Intelligent extraction method for surface deformation of power transmission channel environment

Abstract

The invention discloses a power transmission channel environment surface deformation intelligent extraction method, which belongs to the technical field of power transmission channel deformation extraction, and comprises the following steps: 1, carrying out time sequence difference interference measurement and image preprocessing; 2, constructing a permanent scatterer point selection algorithm and a phase solution algorithm; and step 3, surface deformation extraction. According to another time sequence difference interference method developed for interference decorrelation, all SAR images are grouped according to a baseline minimum principle and are solved in groups respectively, then joint solving is performed by using a least square method, the rank defect problem is solved according to singular value decomposition, and finally deformation information is obtained. According to the method, the problem of correlation loss is solved to a certain extent, the requirement for the number of SAR images is reduced, the density of the selected high-coherence points is greatly increased, and the method can be widely applied to remote mountainous areas.

Description

Technical Field

[0001] The present invention belongs to the technical field of deformation extraction of transmission channels, and particularly relates to an intelligent extraction method for surface deformation of the transmission channel environment. Background Art

[0002] Surface deformation is a phenomenon of displacement on the Earth's surface caused by natural or human factors. Related geological disasters caused by surface deformation have become one of the common concerns at home and abroad. Geological disasters refer to geological phenomena that endanger human life and property and damage the human living environment caused by natural factors or human activities, including earthquakes, volcanoes, land subsidence, landslides, and debris flows. Due to the numerous rivers, complex terrain and landforms, uneven temporal and spatial distribution of rainfall in China, and the increasing intensity of human engineering and economic activities, China has become one of the countries with the most developed geological disasters and the most serious hazards in the world. Traditional surface deformation monitoring technologies are mainly GPS measurement and geodetic leveling. These methods monitor surface changes by arranging discrete points in the deformation area. Although the monitoring accuracy is high, for the extraction of large-area surface deformation, it is difficult to arrange a large number of discrete points. The observation results can only reflect the discrete deformation amounts of each monitoring station, but cannot reflect the overall deformation result of the entire space, which is not conducive to disaster prediction and assessment. Moreover, the observation period is long, and the consumption of human and material resources is large. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art, and provide an intelligent extraction method for surface deformation of the transmission channel environment, which analyzes the phase using spatial correlation and does not use a priori models, meets the requirements of deformation extraction accuracy, and improves the density of PS points. Good deformation results can also be obtained in non-urban areas with severe correlation, solving the problems in the above background art.

[0004] The purpose of the present invention is achieved as follows: An intelligent extraction method for surface deformation of the transmission channel environment, which includes the following steps: Step 1, perform time series differential interferometric measurement and image preprocessing; Step 2, construct a permanent scatterer point selection algorithm and a phase solution algorithm; Step 3, extract surface deformation. Another time series differential interferometry method developed for interferometric decorrelation groups all SAR images according to the principle of minimum baseline, solves within each group respectively, then jointly solves using the least squares method, and solves the rank deficiency problem according to singular value decomposition. Finally, deformation information is obtained. This method not only solves the decorrelation problem to a certain extent, but also reduces the requirements for the number of SAR images, greatly increasing the density of selected high-coherence points, enabling this method to be widely applied to remote mountainous areas.

[0005] The image preprocessing includes common master image selection and image registration. The common master image selection includes time baseline T, spatial baseline B, Doppler centroid frequency difference Fpc, and thermal noise. When there is deformation, the influence of time baseline T, spatial baseline B, and Doppler centroid frequency difference Fpc is considered when selecting the master image, and the problem of master image selection is transformed into an optimization problem of a mathematical model, that is,

[0006]

[0007] In the formula, γ temporal is the decorrelation of the time baseline, γ spatial is the decorrelation of the spatial baseline, γ doppler is the decorrelation of the Doppler centroid, γ thermal is the decorrelation of the thermal noise, and γ total is the total coherence;

[0008] Among them,

[0009]

[0010] In the formula, γ represents the correlation of the corresponding factor, and the superscript C represents the critical value of the corresponding parameter. If a parameter exceeds this critical value, the decorrelation interference fails; different radar data settings have their corresponding critical values. According to the formula, if the thermal noise is a constant, the image that maximizes the sum of the coherences of all interferograms is selected as the common master image, that is,

[0011]

[0012] N represents the number of interference pairs. To maximize the total coherence as much as possible, multiple master images are allowed to form multiple master image interference pairs to improve the utilization rate of the images and increase the total coherence of the interference pairs.

