High-quality pixel selection method for ground-based SAR based on amplitude and phase difference analysis

Abstract

The present invention discloses a method for selecting high-quality pixels of ground-based SAR based on amplitude-phase difference analysis. The present invention can accurately select single-point and regionally stable pixels of amplitude and phase from ground-based SAR images, which is beneficial to improving the accuracy of differential interferometry deformation measurement and providing accurate one-dimensional deformation data for the prediction and early warning of landslide disasters. First, the amplitude deviation method is used to classify the pixels; secondly, based on cluster phase analysis, the pixels with highly stable amplitude and phase are selected; then, based on time series phase analysis, the pixels with high amplitude stability and high phase stability are selected; finally, based on regional amplitude phase analysis, the pixels with regionally stable amplitude and phase are selected.

Description

Technical Field

[0001] The invention belongs to the technical field of synthetic aperture radar, and in particular relates to a ground-based SAR high-quality pixel point selection method based on amplitude and phase difference analysis. Background Art

[0002] As a high-precision deformation measurement instrument, ground-based SAR (Synthetic Aperture Radar) has been widely used in the field of deformation monitoring. Ground-based SAR is usually based on differential interferometry technology. It performs differential interferometry processing on two SAR images acquired at the same location but at different times, and realizes deformation measurement based on phase information. Due to the influence of non-ideal factors such as system thermal noise and atmospheric disturbance, the phase quality of some pixels in the interferometric phase image is very low, and it is necessary to select pixels with high phase quality for deformation analysis.

[0003] The amplitude deviation method has been widely used in the field of ground-based SAR. By calculating the ratio of the amplitude standard deviation to the amplitude mean of each pixel in multiple time-series SAR images, and then setting a reasonable threshold, the selection of PS (Permanent Scatterer) can be achieved. This method can effectively select high-coherence targets such as exposed rocks and concrete buildings in scenes such as mountains, mines, and towns. However, for some less coherent scenes, such as mountain slopes with vegetation, the number of PS selected by the amplitude deviation method is very small due to the lack of high-coherence targets such as exposed rocks in this type of scene. It is impossible to accurately select single-point and regionally stable pixels from ground-based SAR images, which is not conducive to deformation analysis. Summary of the invention

[0004] In view of this, the present invention provides a ground-based SAR high-quality pixel selection method based on amplitude-phase difference analysis, which can accurately select amplitude-phase single-point and regionally stable pixels from ground-based SAR images.

[0005] To achieve the above object, the technical solution of the present invention is as follows:

[0006] A method for selecting high-quality pixels of ground-based SAR based on amplitude-phase difference analysis of the present invention comprises the following steps:

[0007] The amplitude deviation values ​​of multiple ground-based SAR images are calculated to set an amplitude deviation threshold; pixels with high and medium stability in amplitude are selected from the multiple ground-based SAR images according to the threshold; phase screening is performed on pixels with high stability in amplitude, and pixels with high stability in amplitude and phase are extracted based on cluster phase analysis; pixels with high stability in amplitude and phase are used as a benchmark, and pixels with medium stability in amplitude and high stability in phase are extracted based on time series phase analysis; regional amplitude and phase analysis is performed on pixels with high or medium stability in amplitude and low stability in phase, and pixels with regional stability in amplitude and phase are extracted.

[0008] Among them, the amplitude deviation thresholds are 0.2 and 0.4, pixels with amplitude deviation values ​​lower than 0.2 are regarded as high amplitude stable pixels, pixels between 0.2 and 0.4 are regarded as medium amplitude stable pixels, and pixels higher than 0.4 are regarded as low amplitude stable pixels.

[0009] Among them, in the cluster phase analysis, after clustering all the pixels with high amplitude stability, the phase sequence of each cluster center point is first calculated, and then the coherence coefficient of each pixel in the cluster is calculated; the coherence coefficient threshold is set, and the pixels with coherence coefficients higher than the threshold are regarded as pixels with high amplitude stability, and those below the threshold are regarded as pixels with high amplitude stability and low phase stability; after the coherence coefficient is screened, the pixels with high amplitude stability are re-clustered, and the average phase sequence of each cluster center is calculated; then a Delaunay triangulation network is established to connect the cluster centers.

