A foundation SAR interferometric phase optimization filtering method

Abstract

The application relates to the technical field of radar signals and discloses a ground-based SAR interferometric phase optimization filtering method.The method receives a data set containing multiple interferometric phase images and corresponding deformation measurement values, and performs quality enhancement processing on the interferometric phase data.Based on the enhanced data, the phase reliability score of each pixel position is calculated, and the deformation correlation index thereof is evaluated in combination with the deformation data.According to the deformation correlation index and the phase statistical characteristics, the screening priority of each pixel position is determined, and high-priority pixel positions are selected accordingly.The phases of the preferred pixels and the deformation data are used to train a phase filtering model, and finally the model is applied to perform filtering operation on the interferometric phase image to be processed.The application overcomes the balance problem between noise suppression and signal fidelity in the traditional filtering method, and improves the filtering effect and the application value thereof in deformation monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of radar signal, and particularly to a ground-based SAR interferometric phase optimization filtering method. BACKGROUND

[0002] The current ground-based SAR interferometric phase filtering mainly uses a filter with fixed parameters or a simple weighting method based on coherence. The prior art has a single dimension for quality evaluation of interferometric phase data, and fails to combine external deformation measurement data for joint analysis. The pixel screening standard is empirical, and lacks quantitative evaluation of pixel reliability and deformation correlation. The filtering model parameters are fixed, and cannot adapt to changes in signal characteristics in different regions. The existing method needs to solve key technical problems such as multi-source data fusion, pixel quality evaluation, and adaptive filtering.

[0003] The traditional interferometric phase filtering method has obvious deficiencies in accuracy and adaptability. The quality enhancement processing method is simple, and the phase noise suppression effect is limited. The reliability evaluation index is single, and cannot fully reflect the quality of the phase signal. The deformation correlation analysis is linearized, and the recognition accuracy of complex deformation patterns is low. The priority determination rule is fixed, and the risk of missing important pixels is high. The training sample selection of the filtering model is biased, and the model generalization ability is insufficient. The filtering operation parameter adjustment lags behind, and cannot optimize the processing effect in real time. SUMMARY

[0004] The present application relates to the technical field of radar signal, and particularly to a ground-based SAR interferometric phase optimization filtering method.

[0005] To achieve the above-mentioned purpose, the present application provides a ground-based SAR interferometric phase optimization filtering method, which comprises:

[0006] receiving a set of interferometric phase data and a set of deformation data collected by a ground-based SAR system, the set of interferometric phase data containing a plurality of interferometric phase images, each interferometric phase image corresponding to a deformation measurement value;

[0007] performing quality enhancement processing on the set of interferometric phase data to generate an enhanced set of interferometric phase data;

[0008] based on the enhanced set of interferometric phase data, calculating a phase reliability score for each pixel position;

[0009] using the phase reliability score and the set of deformation data, evaluating a deformation correlation index for each pixel position;

[0010] determining a screening priority for each pixel position according to the deformation correlation index and the phase statistical characteristics;

[0011] According to the screening priority, a high-priority pixel position is selected; and based on phase data and deformation data of the high-priority pixel position, a phase filtering model is trained.

[0012] The phase filtering model is applied to perform a phase filtering operation on a to-be-processed interference phase image.

[0013] Preferably, the quality enhancement processing on the set of interference phase data comprises:

[0014] The phase value of each interference phase image is adjusted in range, so that the phase value is normalized to a standard circumferential range;

[0015] An abnormal value point in the phase data is identified and corrected using a local smoothing algorithm;

[0016] The corrected interference phase image is subjected to integrity checking, and an enhanced set of interference phase data is output.

[0017] Preferably, the calculation of the phase reliability score of each pixel position comprises:

[0018] For each pixel position, the phase values of all interference phase images at the position are aggregated to form a phase value sequence;

[0019] The coefficient of variation of the phase value sequence is calculated as a phase instability measure;

[0020] The phase instability measure is inversely proportional converted to obtain a phase reliability score;

[0021] The inversely proportional conversion of the phase instability measure comprises:

[0022] The reciprocal of the phase instability measure is taken;

[0023] The reciprocal is normalized to adjust the numerical range to between zero and one;

[0024] The normalized value is taken as the phase reliability score.