[0013] The image registration includes using the central pixels of the master and slave images as homologous points, and using precise orbit parameters, combined with the ellipsoid equation, Doppler equation, and slant range equation, to calculate the coordinates of the homologous points. Finally, the difference between the two coordinates can obtain the initial offset. The expressions of the three equations are,

[0014] Doppler equation:

[0015]

[0016] Slant range equation:

[0017]

[0018] Ellipsoid equation:

[0019]

[0020] Among them, is the position vector of the radar, is the position vector of the pixel, is the slant range vector between the pixel and the radar, v represents the speed of light, and t represents the transmission time of the signal to the ground object corresponding to the central pixel;

[0021] The steps of the image registration include:

[0022] a) For the homologous point P on the master image, calculate its vector coordinates (x, y, z) in the ellipsoidal reference system;

[0023] b) Using the coordinates of the homologous points on the master image and combining with the Doppler equation, solve the coordinates P of the homologous points on the slave image;

[0024] c) Calculate the overall initial offset offset of the image by taking the difference between the homologous points in the master and slave images;

[0025] The implementation process includes: Select M points from the reference image as the initial control points, and respectively establish an m×m target window centered on them. m is usually taken as an odd number;

[0026] According to the initial offset obtained in the rough registration and the row and column values of the control points on the reference image, calculate the coordinates of the homologous points of the control points on the image to be registered, and then establish a search window centered on the homologous points; The solid line box is for the target window in the reference image, and the solid square points are the control points; The dotted line box is for the search window, the solid line box is the target window, and the solid square points are the corresponding points;

[0027] Place the target window in the search window, starting from the top-left pixel, calculate the intensity correlation value between the homologous points pixel by pixel. After the calculation, take the maximum intensity correlation value, and use the corresponding point as the homologous point of the control point in the reference image, and record its row and column values. Furthermore, the row and column offset between the homologous point and the control point can be calculated.

[0028] The temporal differential interferometry includes the following steps,

[0029] 1) For K + 1 SAR images, based on the optimal principles of time baseline, spatial baseline, and Doppler center frequency, and at the same time according to the specific situation of the study area, use an algorithm to select the master image. Based on the master image, register and process the other K images and perform interference to obtain K interference pairs;

[0030] 2) Perform differential interference on the obtained K interferograms and the external DEM to obtain K differential interferograms;

[0031] 3) On the K + 1 SAR images that have undergone radiometric calibration and registration, use a suitable algorithm to select points;

[0032] 4) Obtain the point differential interferometric phase set from K differential interferometric phase maps and the selected points;

[0033] 5) Analyze the composition of the differential interferometric phase and model the differential interferometric phase;

[0034] 6) According to the point differential interferometric phase set, use a suitable algorithm to solve the differential interferometric phase function model to obtain the deformation rate, elevation error, and atmospheric phase of each point;

[0035] 7) Correct and comprehensively process the deformation rates of the points in each obtained differential interferogram to obtain the final deformation time series.

