12.
[0064] Step S1: Group the acquired N radar images into l groups and perform PS selection processing. N = {N1, N2, ..., N} k |k=1,2,...,l}.
[0065] Grouping diagram as follows Figure 3The flowchart shows the process of grouping and selecting PS points from multiple radar images. The following section uses N1 radar images (N1 is set to 30) as an example to demonstrate the PS point selection process.
[0066] Step S11: From the first image to the completion of monitoring the N1th image, use these N1 radar images as the first set of radar images to perform the first permanent scatterer point selection process.
[0067] Step S12: After step S11, for each new radar image generated, a new set of radar images is formed with the previous N1-1 radar images and permanent scatterer point selection processing is performed.
[0068] In some implementations, the method of this disclosure may include: selecting candidate permanent scatterer points, including:
[0069] Take N1 radar images, use the amplitude information of each pixel to perform preliminary screening, and obtain the amplitude threshold;
[0070] Calculate the minimum amplitude of each pixel extracted from N1 radar images;
[0071] The amplitude threshold of the group is compared with the minimum value. If the amplitude threshold is greater than the minimum value, the pixel is considered a candidate permanent scatterer point.
[0072] Combination Figure 2 The flowchart shown is a detailed step diagram of the PS point selection method based on GMM in this embodiment. Step S2 in this embodiment can be further included in steps S21 to S23 based on the aforementioned step S1.
[0073] Step S2: Selection of candidate PS points.
[0074] Step S21: Take N1 radar images and perform preliminary screening of each pixel using the amplitude information of the pixels.
[0075]
[0076] Where T is the amplitude threshold, and row and col represent the number of rows and columns in a single image, respectively. This represents the amplitude value of pixel (x,y) in the k-th image after radiometric calibration.
[0077] Step S22: Calculate the minimum amplitude A of each pixel extracted from N1 radar images. min .
[0078] Step S23: Compare the amplitude threshold T of the permanent scatterer points in the set of radar images with the minimum value A of the amplitude time series of the pixels in N1 radar images. minCompare and determine whether to consider the pixel as a candidate point for PS. If A min If A > T, then the pixel is considered a candidate PS point; if A min If <T, then the pixel is considered a non-PS candidate point.
[0079] In some implementations, the method of this disclosure can be: for each parameter {α} of two clusters G = {G1, G2} in a Gaussian mixture. i ,μ i ,∑ i Initialize |i=1,2} and set the initial values for iteration;
[0080] Clustering of data points is performed using various parameters, and the probability of each input data point belonging to a different cluster is calculated.
[0081] In some implementations, the method of this disclosure embodiment may be: performing a Gaussian mixture clustering loop:
[0082] Each iteration updates the value of each parameter until the parameter values no longer change, at which point the iteration terminates and the loop ends.
[0083] In some implementations, the method of this disclosure can be: amplitude deviation classification based on Gaussian mixture, including:
[0084] Based on the standard deviation of pixel amplitudes and the mean of pixel amplitudes in N1 radar images, the amplitude deviation index of candidate permanent scatterer points is calculated; Gaussian mixture clustering is then performed to obtain a set of permanent scatterer points with low amplitude deviation.
[0085] Combination Figure 2 The flowchart shown is a detailed step diagram of the PS point selection method based on GMM in this embodiment. Step S3 in this embodiment can further include steps S31 to S32 based on the aforementioned step S2.
[0086] Step S3: Amplitude deviation classification based on GMM; Step S31: Calculate the amplitude deviation index of candidate permanent scatterer points.
[0087]
[0088] In the formula, σ x,y Let m be the standard deviation of the pixel amplitude in N1 radar images. x,y Let N be the average value of the pixel amplitude in N1 radar images. The amplitude deviation index is used to calculate the amplitude deviation index of each candidate PS point in N radar images from group l.
[0089] Step S32: Perform Gaussian mixture clustering.
[0090] Step S321: GMM clustering flowchart as follows Figure 4 As shown. Assume the input data follows two Gaussian distributions, i.e., two clusters G = {G1, G2}. For each cluster in the GMM, the parameters {α}... i ,μ i ,∑ i Initialize |i=1,2} and set the initial values for iteration.
[0091] Step S322: For the candidate PS points in the input data, calculate the magnitude deviation index of each candidate PS point according to the formula in step S31. It is also used as input data for the Gaussian mixture clustering algorithm.
