An Interference Phase Optimization Method Based on Adaptive Spatiotemporal Filtering Fusion

Abstract

The present invention discloses an interference phase optimization method based on adaptive spatio-temporal filtering fusion, belonging to the field of deformation monitoring of radar satellite SAR data. A time series of multiple single-look complex (SLC) images covering the same area is collected, the amplitude information of the time series is extracted, and at the same time, a full-interferometric pair combination network is constructed using the time series data of the SLC. An adaptive spatial dimension phase filtering method based on the extraction of the main frequency takes into account both the noise suppression of the spatial dimension interference phase and the protection of the phase information; a coherence deviation correction algorithm based on the second statistical feature is introduced to improve the estimation accuracy of coherence; an adaptive weighting strategy based on the sigmoid model is introduced to assign reasonable weights to pixels of different qualities and improve the optimization effect; its steps are simple and the application effect is good, especially for the phase optimization processing in high-noise and large-gradient deformation areas, which can effectively improve the optimization effect, noise robustness and self-adaptability of the algorithm, making the interference phase information for time series deformation interpretation smooth and rich.

Description

Technical Field

[0001] The present invention relates to a phase optimization method, in particular to an interferometric phase optimization method based on adaptive spatio-temporal filtering fusion, which is applicable to processing radar satellite time-series SAR data, and belongs to the field of deformation monitoring of time-series InSAR technology. Background Art

[0002] Interferometric synthetic aperture radar (InSAR) technology plays a very important role in geophysics and geodesy disciplines due to its advantages of large range, high precision, all-weather and all-day. The time-series InSAR technology developed based on the analysis of time-series SAR images has shown very broad application prospects in the fields of surface deformation monitoring, geological disaster monitoring, etc. In order to make up for the serious deficiency of the density of permanent scatterer targets in non-urban areas in the time-series InSAR technology, the time-series InSAR technology based on distributed scatterers has been widely promoted. However, there are many random scatterers with similar scattering characteristics in the resolution unit of distributed scatterers, resulting in its susceptibility to temporal and geometric decorrelation. Therefore, in the TSInSAR technology based on distributed scatterers, the phase optimization process aiming to improve the signal-to-noise ratio of the observed phase is a very critical step.

[0003] In the phase optimization process, the covariance matrix or the complex coherence matrix is the basis of the phase optimization model, which has a great influence on the subsequent optimized phase results. It can be decomposed into a coherence matrix and an interferometric phase matrix, and usually, it is obtained by the maximum likelihood estimation of a statistically homogeneous pixel neighborhood set. Regarding the coherence matrix, due to the limitation of the number of samples in the statistically homogeneous pixel neighborhood set and the inevitable interference of heterogeneous pixels, the coherence matrix in the sample covariance matrix obtained by the maximum likelihood estimation always has a deviation. Although the recently proposed advanced coherence estimation method based on the convolutional neural network joint model has shown good estimation results, the estimation accuracy of the coherence matrix is still limited by the number of samples. Therefore, we introduce a coherence deviation correction algorithm based on the second statistical feature to further reduce the coherence estimation deviation and improve the accuracy of the coherence matrix.

[0004] In terms of the interference phase matrix, the maximum likelihood estimation process based on a statistically homogeneous pixel set can be regarded as a homogeneous filtering process for the phase, that is, a spatial averaging filter based on the homogeneous pixel set. However, the selection of homogeneous pixels usually only considers the amplitude information of the SAR image while ignoring the phase information. In a certain scenario, if the target causes a backscattering amplitude similar to that of the background, then in the above-mentioned homogeneous filtering process, the corresponding phase value will be merged with the phase value corresponding to the background, which may lead to blurred phase information and even loss of the true phase information of the target. Especially in the study of deformation time series with a high deformation rate, such as mine deformation monitoring, the phase information contained in the interference pairs corresponding to medium or long baselines is basically independent of the amplitude, that is, the main phase component caused by deformation is independent of the amplitude information. In this case, the dense fringe information caused by large-gradient deformation is easily destroyed by the statistical averaging process of the homogeneous pixel set, resulting in the loss of necessary phase information. Therefore, we introduce an adaptive spatial dimension filtering process based on the extraction of the main frequency to balance phase noise suppression and phase information protection.