[0036] The permanent scatterer point selection algorithm includes setting a window centered on a certain pixel. The correlation coefficient γ of this pixel can use the information of other pixels in the window. The formula is

[0037]

[0038] where M and S respectively represent the pixel blocks corresponding to the master and slave images in the same window, * is the complex conjugate operator, and the correlation coefficient γ ∈ [0, 1] indicates the noise of the interferometric phase. There are N interferometric pairs. For any pixel, calculate its correlation coefficient time series according to the formula: γ1, γ2, …, γ N , and then calculate the average value. The formula is

[0039]

[0040] Set a threshold T γ , and calculate for each pixel. If take this point as a candidate point; the purpose is to filter out points with extremely serious decorrelation, which is used to narrow the selection range and improve the selection efficiency in the subsequent refined selection process;

[0041] The phase unwrapping algorithm includes obtaining a principal value of the phase, the value range is between -π and π, and the true phase is hidden by an integer multiple of 2π. Denote φ(m) as the true phase, as the wrapped phase, and introduce the phase wrapping operator The wrapping formula is

[0042]

[0043] Phase unwrapping is used to restore to φ(m) through an algorithm.

[0044] A new PS point selection algorithm is proposed by combining the PS point selection algorithm and the phase unwrapping algorithm, which overcomes the limitations of the traditional PS selection algorithm. Different from the traditional method based on the amplitude deviation index, the principle of this algorithm for selecting PS points is based on phase stability, enabling a certain number of high-quality PS points to be selected even in remote mountainous areas. The accuracy and robustness of this algorithm are verified through comparative experiments: 3D phase unwrapping further increases the integration path on the basis of 2D phase unwrapping, while enhancing the ability to identify singular points.

[0045] The surface deformation extraction includes selecting master and slave images with a small time base and a short spatial baseline, spectral filtering in the range direction, and removing the non-coincident part of the azimuth Doppler frequency; the obtained points are used for de-correlated phase low-pass filtering; the amplitude difference separation index D is used. ΔA Select SDFP points, and the formula is

[0046]

[0047] In the formula, σ ΔA represents the mean square deviation of the amplitudes of N interferometric pairs, μ A represents the mean of the amplitudes of N interferometric pairs, A m,i,x 、A s,i,x represent the amplitudes of the pixel x in the master and slave images of the interferometric pair numbered i in sequence; the threshold of the value of D ΔA is set to 0.5 - 0.6 to reduce the data volume and select high-coherence candidate points; for any pixel, the atmospheric phase and the deformation phase are spatially correlated within a circle with a radius of L, and the thermal noise phase is spatially uncorrelated within a semi-circle. For high-coherence points, after filtering with the CLAP filter, it is expressed as follows:

[0048]

[0049] In the formula, represents the differential phase, represents the spatially correlated phase after filtering, K ε,x represents the DEM error phase caused by the vertical baseline error; it is solved using the temporal coherence coefficient, and the formula is

[0050]

[0051] Use the solution space search method to find to make γ x the maximum as the optimal solution, and then through the optimal solution, calculate the actual value of the elevation residual error, and at the same time perform iterative calculation filtering.

[0052] Advantages of the present invention: It provides a new idea for extracting deformation. Another time-series differential interferometry method developed for interferometric decorrelation groups all SAR images according to the principle of minimum baseline, solves within each group respectively, then jointly solves using the least squares method, and solves the rank deficiency problem by singular value decomposition. Finally, deformation information is obtained. This method not only solves the decorrelation problem to a certain extent, but also reduces the requirements for the number of SAR images, greatly increasing the density of selected high-coherence points, enabling the method to be widely applied to remote mountainous areas. By combining the PS point selection algorithm and the phase unwrapping algorithm, a new PS point selection algorithm is proposed compared to the limitations of the traditional PS selection algorithm. Different from the traditional method based on the amplitude deviation index for selection, the principle of this algorithm for selecting PS points is based on phase stability, enabling a certain number of high-quality PS points to be selected even in remote mountainous areas. And its accuracy and robustness are verified through comparative experiments: The three-dimensional phase unwrapping further increases the integration path on the basis of two-dimensional phase unwrapping, and at the same time enhances the ability to identify singular points. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is the PS-DInSAR data processing flow chart of the present invention;

[0054] Figure 2 is the SBAS-DInSAR data processing flow chart of the present invention;

[0055] Figure 3 is the StaMPS-MTI data processing flow chart of the present invention;

[0056] Figure 4 is the schematic diagram of phase unwrapping of the present invention;

[0057] Figure 5 is the StaMPS processing flow chart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] The following further describes the present invention in detail with reference to the drawings. It should be noted that this is only for more clearly explaining and interpreting the present invention.