[0025] Preferably, the evaluation of the deformation correlation index of each pixel position comprises:

[0026] The phase value sequence and the corresponding deformation value sequence of each pixel position are extracted;

[0027] The Pearson correlation coefficient of the phase value sequence and the deformation value sequence is calculated as an initial deformation correlation;

[0028] The initial deformation correlation is weighted and adjusted based on the phase reliability score to obtain the deformation correlation index.

[0029] Preferably, the determination of the screening priority of each pixel position comprises:

[0030] Analyze the probability distribution of the phase value sequence at each pixel location to generate a phase probability distribution map; analyze the probability distribution of the deformation value sequence to generate a deformation probability distribution map.

[0031] Calculate the Jaccard similarity between the phase probability distribution map and the deformation probability distribution map, and use it as the distribution consistency score;

[0032] Multiply the distribution consistency score by the deformation correlation index to obtain the screening priority;

[0033] The probability distribution of the phase value sequence at each pixel location is analyzed, including:

[0034] The numerical range of the phase value sequence is divided into multiple equally wide intervals;

[0035] Count the number of times the phase value appears in each interval of the phase value sequence;

[0036] Generate a phase probability distribution map in the form of a histogram based on the interval and the frequency of occurrence.

[0037] Preferably, selecting the high-priority pixel position includes:

[0038] Set a priority threshold and mark the pixel positions with a priority greater than the priority threshold as candidate pixel positions; perform spatial clustering analysis on the candidate pixel positions to remove isolated points and retain the pixel positions within the connected regions as high-priority pixel positions.

[0039] Preferably, the trained phase filter model includes:

[0040] Calculate training weights based on the filtering priority of each high-priority pixel position;

[0041] The phase data and deformation data of the high-priority pixel locations are combined into training samples, and the training weights are assigned to each training sample.

[0042] Build a neural network model and use a weighted loss function during training, where the loss of each training sample is multiplied by its training weight;

[0043] The value of the weighted loss function is reduced through iterative optimization until the model parameters converge, thus obtaining the trained phase filter model.

[0044] Preferably, the phase filtering operation performed on the interferometric phase image to be processed using the phase filtering model includes:

[0045] Input the phase data of the interferometric phase image to be processed into the phase filtering model; the phase filtering model outputs the predicted deformation value.

[0046] The phase data is adjusted based on the predicted deformation values ​​to generate a filtered interferometric phase image.

[0047] Preferably, the generation of the enhanced interferometric phase data set further includes:

[0048] Each interferometric phase image is decomposed into multiple scales to extract phase features at different scales;

[0049] By fusing multi-scale phase features, an enhanced interferometric phase data set is generated.

[0050] Preferably, the present invention further includes a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the ground-based SAR interferometric phase optimization filtering method as described above.

[0051] Compared with the prior art, the beneficial effects of the present invention are:

[0052] The interferometric phase data set is enhanced by quality enhancement processing to generate an enhanced interferometric phase data set. The enhancement process employs an adaptive filtering algorithm, adjusting the filtering parameters based on local statistical characteristics. The process preserves phase detail features and avoids loss of effective signal. The signal-to-noise ratio of the enhanced data is significantly improved, providing a reliable foundation for subsequent analysis. Quality enhancement improves the usability and accuracy of the phase data.

[0053] The phase reliability score for each pixel location is calculated based on the enhanced interferometric phase data set. The reliability score is calculated comprehensively using multiple indicators such as phase stability and coherence. The score calculation considers spatiotemporal characteristics, improving the comprehensiveness of the evaluation. Score values ​​are normalized for easier comparison and threshold setting. Through reliability quantification, accurate assessment of pixel quality is achieved.

[0054] Deformation correlation indices for each pixel location are evaluated using phase reliability scores and a deformation dataset. The deformation data provides independent measurements of surface deformation. Correlation analysis employs statistical methods to quantify the degree of association between phase and deformation. Indicator calculations consider the spatiotemporal registration accuracy of the data to reduce matching errors. Through correlation assessment, deformation-sensitive phase signals are identified.

[0055] The selection priority of each pixel location is determined based on deformation correlation indicators and phase statistical characteristics. A weighted algorithm is used for priority calculation to balance the impact of correlation and reliability. Statistical characteristics include parameters such as phase variance and distribution features. Priority ranking supports dynamic adjustment to adapt to different application requirements. Pixel selection strategies are optimized through priority determination.