[0005] Based on the covariance matrix or the complex coherence matrix, the core difference in the setting of the weight factor in the constructed phase optimization model or algorithm. The process of optimal phase reconstruction can actually be understood as minimizing the difference between the estimated value and the observed value under a specific weight. It is worth noting that the phase optimization algorithm comprehensively considers all possible interference pair information, and advanced research has also shown that even low-coherence pixels can provide indispensable effective information. However, when a higher weight is assigned to low-quality pixels, it will reduce or weaken the role played by high-quality pixels in the phase optimization process, and thus it is difficult to obtain an ideal result in terms of improving the signal-to-noise ratio of the optimized phase. Therefore, it is very important to assign reasonable weight values to the phases of pixels with different qualities. For this purpose, we introduce an adaptive weighting strategy based on the sigmoid function model to assign more reasonable weights to pixels with different qualities, thereby improving the quality of the optimized phase. Summary of the Invention

[0006] Aiming at the deficiencies of the prior art, an interference phase optimization method based on adaptive spatio-temporal filtering fusion for processing radar satellite time series SAR data in high-noise and large-gradient deformation regions is provided. The interference phase information is estimated through adaptive spatial dimension filtering, the coherence estimation deviation is corrected through coherence deviation correction, and a complex coherence matrix is constructed. The phase optimization model based on the sigmoid adaptive weighting strategy is solved to effectively optimize the time series phase, which not only enhances the smoothness of the interference phase but also protects the phase information in the dense fringe region of the interference phase, making the interference phase information used for time series deformation interpretation smooth and rich.

[0007] In order to solve the above technical problems, the present invention provides an interference phase optimization method based on adaptive spatiotemporal filtering fusion and an adaptive spatial phase filtering method based on main frequency extraction, which takes into account both noise suppression and phase information protection of spatial interference phase; introduces a coherence deviation correction algorithm based on the second type of statistical characteristics to improve the estimation accuracy of coherence; introduces an adaptive weighting strategy based on a sigmoid model to assign reasonable weights to pixels of different qualities to improve the optimization effect; and thereby improves the phase quality and phase smoothing effect of the interference phase;

[0008] The specific steps are as follows:

[0009] S1. Collect the time series data of multiple single-view complex images SLC covering the same area, extract the time series amplitude information, and use the time series data of SLC to build a full interference pair combination network. Based on the interference pair combination information, conjugate multiplication is performed on the main image SLC and the auxiliary image SLC to obtain the differential interference phase under the full interference combination corresponding to the SLC time series data;

[0010] S2, using an adaptive spatial phase filtering method based on main frequency extraction to perform spatial phase filtering on the differential interference phase under the full interference combination, and obtaining the filtered interference phase under the full interference combination;

[0011] S3, using the fast homogeneous pixel identification algorithm (FaSHPS) combined with the time series amplitude information to calculate the statistical homogeneous pixel neighborhood set corresponding to each pixel of the SLC image pixel by pixel;

[0012] S4, based on the statistically homogeneous pixel neighborhood set, calculating the coherence coefficients corresponding to all filtered interference phases under the full interference combination, and using the coherence deviation correction method based on the second type of statistical characteristics to perform deviation correction processing on the coherence coefficients to obtain the refined coherence coefficients;

[0013] S5, combining the spatial dimension filtered interference phase and the refined coherence coefficient under the full interference combination, converting the combined interference phase and coherence coefficient into a time dimension interference phase matrix and a coherence matrix for a single pixel point of the SLC image, and constructing a complex coherence matrix for a single pixel point of the SLC image;

[0014] S6. Construct a phase optimization model of the maximum likelihood estimator based on the complex coherence matrix, calculate the adaptive weighting factor based on the sigmoid function model, and solve the phase optimization model in combination with the eigenvalue decomposition strategy to obtain the final optimized interference phase.