[0092] Step S323: Calculate the posterior probability R of each candidate PS point relative to each Gaussian distribution. si
[0093]
[0094] Its probability density function
[0095]
[0096] In the formula, μ i It is the mean, ∑ i It is covariance.
[0097] Step S324: Solve for the parameters in the Gaussian mixture distribution using the maximum likelihood estimation function LL(X).
[0098]
[0099] Step S325: Take the partial derivative of LL(X) with respect to each parameter and set it to 0 to calculate the value of each parameter.
[0100]
[0101]
[0102]
[0103] Step S326: Repeat steps S323 and S325 sequentially to perform GMM clustering. Update the parameter {α′} once for each iteration. i ,μ′ i ,∑' i The iteration terminates and the loop ends when the three parameter values no longer change. At this point, the optimal parameter values for the mixed distribution suitable for the dataset are obtained.
[0104] Step S327: After the iteration is completed, use the optimal parameter values at this time to calculate the probability that each pixel belongs to one of the two clusters, and classify the candidate permanent scattering points in step S31. By comparing the two probability values, select the pixel in the cluster with the higher probability and consider it as a low-amplitude deviation permanent scattering point, denoted as set A1; discard the pixel in the cluster with the lower probability.
[0105] In some implementations, the method of this disclosure can be: coherence coefficient classification based on Gaussian mixture, including:
[0106] Based on a set of low-amplitude deviation permanent scatterers, the coherence coefficient of each pixel in the set is calculated and used as input data for the Gaussian mixture clustering algorithm.
[0107] Perform Gaussian mixture clustering iteratively until the iteration terminates to obtain the optimal parameter values in the mixture distribution suitable for the dataset;
[0108] Calculate the probability that each pixel belongs to one of the two clusters using the optimal parameter values, compare the probability values, and select the pixel from the cluster with the higher probability to obtain the high coherence permanent scatterer point set.
[0109] Combination Figure 2 The flowchart shown is a detailed step diagram of the PS point selection method based on GMM in this embodiment. Step S4 in this embodiment can be further included in steps S41 to S45 based on the aforementioned step S3.
[0110] Step S4: Classification of coherence coefficients based on GMM.
[0111] Step S41: For each parameter {α} of each cluster in the GMM i ,μ i ,∑ i Initialize |i=1,2} and set the initial values for iteration.
[0112] Step S42: For set A1 in step S327, calculate the coherence coefficient γ of each pixel in the set. 1 And γ 1 As input data for the GMM clustering algorithm,
[0113]
[0114] In the formula, x and y represent the pixel positions, M(x,y) and S(x,y) represent the primary and secondary image information, respectively, w and h represent the size of the rectangular sliding window, and * represents the conjugate of the complex number. For low-amplitude deviation index candidate permanent scatterer points in N radar images of group l, the coherence coefficient sequence of each pixel is calculated.
[0115] Step S43: Use the three parameters from step S325 to perform clustering of the data points, and calculate the different clusters G to which each input data point belongs according to step S323. i The probability of.
[0116] Step S44: Execute steps S323 and S325 sequentially to perform GMM clustering iteratively. Update the values of the three parameters once per iteration. The iteration terminates when the three parameter values no longer change, and the loop ends. At this point, the optimal parameter values for the mixture distribution suitable for this dataset are obtained.
[0117] Step S45: After the iteration is complete, use the optimal parameter values at this point to calculate the probability that each pixel belongs to one of the two clusters, and divide the permanent scatterer points from step S42 into two categories. Compare the probability values, select the pixels in the cluster with the higher probability as high-coherence permanent scatterer points, denoted as set A2; discard the pixels in the cluster with the lower probability.
[0118] In some implementations, the method of this disclosure can be: Gaussian mixture-based phase stability classification, including:
[0119] Calculate the phase stability information of each pixel in the low-amplitude deviation permanent scatterer point set, and use it as input data for the Gaussian mixture clustering algorithm;
[0120] Perform Gaussian mixture clustering iteratively until the iteration terminates to obtain the optimal parameter values in the mixture distribution suitable for the dataset;
[0121] The probability of each pixel belonging to one of the two clusters is calculated using the optimal parameter values. The pixel values of the cluster with the higher probability are selected to obtain the high phase stability permanent scatterer point set.
[0122] Combination Figure 2 The flowchart shown is a detailed step diagram of the PS point selection method based on GMM in this embodiment. Step S5 in this embodiment can be further included from step S51 to step S55 based on the aforementioned step S4.