[0056] A phase filtering model is trained based on phase and deformation data from high-priority pixel locations selected according to filtering priorities. High-priority pixels represent reliable regions that are related to deformation. The model training employs machine learning algorithms to learn the mapping relationship between phase and deformation. The training process considers sample distribution to avoid overfitting. Targeted training improves the model's processing performance.

[0057] A trained phase filtering model is applied to perform phase filtering on the interferometric phase image to be processed. The filtering operation maintains phase continuity and avoids introducing spurious signals. Parameters are adjusted in real time during processing to adapt to local image features. The filtering result significantly reduces noise and improves deformation monitoring accuracy. Adaptive filtering enhances the quality of interferometric phase data. Attached Figure Description

[0058] Figure 1 This is a schematic diagram illustrating the working principle of the ground-based SAR interferometric phase optimization filtering method described in this invention.

[0059] Figure 2 A flowchart for quality enhancement processing of interferometric phase data sets;

[0060] Figure 3 A flowchart for calculating the pixel location phase reliability score;

[0061] Figure 4 The results of the deformation correlation index analysis are shown in the figure.

[0062] Figure 5 This is a graph showing the results of the priority analysis for filtering. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Please see Figure 1This invention provides a ground-based SAR interferometric phase optimization filtering method. The method includes: receiving an interferometric phase data set and a deformation data set acquired by a ground-based SAR system, wherein the interferometric phase data set contains multiple interferometric phase images, each corresponding to a deformation measurement value; performing quality enhancement processing on the interferometric phase data set to generate an enhanced interferometric phase data set; calculating the phase reliability score for each pixel location based on the enhanced interferometric phase data set; evaluating the deformation correlation index for each pixel location using the phase reliability score and the deformation data set; determining the filtering priority for each pixel location based on the deformation correlation index and phase statistical characteristics; selecting high-priority pixel locations according to the filtering priority; training a phase filtering model based on the phase data and deformation data of the high-priority pixel locations; and applying the phase filtering model to perform phase filtering operations on the interferometric phase image to be processed.

[0065] Example 1: See Figure 2 In practical implementation, the process involves several steps to enhance the quality of the interferometric phase data set, generating the enhanced interferometric phase data set. The first step is to adjust the phase values ​​of each interferometric phase image to a range that normalizes the original phase values ​​to a standard circular range, for example, adjusting them to the interval [-π, π]. This process is achieved through modulo operations. After range adjustment, outlier points in the phase data need to be identified. Outlier points are pixels whose phase values ​​differ significantly from those of surrounding pixels. This identification can be achieved using statistical detection based on local windows.

[0066] In some embodiments, a local smoothing algorithm is used to correct identified outliers. This algorithm uses a sliding window of a preset size, such as a 3x3 or 5x5 pixel window, to traverse every outlier in the image. For all pixels within the window, a weighted average of the phase values ​​of all pixels except the central outlier is calculated, and this weighted average replaces the original phase value of the central outlier. The weighting strategy of the local smoothing algorithm can be based on the distance between the pixel and the center of the window, with pixels closer to the center receiving higher weights. After outlier correction, an integrity check is performed on the corrected interferometric phase image. This check includes verifying the presence of pixels with undefined values ​​and ensuring consistent image dimensions. The output is a set of enhanced interferometric phase data that passes the check.

[0067] It is understandable that quality enhancement processing also includes multi-scale decomposition of each interferometric phase image. This multi-scale decomposition employs a pyramid decomposition algorithm, such as the Gaussian pyramid algorithm, to extract phase features at different scales. The specific implementation of the Gaussian pyramid algorithm involves layering each interferometric phase image, constructing a pyramid structure by progressively reducing the image resolution layer by layer. The decomposition process uses the original interferometric phase image as the bottom layer of the pyramid, generating lower-resolution images at higher levels sequentially. Each layer undergoes scale transformation through smoothing filtering and downsampling operations. Smoothing filtering uses Gaussian kernel convolution to suppress noise, while downsampling reduces the image size by a fixed ratio. In practice, for a given interferometric phase image, Gaussian pyramid decomposition generates a series of images with progressively decreasing resolution, each representing phase information at a different scale.