[0015] Further, step S2 specifically includes the following sub-steps:

[0016] S2.1, according to the coherence coefficient γ of the pixel to be filtered, adaptively estimate the filter window size Nwin , the formula is:

[0017]

[0018] In the formula, represents the integer operation, ε0 is the noise standard deviation of the phase, and the calculation method is as follows:

[0019]

[0020] In the formula, is the mean value of the coherence coefficient of the interferogram; N min represents the minimum value of the adaptive window, with a value of 5;

[0021] S2.2. Extract a local phase block φ win with a window size of N (m,n) centered on the pixel to be filtered, and use the following formula to transform the local phase block φ (m,n) to the frequency domain by means of fast Fourier transform to obtain the spectral value corresponding to φ (m,n) :

[0022] S (u,v) = FFT2{φ (m,n)} (3)

[0023] In the formula, φ (m,n) represents the local interference phase block with a window size of N win centered on the pixel index of the m-th row and n-th column in the spatial dimension, and S (u,v) is the spectral value after the transformation of the local phase block, and FFT2{·} represents the two-dimensional fast Fourier transform operation;

[0024] S2.3. Calculate the main spectral amplitude of the local window using the spectral value after the transformation of the local phase block

[0025]

[0026] In the formula, thr is the threshold parameter for extracting the main spectrum of the local window, and it is adaptively determined according to the spectral amplitude of the local window:

[0027]

[0028] S2.4. Convolve and multiply the extracted main spectral amplitude with the spectral value S (u,v) to extract the main phase component, and obtain the filtered phase of the local phase block through fast two-dimensional inverse Fourier transform:

[0029]

[0030] In the formula, φ'(m,n) is the filtered local phase block, is the dot product operation, FFT2 -1 {·} is the fast two-dimensional inverse Fourier transform;

[0031] S2.5. Extract the central pixel of the filtered phase block as the filtered phase of the pixel to be filtered, and repeat steps S2.1 - S2.4 until all pixels in the interferogram are traversed to obtain the interferogram filtered in the spatial dimension.

[0032] S2.6. Repeat S2.1 - S2.5 for all interferograms under the full-interference combination. After the traversal is completed, the filtered interference phase under the full-interference combination is obtained.

[0033] Furthermore, step S3 is specifically as follows:

[0034] S3.1. Based on the amplitude vector of the temporal SLC image data, use the following formula to calculate the estimated value of the amplitude mean of the pixel point l to be estimated in the SLC image:

[0035]

[0036] In the formula, is the estimated value of the mean μ(l), A represents the amplitude value of the SLC image, and N SLC represents the number of SLC images;

[0037] S3.2. Use the following formula to calculate the confidence interval at the confidence level of 1 - α to screen for homogeneous pixel points:

[0038]

[0039] In the formula, z 1-α / 2 is the percentile of the probability density function of the 1 - α / 2 standard normal distribution;

[0040] S3.3. Select a 15×15 pixel window centered on the pixel point to be estimated in the SLC image. The connected pixels within the pixel window that satisfy the above confidence interval are selected as homogeneous pixels. Traverse all pixels in the SLC image to obtain the statistical homogeneous pixel neighborhood set corresponding to each pixel.

[0041] Furthermore, step S4 is specifically as follows:

[0042] S4.1. Based on the filtered interference phase and the statistical homogeneous pixel neighborhood set Ω SHP , use the following formula to calculate the coherence coefficient corresponding to the filtered phase

[0043]

[0044] In the formula, Indicates t m The complex value of the SLC image at pixel i at time instant, Indicated by t m Time, t n The filtered phase value of the interference pair of the two SLC images at pixel i at time;

[0045] S4.2. Using the coherence bias correction method based on the second type of statistical characteristics to correct the estimated coherence Perform deviation correction processing to obtain the refined coherence coefficient

[0046]

[0047] Where N SHP is the statistically homogeneous pixel neighborhood set Ω SHP The number of pixels;

[0048] S4.3, repeat steps S4.1 to S4.2 until all filtered interference patterns under the full interference combination are traversed to obtain a refined coherence pattern corresponding to the filtered phase under the full interference combination.