[0123] Step S5: Phase stability classification based on GMM.
[0124] Step S51: Initialize each parameter of each cluster in the GMM and set the initial values for iteration.
[0125] Step S52: Calculate the phase stability information γγ of each pixel in the low-amplitude deviation permanent scatterer point set. 1 And γγ 1 As input data for the Gaussian mixture clustering algorithm,
[0126]
[0127] In the formula, γ 1 For the temporal correlation of pixels, Let be the k-th interference phase value of a given pixel. For low-amplitude deviation index candidate permanent scatterer points in l groups of radar images, the phase stability information sequence of each pixel is calculated.
[0128] Step S53: Use the three parameters from step S325 to perform clustering of data points, and calculate the probability of each input data belonging to a different cluster according to step S323.
[0129] Step S54: Execute steps S323 and S325 sequentially to perform GMM clustering iteratively. Update the parameter values once per iteration, and update the data point assignments after each iteration. The iteration terminates and the loop exits when the three parameter values no longer change. At this point, the optimal parameter values in the mixture distribution suitable for the dataset are obtained.
[0130] Step S55: After the iteration is complete, use the optimal parameter values at this point to calculate the probability that each pixel belongs to one of the two clusters, and divide the permanent scatterer points from step S52 into two categories. By comparing the probability values, select the pixels in the cluster with the higher probability as the set of high phase-stability permanent scatterer points, denoted as set A3.
[0131] In some implementations, the method of this disclosure may include: selecting target permanent scattering points based on a set of highly coherent permanent scattering points and a set of highly phase-stable permanent scattering points, including:
[0132] The union of the high coherence permanent scatterer point set and the high phase stability permanent scatterer point set is taken as the final target permanent scatterer point.
[0133] Combination Figure 2 The flowchart shown in this embodiment illustrates the specific steps of the PS point selection method based on GMM. Step S6 in this embodiment can be implemented based on the aforementioned step S5 as follows:
[0134] Step S6: Take the union of the high coherence PS point set A2 obtained in step S45 and the high phase stability PS point set A3 obtained in step S55, and use it as the final target PS point.
[0135] See Figure 5 The graph shows a comparison of the quantitative results of the amplitude deviation method (left) and the PS point selection method based on GMM (right). Figure 6 The diagram shows a comparison of phase results between the amplitude deviation method (left) and the GMM-based PS point selection method (right). Figure 7The diagram shows a comparison of the number of PS points in each group. The embodiments of this disclosure can improve the stability of the number of PS points between groups to a certain extent.
[0136] This disclosure provides a method for selecting permanent scatterers (PS) in micro-deformation monitoring radar images based on Geometric Matrix (GMM) clustering. The method introduces a GMM clustering model to classify and select PS points. By iteratively updating the probability values of the model and continuously updating the parameter values, the iteration terminates when the parameters no longer change. The classification of PS points is determined based on the probability values. These embodiments are significant for the accurate selection of PS points and subsequent deformation inversion in micro-deformation monitoring radar, compared to existing methods:
[0137] 1. First, the acquired radar images are grouped, which increases the number of PS points within a group to a certain extent, and effectively improves the stability of the number of PS points between groups, laying a good foundation for high-precision deformation inversion.
[0138] 2. The amplitude and phase information of radar images are combined with the GMM clustering idea to select PS points. The PS points are iteratively classified according to the multi-faceted characteristics of the candidate PS points, which effectively solves the problem of setting the same threshold for selecting PS points in different areas of the same scene.
[0139] As one solution, embodiments of this disclosure provide an apparatus for selecting permanent scatterers in micro-variation monitoring radar images based on a Gaussian mixture model, comprising one or more processing modules configured as follows:
[0140] The acquired radar images are grouped and their permanent scatterer points are selected.
[0141] Select candidate permanent scatterer points;
[0142] Amplitude deviation classification based on Gaussian mixture;
[0143] Classification of coherence coefficients based on Gaussian mixtures;
[0144] Phase stability classification based on Gaussian mixtures;
[0145] The target permanent scatterer point is selected based on the set of highly coherent permanent scatterers and the set of highly phase-stable permanent scatterers.
[0146] As one implementation, one or more processing modules of this disclosure are further configured to: group the acquired radar images and perform permanent scatterer selection processing, including:
[0147] After monitoring is completed from the first radar image to the N1st radar image, these N1 radar images are used as the first set of radar images for the first permanent scatterer point selection process.