[0059] As Figures 1-5 shown, this embodiment discloses an intelligent extraction method for surface deformation in the transmission channel environment, which includes the following steps:

[0060] Step 1: Conduct time series differential interferometric measurement and image preprocessing; Permanent Scatterer (PS) refers to hard targets with stable scattering characteristics and strong radar wave reflection capabilities, such as artificial buildings, bare rocks, etc. In the power transmission corridor, it of course also includes transmission towers. Due to their stable scattering characteristics and strong reflection capabilities, these ground objects have high signal-to-noise ratios and good temporal high coherence. PS-DInSAR uses the permanent scatterers in the study area, analyzes the interferometric phase sets of them through multi-temporal analysis, extracts their deformation information, and thus obtains the surface deformation situation of the study area.

[0061] The time series differential interferometric measurement includes the following steps:

[0062] 1) For K + 1 SAR images, based on the principles of optimal time baseline, spatial baseline, and Doppler center frequency, and according to the specific situation of the study area, select the master image using an algorithm. Based on the master image, register and process the other K images and perform interference to obtain K interference pairs;

[0063] 2) Perform differential interference on the obtained K interferograms and an external DEM to obtain K differential interferograms;

[0064] 3) On the K + 1 SAR images that have undergone radiometric calibration and registration, select points using a suitable algorithm;

[0065] 4) Obtain the point differential interferometric phase set from the K differential interferometric phase diagrams and the selected points;

[0066] 5) Analyze the composition of the differential interferometric phase and model the differential interferometric phase;

[0067] 6) According to the point differential interferometric phase set, use a suitable algorithm to solve the differential interferometric phase function model to obtain the deformation rate, elevation error, and atmospheric phase of each point;

[0068] 7) Correct and comprehensively process the deformation rates of the points in each obtained differential interferogram to obtain the final deformation time series.

[0069] The image preprocessing includes the selection of a common master image and image registration. Before performing interferometry on two SAR images, the first thing to do is image registration. The purpose of registration is to ensure that the pixel points at the same position in the two matrices correspond to the same target point on the ground, so that the interferometric phase information obtained after interference has a higher signal-to-noise ratio. After accurately registering the two SAR images, the complex values at the corresponding positions are multiplied conjugately, i.e., interfered, to obtain an interferometric phase diagram reflecting the range information of the radar's line of sight to the ground. Then, based on the formula, the terrain and deformation information of the study area can be obtained by inverting the unwrapped interferometric phase. Therefore, the accuracy of SAR image registration directly affects the quality of subsequent data processing, such as phase unwrapping, extraction of deformation information, etc. The key to registration is how to determine the relative offset of the corresponding pixels between the two images, and then resample the secondary image based on this to ensure that the pixels of the two images can correspond well.

[0070] For the precise registration of images based on correlation measures, the rough registration of images introduced earlier only obtains the overall offset of the images, and its accuracy is only at the pixel level, which cannot meet the sub-pixel level required for interferometry. Sub-pixel registration is more precise. It no longer processes the entire image, but calculates the offset locally. The local offsets are different, and the offsets are no longer integers, and there are also different offsets in the azimuth and range directions. In addition, when resampling SAR images, it is required to resample the real and imaginary parts separately. Commonly used two-dimensional resampling methods include: bilinear interpolation, bicubic interpolation, and nearest-neighbor interpolation. Among them, the bicubic interpolation method takes the most time, followed by the bilinear interpolation method. Therefore, in practical applications, which method to choose should be determined according to the requirements of accuracy and efficiency.