[0049] Further, step S5 is specifically as follows:

[0050] S5.1. Combine all spatial dimension filtered interferometric phase matrices and convert them uniformly into N for a single pixel. SLC ×N SLC dimensional time-dimensional interference phase matrix After traversing all pixels, the time-dimensional interference phase matrix is ​​expressed as:

[0051]

[0052] S5.2. Combine all spatial dimension filtering phase corresponding coherence matrices and convert them uniformly into N for a single pixel. SLC ×N SLC The time-dimensional coherence matrix After traversing all pixels, the time dimension coherence matrix is ​​expressed as:

[0053]

[0054] S5.3. Time-dimensional interferometric phase matrix based on all pixels and the coherence matrix Constructing the complex coherence matrix

[0055]

[0056] Furthermore, step S6 is specifically as follows:

[0057] S6.1. Based on the complex coherence matrix and combined with the sigmoid function model, calculate the adaptive weight factor w using the following formula adaptive :

[0058]

[0059] In the formula, b represents the inflection point, and k represents the severity of data change before and after the inflection point, which is set to 40;

[0060] S6.2. Build a phase optimization function model based on the adaptive weight factor, and solve the optimization function model in combination with the eigenvalue decomposition strategy. The solution of the optimization model, that is, the optimal phase solution, is expressed as: The eigenvector corresponding to the minimum eigenvalue of

[0061] Beneficial effects:

[0062] The present invention fully analyzes the principle process of the phase optimization algorithm, converts it into an optimization processing step combined with spatio-temporal filtering, and through spatial adaptive interference phase filtering, takes into account both the suppression of phase noise and the protection of phase fringe information of the interferogram in the spatial dimension; introduces a coherence deviation correction algorithm based on the second type of statistical features to improve the coherence estimation accuracy corresponding to the filtered phase; through the adaptive weighting factor based on the sigmoid function model and the model solution strategy based on eigenvalue decomposition in the optimization model, improves the phase filtering effect and efficiency in the time dimension; combining the advantages of the above core steps, this method as a whole has better optimization effect, higher noise robustness and self-adaptability than the traditional phase optimization algorithm, especially for high-noise regions and large-gradient deformation regions. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is a schematic flow chart of the interference phase optimization method based on adaptive spatio-temporal filtering fusion of the present invention;

[0064] Figure 2 is a geographical location area map of the study area;

[0065] Figure 3 is a schematic diagram of the number of pixels included in the neighborhood set corresponding to each pixel;

[0066] Figure 4 is the coherence diagram corresponding to the filtered phase;

[0067] Figure 5 is the coherence diagram after deviation correction;

[0068] Figure 6 is a schematic diagram of the time-dimensional coherence matrix;

[0069] Figure 7 The original interferogram with a time baseline of 240 days;

[0070] Figure 8 The interferogram with a time baseline of 240 days obtained from the PTA optimization result;

[0071] Figure 9 The interferogram with a time baseline of 240 days obtained from the EMI optimization result;

[0072] Figure 10 The interferogram with a time baseline of 240 days obtained from the optimization result using the method of the present invention;

[0073] Figure 11 Schematic diagram of the mean and standard deviation of the interferogram residual points and the percentage increase in SPD under the full interferogram combination obtained from the optimized phases corresponding to different algorithms. Detailed implementation manner

[0074] The technical solution of the present invention will be further described below with reference to the accompanying drawings:

[0075] As Figure 1 shown in the flow, an interferometric phase optimization method based on adaptive spatio-temporal filtering fusion according to the present invention has input information of 24 scenes of time series SLC images covering the study area, and the corresponding actual geographical location area map of the study area is as Figure 2 shown, and the method specifically includes the following steps:

[0076] S1. Collect time series data of multiple single-look complex images (SLC) covering the same area, extract the time series amplitude information, construct a full interferogram pair combination network, and perform conjugate multiplication on the master and slave SLCs based on the interferogram pair combination information to obtain the differential interferometric phase under the full interferogram combination;

[0077] S2. Perform spatial dimension phase filtering on the interferometric phase diagram under the full interferogram combination by using an adaptive spatial dimension phase filtering method based on the main frequency extraction to obtain the filtered phase under the full interferogram combination;

[0078] S2.1. Adaptively estimate the filtering window size N win according to the coherence coefficient γ of the pixel to be filtered, and the formula is:

[0079]

[0080] In the formula, represents the rounding operation, ε0 is the standard deviation of the phase noise, which is calculated by the following formula,

[0081]

[0082] In the formula, is the average coherence coefficient of the interference pattern; N min represents the minimum value of the adaptive window, set to 5.

[0083] S2.2. Extract a local phase block φ of window size N centered on the pixel to be filtered win and use the fast Fourier transform to transform the local phase block φ (m,n) to the frequency domain and obtain the corresponding spectral values of φ (m,n) , with the formula (m,n) being

[0084] S (u,v) = FFT2{φ (m,n)} (3)

[0085] where φ (m,n) represents the local interference phase block of window size N centered on the pixel index in the m-th row and n-th column of the spatial dimension, and S win is the spectral value after transformation of the local phase block, and FFT2{·} represents the two-dimensional fast Fourier transform operation. (u,v) (u,v)

[0086] S2.3. Calculate the main spectral amplitude of the local window The formula is

[0087]

[0088] where thr is the threshold parameter for extracting the main spectrum, which can be adaptively determined according to the spectral amplitude of the local window, with the formula

[0089]

[0090] S2.4. Convolve the extracted main spectral amplitude with the spectral value S (u,v) to extract the main phase component, and obtain the filtered phase of this local phase block through the fast two-dimensional inverse Fourier transform. The formula is

[0091]

[0092] where φ' (m,n) is the filtered local phase block, is the dot product operation, and FFT2 -1 {·} is the fast two-dimensional inverse Fourier transform.

[0093] S2.5. Extract the central pixel of the filtered phase block as the filtered phase of the pixel to be filtered, and loop through steps S2.1 - S2.4 until all pixels in the interference pattern are traversed to obtain the interference pattern filtered in the spatial dimension

[0094] S2.6. Repeat steps S2.1 - S2.5 until all interferograms under the complete interference combination are traversed, and obtain the filtered interferometric phase under the complete interference combination.

[0095] S3. Calculate the corresponding statistical homogeneous pixel neighborhood set for each pixel using the Fast and Simple Homogeneous Pixel Selection algorithm (FaSHPS) based on the time - series amplitude information;

[0096] S3.1. Based on the amplitude vector of the time - series SLC image data, calculate the estimated value of the amplitude mean of the pixel point l to be estimated in the SLC image. The formula is

[0097]

[0098] where is the estimated value of the mean μ(l), A represents the amplitude value of the SLC image, and N SLC represents the number of SLC images.

[0099] S3.2. Calculate the confidence interval at the confidence level 1 - α to screen for homogeneous pixel points. The formula is

[0100]

[0101] where z 1-α / 2 is the percentile of the probability density function of the 1 - α / 2 standard normal distribution.

[0102] S3.3. Select a 15×15 pixel window centered on the pixel point to be estimated in the SLC image. The connected pixels within the pixel window that satisfy the above - mentioned confidence interval are selected as homogeneous pixels; traverse all pixels in the SLC image to obtain the statistical homogeneous pixel neighborhood set corresponding to each pixel. The number of pixels included in the neighborhood set corresponding to each pixel is as Figure 3 shown.