[0148] Each time a new radar image is generated, it is combined with the previous N1-1 radar images to form a new set of radar images for permanent scatterer point selection processing.
[0149] As one implementation, one or more processing modules of this disclosure are further configured to: select candidate permanent scatterer points, including:
[0150] Take N1 radar images, use the amplitude information of each pixel to perform preliminary screening, and obtain the amplitude threshold;
[0151] Calculate the minimum amplitude of each pixel extracted from N1 radar images;
[0152] The amplitude threshold of the group is compared with the minimum value. If the amplitude threshold is greater than the minimum value, the pixel is considered a candidate permanent scatterer point.
[0153] As one implementation, one or more processing modules of this disclosure are further configured to: perform amplitude deviation classification based on Gaussian mixture, including:
[0154] The amplitude deviation index of candidate permanent scatterer points is calculated based on the standard deviation of pixel amplitudes in N1 radar images and the mean of pixel amplitudes in N1 radar images.
[0155] Gaussian mixture clustering is performed to obtain a set of low-amplitude deviation permanent scatterers.
[0156] As one implementation, one or more processing modules of this disclosure are further configured to: classify coherence coefficients based on Gaussian mixtures, including:
[0157] Based on a set of low-amplitude deviation permanent scatterers, the coherence coefficient of each pixel in the set is calculated and used as input data for the Gaussian mixture clustering algorithm.
[0158] Perform Gaussian mixture clustering iteratively until the iteration terminates to obtain the optimal parameter values in the mixture distribution suitable for the dataset;
[0159] Calculate the probability that each pixel belongs to one of the two clusters using the optimal parameter values, compare the probability values, and select the pixel from the cluster with the higher probability to obtain the high coherence permanent scatterer point set.
[0160] As one implementation, one or more processing modules of this disclosure are further configured to: perform phase stability classification based on Gaussian mixtures, including:
[0161] Calculate the phase stability information of each pixel in the low-amplitude deviation permanent scatterer point set, and use it as input data for the Gaussian mixture clustering algorithm;
[0162] Perform Gaussian mixture clustering iteratively until the iteration terminates to obtain the optimal parameter values in the mixture distribution suitable for the dataset;
[0163] The probability of each pixel belonging to one of the two clusters is calculated using the optimal parameter values. The pixel values of the cluster with the higher probability are selected to obtain the high phase stability permanent scatterer point set.
[0164] As one implementation, one or more processing modules of this disclosure are further configured to: process the parameters {α} of each cluster of the two clusters G = {G1, G2} in the Gaussian mixture. i ,μ i ,∑ i Initialize |i=1,2} and set the initial values for iteration;
[0165] Clustering of data points is performed using various parameters, and the probability of each input data point belonging to a different cluster is calculated.
[0166] As one implementation, one or more processing modules of this disclosure are further configured to: perform Gaussian mixture clustering loops:
[0167] Each iteration updates the value of each parameter until the parameter values no longer change, at which point the iteration terminates and the loop ends.
[0168] As one implementation, one or more processing modules of this disclosure are further configured to: select target permanent scattering points based on a set of highly coherent permanent scattering points and a set of highly phase-stable permanent scattering points, including:
[0169] The union of the high coherence permanent scatterer point set and the high phase stability permanent scatterer point set is taken as the final target permanent scatterer point.
[0170] In some specific implementations, one or more processing modules of this disclosure are configured to perform specific implementations of steps S1 to S6 described above.
[0171] Specifically, one of the inventive concepts disclosed herein aims to at least group acquired radar images and perform permanent scattering point (PS) selection processing; select candidate PS points; classify amplitude deviation based on Gaussian mixture; classify coherence coefficient based on Gaussian mixture; classify phase stability based on Gaussian mixture; and select target PS points based on a set of highly coherent PS points and a set of highly phase stable PS points. This is achieved by employing a Gaussian mixture clustering model, which comprehensively considers multiple features of pixels to effectively classify the data and select PS points. On one hand, grouping multiple acquired radar images increases the number of PS points within a group to a certain extent while effectively improving the stability of the number of PS points between groups, laying a good foundation for high-precision deformation inversion. On the other hand, combining the amplitude and phase information of radar images with the GMM clustering idea for PS point selection, and iteratively classifying the candidate PS points based on multiple features, solves the problem of setting the same threshold for selecting PS points in different areas of the same scene.