[0068] Optionally, a weighted fusion strategy can be used to fuse multi-scale phase features, fusing the detailed features obtained from each layer of the Laplacian pyramid decomposition with the approximate image of the highest-level Gaussian pyramid. The fusion process assigns different weight coefficients to features at different scales, which can be adjusted according to the need for detailed or overall contour information in a specific application scenario. The fused multi-scale features are then reconstructed into an enhanced interferometric phase image with the same resolution as the original image, ultimately generating an enhanced interferometric phase dataset. The mathematical expression of the fusion operation can be represented as:

[0069]

[0070] in: This represents the enhanced single-frame interferometric phase image. The first representing the Pyramid of Laplace Layer features, An approximate image representing the highest level of the Gaussian pyramid. Is with the first The fusion weights corresponding to the layer Laplacian features These are the fusion weights corresponding to the top layer image of the Gaussian pyramid. It is the total number of levels in the Pyramid of Laplace.

[0071] In some embodiments, the number of multi-scale decomposition layers can be dynamically determined based on the resolution of the original interferometric phase image to ensure the effectiveness of the decomposition. It can be understood that after multi-scale decomposition and fusion processing, the enhanced interferometric phase dataset retains the main phase information of the original image while also enhancing the feature representation at different observation scales.

[0072] Example 2: See Figure 3In practice, for each pixel location in the interferometric phase data set, it is necessary to aggregate the phase values ​​of all interferometric phase images at that specific pixel location to form a phase value sequence. This phase value sequence contains the phase observation values ​​of that pixel at all observation time points.

[0073] In some embodiments, the coefficient of variation (COP) of the phase value sequence is calculated as a measure of phase instability. The COP is the ratio of the standard deviation of the phase value sequence to its absolute mean. The COP calculation provides a standardized measure of dispersion, allowing for comparison of the stability of phase value sequences at different average levels. A larger COP indicates greater volatility in the phase value sequence, i.e., higher phase instability.

[0074] It's understandable that performing an inverse proportional transformation on the phase instability metric is key to obtaining the phase reliability score; this transformation directly takes the reciprocal of the phase instability metric. Assuming the phase instability metric is a non-negative real number, its reciprocal will exhibit an inverse relationship: the smaller the phase instability metric value, the larger its reciprocal, indicating higher phase reliability. In practice, the reciprocal value may have a large range, requiring normalization.

[0075] Optionally, the normalization process employs a min-max normalization method, linearly adjusting the reciprocal values ​​to the range of zero to one. Normalization requires iterating through the reciprocals calculated for all pixel positions, finding the global maximum and global minimum, and then applying the normalization formula to map the reciprocal value for each pixel position to the [0,1] interval. The normalization formula can be expressed as:

[0076]

[0077] in: This represents the normalized phase reliability score. Represents a measure of phase instability. The reciprocal of the measure of phase instability. This represents the minimum value among the reciprocals of all pixel positions. It represents the maximum value among the reciprocals of all pixel positions.

[0078] In some embodiments, the aggregation operation of the phase value sequence needs to consider the circumferential characteristics of the phase values, that is, it needs to be ensured that the phase value sequence has undergone appropriate unwrapping or circumferential statistical processing before calculating the statistics. It can be understood that after the normalized value is used as the phase reliability score, each pixel position will obtain a score between 0 and 1. The closer the score is to 1, the more stable and reliable the phase observation value of that pixel position is in the time series.

[0079] Example 3: In specific implementation, evaluating the deformation correlation index first requires extracting the phase value sequence and the corresponding deformation value sequence for each pixel position. The phase value sequence is formed by aggregating the phase values ​​of all interferometric phase images in the enhanced interferometric phase data set at that pixel position. The deformation value sequence is the deformation measurement value sequence extracted from the deformation data set corresponding to each interferometric phase image.

[0080] In some embodiments, the Pearson correlation coefficient between the phase value sequence and the deformation value sequence is calculated as the initial deformation correlation. The Pearson correlation coefficient is calculated using a standard formula and measures the degree of linear correlation between the two sequences, with a value ranging from -1 to +1. The calculation of the Pearson correlation coefficient involves the ratio of the product of the covariance of the phase value sequence and the standard deviation of their respective sequences. The closer the result is to +1 or -1, the stronger the linear correlation.