[0103] S4. Based on the filtered interferometric phase and the statistical homogeneous pixel neighborhood set, calculate the coherence coefficient corresponding to the filtered phase for each interferogram, and use the coherence deviation correction method based on the second - type statistical features to correct the deviation of the estimated coherence coefficient to obtain the refined coherence coefficient;

[0104] S4.1. Based on the filtered interferometric phase and the statistical homogeneous pixel neighborhood set Ω SHP , calculate the coherence coefficient corresponding to the filtered phase. The formula is

[0105]

[0106] where represents t mThe complex value of the SLC image at pixel i at time instant, Indicated by t m Time, t n The filtered phase value of the interference pair of the two SLC images at pixel i at time t. Taking an interference pair as an example, the coherence graph corresponding to the filtered phase is as follows: Figure 4 shown.

[0107] S4.2. Using the coherence bias correction method based on the second type of statistical characteristics to correct the estimated coherence Perform deviation correction processing to obtain the refined coherence coefficient The formula is

[0108]

[0109] Where N SHP is the statistically homogeneous pixel neighborhood set Ω SHP The number of pixels. Figure 4 The corresponding deviation corrected coherence diagram is shown in Figure 5 shown.

[0110] S4.3, repeat steps S4.1-S4.2 until all filtered interference images under the full interference combination are traversed to obtain the refined coherence image corresponding to the filtered phase under the full interference combination.

[0111] S5. Combine the spatial dimension filtered interference phase image and the corresponding refined coherence image under the full interference combination, convert them into the time dimension interference phase matrix and coherence matrix for a single pixel, and construct a complex coherence matrix for a single pixel;

[0112] S5.1. Combine all spatial dimension filtered interferometric phase matrices and convert them uniformly into N for a single pixel. SLC ×N SLC dimensional time-dimensional interference phase matrix After traversing all pixels, the time-dimensional interference phase matrix is ​​expressed as:

[0113]

[0114] S5.2. Combine all spatial dimension filtering phase corresponding coherence matrices and convert them uniformly into N for a single pixel. SLC ×N SLC The time-dimensional coherence matrix After traversing all pixels, the time dimension coherence matrix is ​​expressed as:

[0115]

[0116] Taking a pixel as an example, its corresponding time dimension coherence matrix is ​​as follows: Figure 6as shown

[0117] S5.3, Based on the time - dimension interference phase matrix of all pixels and the coherence matrix construct a complex coherence matrix The formula is

[0118]

[0119] S6. Construct a phase optimization model of the maximum likelihood estimator based on the complex coherence matrix, calculate the adaptive weighting factor based on the sigmoid function model, and solve the phase optimization model by combining the eigenvalue decomposition strategy to obtain the final optimized phase.

[0120] S6.1. Based on the complex coherence matrix, combined with the sigmoid function model, calculate the adaptive weight factor. The formula is

[0121]

[0122] In the formula, b represents the inflection point, and k represents the severity of data change before and after the inflection point, which is set to 40.

[0123] S6.2. Based on the adaptive weight factor, build a phase optimization function model, and solve the optimization function model by combining the eigenvalue decomposition strategy. The solution of its optimization model, that is, the optimal phase solution, is expressed as The eigenvector corresponding to the minimum eigenvalue of