[0071] The selection of a common master image includes the temporal baseline T, spatial baseline B, Doppler centroid frequency difference Fpc, and thermal noise. When there is deformation, the influence of the temporal baseline T, spatial baseline B, and Doppler centroid frequency difference Fpc should be considered when selecting the master image. The problem of selecting the master image is transformed into an optimization problem of a mathematical model, that is,

[0072]

[0073] In the formula, γ temporal is the decorrelation of the temporal baseline, γ spatial is the decorrelation of the spatial baseline, γ doppler is the decorrelation of the Doppler centroid, γ thermal is the decorrelation of thermal noise, and γ total is the total coherence;

[0074] Among them,

[0075]

[0076] In the formula, γ represents the correlation of the corresponding factor, and the superscript C represents the critical value of the corresponding parameter. If a parameter exceeds this critical value, the decorrelation interference fails; different radar data settings have their corresponding critical values. According to the formula, if the thermal noise is a constant, the image that maximizes the sum of the coherences of all interferograms is selected as the common master image, that is,

[0077]

[0078] N represents the number of interferometric pairs. To maximize the total coherence as much as possible, multiple master images are allowed to form multi-master image interferometric pairs to improve the utilization rate of the images and increase the total coherence of the interferometric pairs.

[0079] The image registration includes taking the central pixels of the master and slave images as homologous points, using precise orbit parameters, and combining the ellipsoid equation, Doppler equation, and slant range equation to calculate the coordinates of the homologous points. Finally, the initial offset can be obtained by taking the difference between the two coordinates. The expressions of the three equations are as follows:

[0080] Doppler equation:

[0081]

[0082] Slant range equation:

[0083]

[0084] Ellipsoid equation:

[0085]

[0086] Among them, is the position vector of the radar, is the position vector of the pixel, is the slant range vector between the pixel and the radar. v represents the speed of light, and t represents the transmission time of the signal to the ground object corresponding to the central pixel;

[0087] The steps of the image registration include:

[0088] b) For the homologous point P on the master image, calculate its vector coordinates (x, y, z) in the ellipsoid reference system;

[0089] b) Using the coordinates of the homologous point on the master image and combining the Doppler equation, solve the coordinates of the homologous point P on the slave image;

[0090] c) Calculate the initial offset offset of the whole image by taking the difference between the homologous points in the master and slave images;

[0091] The implementation process includes: selecting M points from the reference image as initial control points, and respectively establishing an m×m target window centered on them. m is usually taken as an odd number;

[0092] According to the initial offset obtained in the rough registration and the row and column values of the control points on the reference image, calculate the coordinates of the corresponding points of the control points on the image to be registered, and then establish a search window centered on the corresponding points; the solid line frame is used for the target window in the reference image, and the solid square points are the control points; the dashed line frame is used for the search window, the solid line frame is the target window, and the solid square points are the corresponding points;

[0093] Place the target window in the search window, starting from the top-left pixel, calculate the intensity correlation value between the corresponding points pixel by pixel. After the calculation is completed, take the maximum intensity correlation value, and use the corresponding point as the corresponding point of the control point in the reference image, and record its row and column values. Furthermore, the row and column offset between the corresponding point and the control point can be calculated.

[0094] Step 2: Construct a permanent scatterer point selection algorithm and a phase unwrapping algorithm;

[0095] The permanent scatterer point selection algorithm includes setting a window centered on a certain pixel. The correlation coefficient γ of this pixel can use the information of other pixels in the window. The formula is

[0096]

[0097] In the formula, M and S respectively represent the pixel blocks corresponding to the master and slave images in the same window, * is the complex conjugate operator, and the correlation coefficient γ ∈ [0, 1] indicates the noise of the interference phase. There are N interference pairs. For any pixel, calculate its correlation coefficient time series according to the formula: γ1, γ2, …, γ N , and then calculate the average value. The formula is

[0098]

[0099] Set a threshold T γ , and calculate for each pixel. If regard this point as a candidate point; the purpose is to filter out the points with extremely serious decorrelation, and in the subsequent refinement process, it is used to narrow the selection range and improve the selection efficiency;

[0100] The phase unwrapping algorithm includes obtaining a principal value of the phase, and the value range is between -π and π, and the true phase is hidden by an integer multiple of 2π. Denote φ(m) as the true phase as the wrapped phase, and introduce the phase wrapping operator The wrapping formula is

[0101]

[0102] Phase unwrapping is used to restore to φ(m) through an algorithm.