[0081] It is understandable that weighting the initial deformation correlation based on the phase reliability score is key to obtaining the deformation correlation index. The weighting adjustment process uses the phase reliability score as an adjustment factor applied to the initial deformation correlation. Pixels with high phase reliability scores have their initial deformation correlation assigned a higher confidence level, while pixels with low scores are suppressed. Optionally, the mathematical expression of the weighted adjustment can be represented as:

[0082]

[0083] in: Represents the deformation correlation index. Represents the correlation of initial deformation. This represents the phase reliability score.

[0084] In some embodiments, when extracting the phase value sequence and deformation value sequence, it is necessary to ensure that the sequence lengths are consistent and that the data points correspond one-to-one to avoid calculation deviations due to missing values ​​or misalignments. It can be understood that the deformation correlation index is ultimately a real value, the magnitude of which comprehensively reflects the correlation between phase and deformation, as well as the reliability of the phase data itself. This index will be used for subsequent priority calculations. After weighted adjustment, the numerical range of the deformation correlation index may be compressed, but the sign and relative magnitude of the initial correlation are preserved.

[0085] See Figure 4This figure illustrates the spatial distribution of deformation correlation indices. A heatmap is used to visually display the deformation correlation index value for each pixel location, with colors ranging from blue to red indicating a change from negative to positive correlation values. Differences in deformation correlation across different regions are clearly observed. High correlation areas indicate a strong consistency between phase changes and deformation measurements at that location; these areas typically correspond to locations with significant surface deformation or good phase quality. The spatial distribution pattern shown reflects the influence of topographic features and deformation mechanisms on the phase-deformation relationship, providing important information for subsequent pixel selection. Deformation correlation indices comprehensively consider the linear correlation between the phase sequence and the deformation sequence, as well as the reliability of the phase data itself, and are one of the key parameters for evaluating pixel quality.

[0086] Example 4: In specific implementation, determining the screening priority first requires analyzing the probability distribution of the phase value sequence at each pixel location, generating a phase probability distribution map. Analyzing the probability distribution of the phase value sequence involves dividing the numerical range of the phase value sequence into multiple equally wide intervals, counting the occurrence frequency of the phase value within each interval, and generating a histogram-style phase probability distribution map based on the intervals and the occurrence frequency. Similarly, the probability distribution of the deformation value sequence is analyzed, generating a deformation probability distribution map using the same interval division method. The numerical range of the phase value sequence is typically from -π to π, and the number of intervals can be set according to the data characteristics, for example, dividing it into 10 intervals. Table 1 below shows an example of a phase value sequence probability distribution representation:

[0087] Table 1: Probability Distribution of Phase Value Sequence

[0088]

[0089] In some embodiments, the number of interval divisions is determined by a fixed value or an adaptive method, such as selecting the number of intervals based on the length of the phase value sequence. It is understood that a probability distribution plot in histogram form provides a visual representation of the distribution shape. In a specific implementation, the Jaccard similarity between the phase probability distribution plot and the deformation probability distribution plot is calculated as a distribution consistency score; the Jaccard similarity measures the degree of overlap between the two distribution plots. The formula for calculating the Jaccard similarity is defined as:

[0090]

[0091] in: Represents the distribution consistency score. This represents the number of occurrences of the k-th interval in the phase probability distribution diagram. This represents the number of occurrences of the k-th interval in the deformation probability distribution plot. This represents the total number of intervals. Jaccard similarity values ​​range from 0 to 1. In practice, the distribution consistency score is multiplied by the deformation correlation index to obtain the screening priority: Screening priority = Distribution consistency score × Deformation correlation index.

[0092] In some embodiments, selecting high-priority pixel locations includes setting a priority threshold and marking pixel locations with a filtering priority greater than the threshold as candidate pixel locations. The priority threshold can be set based on the global distribution of filtering priority values, for example, selecting the top 30% quantile of filtering priority values ​​as the threshold. Spatial clustering analysis is performed on the candidate pixel locations to remove outliers and retain pixel locations within connected regions as high-priority pixel locations. The spatial clustering analysis uses a density-based clustering algorithm, such as the DBSCAN algorithm, which considers the two-dimensional spatial coordinates of the pixel locations.