[0124] To make the exemplary embodiments of the invention of an interference phase optimization method based on adaptive spatio - temporal filtering fusion more persuasive, we use the traditional PTA and EMI phase optimization algorithms to perform phase optimization processing on the same example data, and conduct a comparative evaluation analysis on the interferograms re - obtained using the optimized phase. Taking the interference pair with a time baseline of 240 days as an example, the original interferogram is as Figure 7 shown, the interferogram obtained by optimizing the phase with the PTA algorithm is as Figure 8 shown, the interferogram obtained by optimizing the phase with the EMI algorithm is as Figure 9 shown, and the interferogram obtained by optimizing the phase with the method of the present invention is as Figure 10As shown. From the comparative analysis in the figure, it can be seen that the interferogram obtained by the optimized phase acquisition corresponding to the method proposed in this patent is smoother, and the noise suppression effect is the most obvious; through the local enlarged view of the dense fringe area on the right, it can be found that the method proposed in this patent can effectively protect the phase information at the dense fringes. In order to quantitatively describe the optimization effect of the algorithm, we obtained the interferograms of the optimized phase estimated by different methods under the full-interference combination, calculated the number of residual points and the sum of the phase differences (sum of the phase difference, SPD) corresponding to the interferograms. The fewer the number of residual points and the smaller the SPD value, the higher the phase quality of the interferogram, and the better the corresponding optimized phase effect. Calculate the improvement percentage of the optimized interferogram compared with the original interferogram, and calculate the mean and standard deviation of the improvement percentages corresponding to all interference pairs. The results are as Figure 11 shown. Taking the residual point index as an example, the mean improvement percentage of the PTA algorithm is 55.52%, and the standard deviation is 8.2%. The mean improvement percentage of the EMI algorithm is 54.05%, and the standard deviation is 9.6%. The improvement percentage of an interferometric phase optimization method based on adaptive spatio-temporal filtering fusion of the present invention is 93.8%, and the standard deviation is 2.0%. The results show that an interferometric phase optimization method based on adaptive spatio-temporal filtering fusion of the present invention has better optimization effect and higher noise robustness.

Claims

1. An interference phase optimization method based on adaptive spatio-temporal filtering fusion, characterized in that, The specific steps are as follows: S1. Collect the time series data of multiple single-view complex images SLC covering the same area, extract the time series amplitude information, and use the time series data of SLC to build a full interference pair combination network. Based on the interference pair combination information, conjugate multiply the main image SLC and the auxiliary image SLC to obtain the differential interference phase under the full interference combination corresponding to the SLC time series data; S2, using an adaptive spatial phase filtering method based on main frequency extraction to perform spatial phase filtering on the differential interference phase under the full interference combination, and obtaining the filtered interference phase under the full interference combination; S2.

1. Adaptive estimation of the filtering window size N according to the coherence coefficient γ of the pixel to be filtered win , and the formula is: In the formula, represents the integer operation, and ε0 is the noise standard deviation of the phase. The calculation method is as follows: In the formula, is the average coherence coefficient of the interference pattern; N min represents the minimum value of the adaptive window, with a value of 5; S2.

2. Extract a local phase block φ with a window size of N centered on the pixel to be filtered win ; and use the following formula to transform the local phase block φ (m,n) to the frequency domain by means of fast Fourier transform, and obtain the spectral value corresponding to φ (m,n) : (m,n) ​ S (u,v) = FFT2{φ (m,n)} (3) where φ (m,n) represents a local interference phase block with a window size of N centered on the pixel index of the m-th row and n-th column spatial dimension, win S (u,v) is the spectral value after the conversion of the local phase block, and FFT2{·} represents a two-dimensional fast Fourier transform operation; S2.

3. Calculate the main spectral amplitude of the local window using the spectral values after the local phase block transformation Where thr is the threshold parameter for extracting the main spectrum of the local window, and is adaptively determined according to the spectrum amplitude of the local window: S2.

4. Multiply the extracted main spectral amplitude by the spectral value S (u,v) to extract the main phase component through convolution multiplication, and obtain the filtered phase of this local phase block through fast two-dimensional inverse Fourier transform: where, φ' (m,n) is the filtered local phase block, is the dot product operation, FFT2 -1 {·} is the fast two-dimensional inverse Fourier transform; S2.

5. Extract the central pixel of the filtered phase block as the filtered phase of the pixel to be filtered, and repeat steps S2.1 - S2.4 until all pixels in the interferogram are traversed to obtain the interferogram filtered in the spatial dimension. S2.6, repeat S2.1 to S2.5 for all interference patterns under the full interference combination, and after the traversal, obtain the filtered interference phase under the full interference combination; S3, using the fast homogeneous pixel identification algorithm FaSHPS combined with the time series amplitude information to calculate the statistical homogeneous pixel neighborhood set corresponding to each pixel of the SLC image pixel by pixel; S4, based on the statistically homogeneous pixel neighborhood set, calculating the coherence coefficients corresponding to all filtered interference phases under the full interference combination, and using the coherence deviation correction method based on the second type of statistical characteristics to perform deviation correction processing on the coherence coefficients to obtain the refined coherence coefficients; S4.