[0103] PS-DInSAR is used to achieve the need for a sufficient number of SAR images. At least 25 SAR images are required to obtain high-precision results, and the increase in the number of images correspondingly leads to a decrease in processing efficiency. Having a sufficient number of high-quality PS points is the key to the PS-DInSAR technique. In mountainous areas with sufficient vegetation cover, extracting a sufficient number of PS points to solve the differential interference phase model enables this method to be applied to such areas and has a wider application. For the PS point selection algorithm, there are the following four traditional algorithms: the coherence coefficient method, the amplitude deviation index threshold method, the multi-polarization image selection method, and the phase deviation threshold method. However, high-quality PS points require meeting multiple characteristics simultaneously, while these methods only use a certain characteristic of the PS points and may be affected by the signals of nearby points during the PS point selection process. These factors may cause missed and misjudged PS points, resulting in insufficient quantity and quality of the selected PS points. This method greatly enhances the quantity and quality.

[0104] The networking of PS points and determining the neighborhood relationship between PS point pairs is one of the key issues in solving the PS point neighborhood differential phase model. The method of establishing a two-dimensional Delaunay triangulation network through the image plane coordinates of PS points is used to construct the PS network, solve the problem of inconsistent resolutions in the range and azimuth directions of SAR images, and thus reduce the error of the deformation result.

[0105] To solve the limitations of traditional DInSAR technology in terms of time and space baselines, as well as the limitations brought by decorrelation, atmospheric delay, etc., the small baseline subset differential interferometry method of the present invention, abbreviated as SBAS-DInSAR, is based on the principle of smaller space and time baselines, forms differential interference pairs pairwise between images, selects high-coherence target points, establishes a differential interference phase model, and uses a suitable algorithm to solve the model, finally obtaining an accurate deformation result.

[0106] Step 3: Surface deformation extraction;

[0107] The surface deformation extraction includes selecting master and slave images with a small time base and a short space baseline, performing spectral filtering in the range direction and removing the non-coincident part of the azimuth Doppler frequency; the obtained points are used as the low-pass filtering of the decorrelated phase; using the amplitude difference separation index D ΔA Select SDFP points, and the formula is

[0108]

[0109] In the formula, σ ΔA represents the mean square deviation of the amplitudes of N interference pairs, μ A represents the mean of the amplitudes of N interference pairs, A m,i,x 、As,i,x successively represent the amplitudes of the pixel x in the master and slave images of the interference pair numbered i; D ΔA The value of D is set with a threshold of 0.5 - 0.6 to reduce the data volume and select high-coherence candidate points; for any pixel, the atmospheric phase and the deformation phase are spatially correlated within a circle with a radius of L, and the thermal noise phase is spatially uncorrelated within a semi-circle. For high-coherence points, after filtering with the CLAP filter, it is expressed as follows:

[0110]

[0111] In the formula, represents the differential phase, represents the spatially correlated phase after filtering, K ε,x represents the DEM error phase caused by the vertical baseline error; solved using the temporal coherence coefficient, the formula is

[0112]

[0113] Use the solution space search method to find such that γ x is maximized as the optimal solution, and then through the optimal solution, calculate the actual value of the elevation residual error, and at the same time perform iterative calculation filtering.

[0114] In the actual processing process, the amplitude deviation index threshold is set to 0.5, and pixels smaller than this threshold are used as PS points. However, in some research areas, points with an amplitude deviation index greater than 0.5 may also be selected as PS points.

[0115] In the process of identifying PS points by the traditional PS algorithm, a time-based deformation function model is used. This model requires prior knowledge of the deformation rate in the research area, and this algorithm often assumes that the surface deformation process is uniform or periodic. However, prior knowledge of the deformation rate is not available for every research area, and moreover, the deformation characteristics are often different at different times. Therefore, for some research areas with non-uniform and anisotropic deformation characteristics, the traditional PS algorithm cannot measure.