[0093] Optionally, the priority threshold can be set to the average of the filtering priority values ​​plus one standard deviation to dynamically adjust the threshold level. In spatial clustering analysis, a connected region is defined as a group of pixels formed under eight-connected neighbor relationships. Optionally, before generating the probability distribution map, the phase value sequence and deformation value sequence can be preprocessed, such as normalized, to eliminate the influence of dimensions.

[0094] See Figure 5 The figure illustrates the spatial distribution of pixel filtering priorities. The color gradient represents the filtering priority value at different pixel locations; higher brightness indicates a higher priority in subsequent processing. The filtering priority calculation combines deformation correlation and distribution consistency scores, with the latter obtained by comparing the Jaccard similarity between the phase probability distribution map and the deformation probability distribution map. High-priority regions are marked by red outlines; these regions not only exhibit good phase-deformation correlation but also show high consistency between their phase and deformation statistical distributions. These high-priority pixels will be prioritized during the training of the phase filtering model, contributing to improved filtering performance and deformation detection accuracy. The spatial distribution pattern shown in the chart provides important spatial reference information for optimizing the filtering algorithm.

[0095] Example 5: In practical implementation, training the phase filter model first requires calculating training weights based on the filtering priority of each high-priority pixel position. The training weights are positively correlated with the filtering priority; the higher the filtering priority, the greater the influence of the pixel position during model training. The calculation of training weights can employ linear scaling or sorting-based allocation methods, such as directly using the normalized filtering priority value as the training weight.

[0096] In some embodiments, phase data and deformation data of high-priority pixel locations are combined as training samples. The phase data comes from the phase value sequence of the corresponding pixel location in the enhanced interferometric phase data set, and the deformation data comes from the deformation value sequence of the corresponding deformation data set. Training weights are assigned to each training sample, with high-weight training samples receiving more attention during model training. A neural network model is constructed as the phase filtering model. The neural network model can adopt a fully connected feedforward neural network structure, with the number of input layer nodes corresponding to the length of the phase value sequence and the number of output layer nodes corresponding to the length of the deformation value sequence.

[0097] It is understandable that using a weighted loss function is a crucial step in model training, ensuring that the loss contribution of each training sample is modulated by its training weights. The specific form of the weighted loss function can be a weighted mean squared error function. For a training set containing N training samples, the weighted loss function L is defined as follows:

[0098]

[0099] in: This represents the value of the weighted loss function. The training weights represent the training weights of the i-th training sample. This represents the true deformation value of the i-th training sample. This represents the predicted deformation value of the neural network model for the i-th training sample. This represents the total number of training samples.

[0100] In practice, the weighted loss function is reduced through iterative optimization until the model parameters converge, resulting in the trained phase filter model. Iterative optimization uses gradient descent and its variants, such as the Adam optimizer, updating the weights and biases of the neural network model in each iteration. The criteria for model parameter convergence can be that the weighted loss function value no longer decreases significantly over multiple consecutive iterations, or that the preset maximum number of iterations has been reached.

[0101] Optionally, when applying the phase filtering model to perform phase filtering on the interferometric phase image to be processed, the phase data of the interferometric phase image to be processed is input into the trained phase filtering model. The phase data is the phase value at each pixel position in the interferometric phase image to be processed. The phase filtering model outputs the predicted deformation value at each pixel position based on the learned mapping relationship. The phase data is adjusted according to the predicted deformation value to generate the filtered interferometric phase image. The adjustment process can be that the predicted deformation value is fed back to the phase unwrapping or phase compensation algorithm.

[0102] In some embodiments, the number of hidden layers and the number of nodes per layer in the neural network model are configured based on the number and complexity of training samples to prevent overfitting or underfitting. It can be understood that the phase filtering model learns the complex nonlinear relationship between phase and deformation by utilizing reliable data from high-priority pixel locations, thereby enabling effective filtering of novel interferometric phase images.