[0172] This disclosure also provides a computer-readable storage medium storing computer-executable instructions thereon. When executed by a processor, the computer-executable instructions mainly implement the above-described method for selecting permanent scatterers in micro-variation monitoring radar images based on a Gaussian mixture model, including at least:
[0173] The acquired radar images are grouped and their permanent scatterer points are selected.
[0174] Select candidate permanent scatterer points;
[0175] Amplitude deviation classification based on Gaussian mixture;
[0176] Classification of coherence coefficients based on Gaussian mixtures;
[0177] Phase stability classification based on Gaussian mixtures;
[0178] The target permanent scatterer point is selected based on the set of highly coherent permanent scatterers and the set of highly phase-stable permanent scatterers.
[0179] The above embodiments are merely exemplary embodiments of this disclosure and are not intended to limit this disclosure. The scope of protection of this disclosure is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this disclosure within its substance and scope, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this disclosure.
Claims
1. A method for selecting permanent scatterers in micro-variation monitoring radar images based on Gaussian mixture models, including: The acquired radar images are grouped and their permanent scatterer points are selected. Select candidate permanent scatterer points; Amplitude deviation classification based on Gaussian mixture; Classification of coherence coefficients based on Gaussian mixtures; Phase stability classification based on Gaussian mixtures; The target permanent scattering point is selected based on the high coherence permanent scattering point set and the high phase stability permanent scattering point set; Candidate permanent scatterer points were selected, including: Pick A radar image is used to initially filter each pixel using the amplitude information of the pixels, and an amplitude threshold is obtained. calculate The minimum amplitude of each pixel extracted from a radar image; The amplitude threshold of the group is compared with the minimum value. If the amplitude threshold is greater than the minimum value, the pixel is considered a candidate permanent scatterer point. Gaussian mixture-based amplitude deviation classification includes: based on The standard deviation of pixel amplitude in a radar image The mean amplitude of pixels in the radar image is used to calculate the amplitude deviation index of candidate permanent scatterer points; Gaussian mixture clustering is performed to obtain a set of low-amplitude deviation permanent scattering points; The classification of coherence coefficients based on Gaussian mixtures includes: Based on a set of low-amplitude deviation permanent scatterers, the coherence coefficient of each pixel in the set is calculated and used as input data for the Gaussian mixture clustering algorithm. Perform Gaussian mixture clustering iteratively until the iteration terminates to obtain the optimal parameter values in the mixture distribution suitable for the dataset; Calculate the probability that each pixel belongs to one of the two clusters using the optimal parameter values, compare the probability values, and extract the pixels from the cluster with the higher probability to obtain the high coherence permanent scatterer point set. Phase stability classification based on Gaussian mixtures includes: Calculate the phase stability information of each pixel in the low-amplitude deviation permanent scatterer point set, and use it as input data for the Gaussian mixture clustering algorithm; Perform Gaussian mixture clustering iteratively until the iteration terminates to obtain the optimal parameter values in the mixture distribution suitable for the dataset; Calculate the probability that each pixel belongs to one of the two clusters using the optimal parameter values, compare the probability values, and select the pixel from the cluster with the higher probability to obtain the high phase stability permanent scatterer point set; The method further includes: For two clusters in a Gaussian mixture Parameters of each cluster Initialization, setting initial values for iteration; Clustering of data points is performed using various parameters, and the probability of each input data point belonging to a different cluster is calculated.
2. The method according to claim 1, wherein, The acquired radar images are grouped and their permanent scatterer selection processes are performed, including: From the first radar image to the... The radar image monitoring is complete, and this... The radar images are used as the first set of radar images for the first permanent scatterer point selection process. Each time a new radar image is generated, it is compared with the previous one. The radar images are used to form a new set of radar images for permanent scatterer point selection processing.
3. The method according to claim 1, wherein, In the case of performing Gaussian mixture clustering loop: Each iteration updates the value of each parameter until the parameter values no longer change, at which point the iteration terminates and the loop ends.
4. The method according to claim 1, wherein, Based on the set of highly coherent permanent scatterers and the set of highly phase-stable permanent scatterers, target permanent scatterer points are selected, including: The union of the high coherence permanent scatterer point set and the high phase stability permanent scatterer point set is taken as the final target permanent scatterer point.
5. A computer-readable storage medium having stored thereon computer-executable instructions, which, when executed by a processor, implement: The method for selecting permanent scatterers in micro-variation monitoring radar images based on Gaussian mixture models according to any one of claims 1 to 4.