1. Based on the filtered interference phase and the statistically homogeneous pixel neighborhood set Ω SHP , calculate the coherence coefficient corresponding to the filtered phase using the following formula In the formula, represents the complex value of the SLC image at pixel i at time t, m and represents the filtered phase value at pixel i of the interferometric pair formed by combining two SLC images at times m t n and S4.

2. Adopt the coherence deviation correction method based on the second type of statistical feature to correct the deviation of the estimated coherence and obtain the refined coherence coefficient Where N SHP is the number of pixels in the statistically homogeneous pixel neighborhood set Ω SHP ; S4.3, repeating steps S4.1 to S4.2 until all filtered interference images under the full interference combination are traversed to obtain a refined coherence image corresponding to the filtered phase under the full interference combination; S5, combining the spatial dimension filtered interference phase and the refined coherence coefficient under the full interference combination, converting the combined interference phase and coherence coefficient into a time dimension interference phase matrix and a coherence matrix for a single pixel point of the SLC image, and constructing a complex coherence matrix for a single pixel point of the SLC image; S6. Construct a phase optimization model of the maximum likelihood estimator based on the complex coherence matrix, calculate the adaptive weighting factor based on the sigmoid function model, and solve the phase optimization model in combination with the eigenvalue decomposition strategy to obtain the final optimized interference phase.

2. The interference phase optimization method based on adaptive spatio-temporal filtering fusion according to claim 1, characterized in that, Step S3 is specifically as follows: S3.

1. Based on the amplitude vector of the time-series SLC image data, the amplitude mean estimation value of the pixel point l to be estimated in the SLC image is calculated using the following formula: In the formula, is the estimated value of the mean μ(l), A represents the amplitude value of the SLC image, and N SLC represents the number of SLC images; S3.

2. Calculate using the following formula The confidence interval at the confidence level of 1-α is used to screen homogeneous pixel points: where z 1-α2 is the percentile of the probability density function of the 1-α / 2 standard normal distribution; S3.

3. Select a 15×15 pixel window centered on the pixel to be estimated in the SLC image. The connected pixels in the pixel window that satisfy the above confidence interval are selected as homogeneous pixels. Traverse all pixels in the SLC image to obtain the statistically homogeneous pixel neighborhood set corresponding to each pixel.

3. A method for optimizing interference phase based on adaptive spatio-temporal filtering fusion according to claim 1, characterized in that, Step S5 is specifically as follows: S5.

1. Combine all spatial dimension filtered interferometric phase matrices and convert them uniformly into N for a single pixel. SLC ×N SLC dimensional time-dimensional interference phase matrix After traversing all pixels, the time-dimensional interference phase matrix is ​​expressed as: S5.

2. Combine the coherence matrices corresponding to the phases after spatial dimension filtering for all, and uniformly convert them into an N SLC ×N SLC -dimensional temporal dimension coherence matrix After traversing all pixels, the temporal dimension coherence matrix is expressed as S5.

3. Temporal Interference Phase Matrix Based on All Pixel Points and Coherence Matrix Construct Complex Coherence Matrix 4. A method for optimizing interference phase based on adaptive spatio-temporal filtering fusion according to claim 3, characterized in that, Step S6 is specifically as follows: S6.

1. Based on the complex coherence matrix and combined with the sigmoid function model, calculate the adaptive weight factor w using the following formula adaptive :[[]]END]] In the formula, b represents the inflection point, k represents the severity of the data change before and after the inflection point, and is set to 40; S6.

2. Build a phase optimization function model based on the adaptive weight factor, and solve the optimization function model in combination with the eigenvalue decomposition strategy. The solution of the optimization model, that is, the optimal phase solution, is expressed as: The eigenvector corresponding to the minimum eigenvalue of.

Images