[0116] The present invention uses the traditional PS algorithm for the PS-DInSAR process to ensure the selection of reliable PS points and the convergence of the algorithm, so as to obtain accurate deformation results. Generally, the number of images is required to be more than 25. By using the amplitude characteristics of the target points and the phase space correlation characteristics of the interferogram, the points with stable phase in the time series interferometric data are identified, and then the deformation information is inversely calculated through these points. Different from the traditional algorithm, in the selection process, an iterative method is adopted to obtain a new phase estimate value, thereby reducing the false alarm rate of the selected PS points. The stable operation of the transmission line affects the safety of the entire transmission network. However, due to geographical conditions, a large part of the transmission channels pass through areas prone to geological disasters, such as debris flows, collapses, landslides, ground collapses, etc. These disasters seriously threaten the safety of transmission towers. Therefore, it is necessary to monitor and warn of geological disasters in the transmission channels. When these disasters occur, there is a common feature, that is, surface deformation. Therefore, the warning and assessment of disasters are transformed into the monitoring of surface deformation of the transmission channels.

[0117] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and its concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for intelligent extraction of ground deformation in a power transmission channel environment, characterized in that: The following steps are involved: S1, performing time-series differential interferometry measurement and image preprocessing; S2, construct permanent scatterer point selection algorithm and phase solution algorithm; S3. Surface deformation extraction.

2. The method for intelligently extracting ground deformation in a power transmission channel environment according to claim 1 is characterized in that: The image preprocessing includes public main image selection and image registration. The public main image selection includes time baseline T, space baseline B, Doppler centroid frequency difference Fpc and thermal noise. When deformation occurs, the influence of time baseline T, space baseline B and Doppler centroid frequency difference Fpc is considered when selecting the main image, and the main image selection problem is converted into an optimization problem of a mathematical model, that is, In the formula, γ temporal is the time baseline decorrelation, γ spatial is the spatial baseline decorrelation, γ doppler is the Doppler centroid decorrelation, γ thermal is the thermal noise decorrelation, γ total is the total coherence; in, In the formula, γ represents the correlation of the corresponding factors, and the superscript C represents the critical value of the corresponding parameter. If a parameter exceeds this critical value, the de-correlation interference fails. Different radar data are set with corresponding critical values. According to the formula, if the thermal noise is a constant, the image with the largest sum of the coherence of all interference patterns is selected as the common main image, that is, N represents the number of interference pairs. In order to maximize the overall coherence, multiple main images are allowed to form multiple main image interference pairs, so as to improve the utilization rate of the images and the overall coherence of the interference pairs.

3. The method for intelligently extracting ground deformation in a power transmission channel environment according to claim 2 is characterized in that: The image registration includes taking the central pixels of the master and slave images as the same-name points, using the precise orbit parameters, the combined ellipsoid equation, the Doppler equation and the slant range equation to obtain the coordinates of the same-name points, and finally the initial offset can be obtained by subtracting the two coordinates. The expressions of the three equations are: Doppler equation: Slope range equation: Ellipsoid equation: in, is the radar's position vector, is the position vector of the pixel, is the slant distance vector between the pixel and the radar, v represents the speed of light, and t represents the transmission time of the signal to the ground object corresponding to the central pixel; The steps of image registration include: a) For the same-name point P on the main image, calculate its vector coordinates (x, y, z) in the ellipsoid reference system; b) Using the coordinates of the same-name points in the main image and the Doppler equation, solve the coordinates P of the same-name points in the slave image; c) Subtract the same-name points in the master and slave images to obtain the initial offset of the entire image; The implementation process includes: selecting M points from the reference image as initial control points, and establishing an m×m target window with them as the center, where m is usually an odd number; According to the initial offset obtained in the rough registration and the row and column values ​​of the control points on the reference image, the coordinates of the same-name points on the image to be registered are calculated, and then a search window is established with the same-name points as the center; the solid-line frame is used for the target window in the reference image, and the solid square points are the control points; the dotted-line frame is used for the search window, the solid-line frame is the target window, and the solid square points are the corresponding points; Place the target window in the search window, start from the upper left pixel, and calculate the intensity correlation value between the same-name points pixel by pixel. After the calculation is completed, take the largest intensity correlation value, and use its corresponding point as the same-name point of the control point in the reference image, and record its row and column values, and then calculate the row and column offset between the same-name point and the control point.