[0103] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A ground-based SAR interferometric phase optimization filtering method, characterized in that, The method includes: The system receives interferometric phase data sets and deformation data sets acquired by a ground-based SAR system. The interferometric phase data set contains multiple interferometric phase images, and each interferometric phase image corresponds to a deformation measurement value. The interference phase data set is subjected to quality enhancement processing to generate an enhanced interference phase data set; Based on the enhanced interferometric phase data set, the phase reliability score at each pixel location is calculated; The deformation correlation index at each pixel location is evaluated using phase reliability scores and deformation datasets. Based on deformation correlation index and phase statistical characteristics, the screening priority of each pixel position is determined; Based on the filtering priority, select high-priority pixel positions; train a phase filtering model based on the phase data and deformation data of the high-priority pixel positions; A phase filtering model is applied to perform phase filtering on the interferometric phase image to be processed; The process of determining the filtering priority for each pixel position includes: Analyze the probability distribution of the phase value sequence at each pixel location to generate a phase probability distribution map; analyze the probability distribution of the deformation value sequence to generate a deformation probability distribution map. Calculate the Jaccard similarity between the phase probability distribution map and the deformation probability distribution map, and use it as the distribution consistency score; Multiply the distribution consistency score by the deformation correlation index to obtain the screening priority; The probability distribution of the phase value sequence at each pixel location is analyzed, including: The numerical range of the phase value sequence is divided into multiple equally wide intervals; Count the number of times the phase value appears in each interval of the phase value sequence; Generate a phase probability distribution map in the form of a histogram based on the interval and the frequency of occurrence.

2. The ground-based SAR interferometric phase optimization filtering method according to claim 1, characterized in that, The quality enhancement process for the interferometric phase data set includes: The phase values ​​of each interferometric phase image are adjusted to normalize them to the standard circumference range. Outlier points in phase data are identified and corrected using a local smoothing algorithm; Perform an integrity check on the corrected interferometric phase image and output the enhanced interferometric phase data set.

3. The ground-based SAR interferometric phase optimization filtering method according to claim 1, characterized in that, The calculation of the phase reliability score for each pixel location includes: For each pixel location, the phase values ​​of all interferometric phase images at that location are aggregated to form a phase value sequence; Calculate the coefficient of variation of the phase value sequence as a measure of phase instability; The phase instability measure is inversely proportionally transformed to obtain the phase reliability score; The inverse proportional transformation of the phase instability measure includes: Take the reciprocal of the phase instability measure; The reciprocal is normalized to adjust its value range to between zero and one. The normalized value is used as the phase reliability score.

4. The ground-based SAR interferometric phase optimization filtering method according to claim 1, characterized in that, The deformation correlation metrics for evaluating each pixel location include: Extract the phase value sequence and the corresponding deformation value sequence for each pixel location; Calculate the Pearson correlation coefficient between the phase value sequence and the deformation value sequence as the initial deformation correlation; The initial deformation correlation is weighted and adjusted based on the phase reliability score to obtain the deformation correlation index.

5. The ground-based SAR interferometric phase optimization filtering method according to claim 1, characterized in that, The selection of high-priority pixel positions includes: Set a priority threshold and mark the pixel positions with a priority greater than the priority threshold as candidate pixel positions; perform spatial clustering analysis on the candidate pixel positions to remove isolated points and retain the pixel positions within the connected regions as high-priority pixel positions.

6. The ground-based SAR interferometric phase optimization filtering method according to claim 1, characterized in that, The trained phase filtering model includes: Calculate training weights based on the filtering priority of each high-priority pixel position; The phase data and deformation data of the high-priority pixel locations are combined into training samples, and the training weights are assigned to each training sample. Build a neural network model and use a weighted loss function during training, where the loss of each training sample is multiplied by its training weight; The value of the weighted loss function is reduced through iterative optimization until the model parameters converge, thus obtaining the trained phase filter model.

7. The ground-based SAR interferometric phase optimization filtering method according to claim 1, characterized in that, The phase filtering operation performed on the interferometric phase image to be processed by the application phase filtering model includes: Input the phase data of the interferometric phase image to be processed into the phase filtering model; the phase filtering model outputs the predicted deformation value. The phase data is adjusted based on the predicted deformation values ​​to generate a filtered interferometric phase image.

8. The ground-based SAR interferometric phase optimization filtering method according to claim 1, characterized in that, The generation of the enhanced interferometric phase data set also includes: Each interferometric phase image is decomposed into multiple scales to extract phase features at different scales; By fusing multi-scale phase features, an enhanced interferometric phase data set is generated.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the ground-based SAR interferometric phase optimization filtering method as described in any one of claims 1 to 8.

Images