4. The method for intelligently extracting ground deformation in a power transmission channel environment according to claim 1, characterized in that: The time-series differential interferometry The following steps are included: 1) For K+1 SAR images, according to the optimal principles of time baseline, space baseline and Doppler center frequency, and according to the specific conditions of the study area, an algorithm is used to select the main image, and based on the main image, the other K images are registered and interfered to obtain K interference pairs; 2) Perform differential interference on the obtained K interference patterns and the external DEM to obtain K differential interference patterns; 3) Select points using appropriate algorithms on the K+1 SAR images that have been radiometrically calibrated and registered; 4) Obtaining a point differential interference phase set from K amplitude differential interference phase images and the selected points; 5) Analyze the differential interference phase composition and model the differential interference phase; 6) According to the point differential interferometry phase set, a suitable algorithm is used to solve the differential interferometry phase function model to obtain the deformation rate, elevation error and atmospheric phase of each point; 7) The deformation rates of the points in each differential interference pattern are corrected and comprehensively processed to obtain the final deformation time series.

5. The method for intelligent extraction of ground deformation in a power transmission channel environment according to claim 1 is characterized by: The permanent scatterer point selection algorithm includes setting a window with a certain pixel as the center. The correlation coefficient γ of the pixel can use the information of other pixels in the window. The formula is: Where M and S represent the pixel blocks corresponding to the main and auxiliary images in the same window, respectively. * is the complex conjugate operator. The correlation coefficient γ∈[0,1] indicates the noise of the interference phase. There are N interference pairs. For any pixel, its correlation coefficient time series is calculated according to the formula: γ1,γ2,…,γ N Then calculate the average value, the formula is, Set a threshold T γ , calculate the value of each pixel like The point is taken as a candidate point; The purpose is to filter out the points with extremely serious loss of correlation, which will be used to narrow the selection range and improve the selection efficiency in the subsequent selection process; The phase solution algorithm includes obtaining a main value of the phase, the value range is between -π and π, and the real phase is hidden by an integer 2π, denoted by φ(m) as the real phase, For the wrapped phase, the phase wrapping operator is introduced The winding formula is, Phase unwrapping is used to convert Restore to φ(m).

6. The method for intelligently extracting ground deformation in a power transmission channel environment according to claim 1, characterized in that: The surface deformation extraction includes selecting the main and auxiliary images with hourly base and short spatial baseline, filtering the spectrum in the range direction and removing the non-overlapping parts of the Doppler frequency in the azimuth direction; the obtained points are used as phase low-pass filtering for de-correlation; using amplitude difference to separate the difference separation index D ΔA Select the SDFP point, the formula is, In the formula, σ ΔA represents the mean square error of the amplitudes of N interference pairs, μ A represents the mean value of the amplitude of N interference pairs, A m,i,x , A s,i,x represents the amplitude of pixel x in the main and auxiliary images of the interference pair numbered i; D ΔA The value of is set to a threshold of 0.5-0.6 to reduce the amount of data and select high coherence candidate points; for any pixel, the atmospheric phase and deformation phase are spatially correlated within a circle with a radius of L, and the thermal noise phase is spatially uncorrelated within the semicircle. For high coherence points, the CLAP filter is used to filter and is expressed as follows: In the formula, represents the differential phase, represents the spatial correlation phase after filtering, K ε,x Represents the DEM error phase caused by the vertical baseline error; it is solved using the time series coherence coefficient, and the formula is: Using solution space search method to find Make γ x The maximum is the optimal solution, and then the actual value of the elevation residual error is calculated through the optimal solution, and the filter is calculated through iterative calculation.

Images