[0010] The K-means clustering algorithm is used to divide the pixels into K clusters. The average number of pixels with high amplitude stability in each cluster is set to 100, and the value of K depends on the total number of pixels with high amplitude stability and their distribution in the image. The cluster center point C k The complex phase sequence Expressed as

[0011]

[0012] Among them, r n Represents the nth pixel in the cluster and the cluster center point C k The distance between () represents a complex function, Ang() represents the phase of a complex number; Characterizes the N in the kth cluster k The inverse distance weighted average phase of pixels with high stability amplitude; the coherence coefficient of the nth pixel in this cluster is

[0013]

[0014] Among them, if a stable pixel point in amplitude is in a certain triangle, calculate the distance between it and the three vertices of the triangle, and then calculate the average phase sequence and the coherence coefficient of the stable pixel point in amplitude; if a stable pixel point in amplitude is outside all triangles, select the cluster center closest to it, and calculate its coherence coefficient in the same way; set the coherence coefficient threshold, and the pixel points with coherence coefficients higher than the threshold are stable pixels in amplitude and high phase stability, and those below the threshold are stable pixels in amplitude and low phase stability.

[0015] Among them, all homogeneous pixel sets are obtained from all pixels with high or medium stability in amplitude and low stability in phase; for each homogeneous pixel set, circular periodic mean filtering is performed to obtain the filtered phase sequence of the center point of the set Expressed as

[0016]

[0017] in, represents the mean operation, Represents the phase of the i-th pixel in the homogeneous pixel set; Represents the average vector of the unit phase vector of a set of homogeneous pixels, expressed as

[0018]

[0019] The coherence coefficients of all homogeneous pixel sets are calculated, and a homogeneous pixel set with a coherence coefficient higher than a set threshold is a regionally stable pixel with amplitude and phase.

[0020] The homogeneous pixel set is obtained as follows:

[0021] Based on the KS test, we determine whether the amplitudes of point P and point Q follow the same distribution. The discriminant function of the KS test is expressed as

[0022]

[0023] Among them, D N Represents the maximum vertical distance between two probability distribution functions; the amplitude sequence d of point P and point Q P d Q The probability distribution functions of P ) and F(d Q );

[0024] When D N When the value of is greater than the set threshold, the amplitude distribution of point P and point Q is different; when D N When the value of is less than the set threshold, the amplitude distributions of point P and point Q are consistent;

[0025] Based on the KS test, all the pixel points in the rectangular window centered on P that have the same amplitude distribution as point P are judged. If the number of these pixel points is not less than 10, they together with point P constitute a homogeneous pixel point set.

[0026] Beneficial Effects

[0027] The present invention first adopts the amplitude deviation method to classify the pixel points; secondly, based on the cluster phase analysis, the pixel points with highly stable amplitude and phase are selected; then based on the time series phase analysis, the pixel points with high amplitude stability and high phase stability are selected; finally, based on the regional amplitude phase analysis, the pixel points with regional amplitude and phase stability are selected. The present invention can accurately select the single-point and regionally stable pixel points with amplitude and phase from the ground-based SAR image, which is conducive to improving the accuracy of differential interferometry deformation measurement and providing accurate one-dimensional deformation data for the prediction and early warning of landslide disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of the high-quality pixel selection method of the present invention.

[0029] Figure 2 This is a schematic diagram of clustering division of the present invention.

[0030] Figure 3 It is a schematic diagram of the Delaunay triangulation of the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0032] The flow chart of the high-quality pixel selection method of the present invention is as follows: Figure 1 As shown, the present invention specifically includes the following steps:

[0033] Step 1: Amplitude classification:

[0034] The amplitude deviation value D of a pixel A Expressed as where σ A and m A They represent the standard deviation and mean of the amplitude sequence of the pixel in the M images. To ensure the accuracy of the amplitude deviation estimation, the number M should be greater than 20.

[0035] The amplitude deviation thresholds of 0.2 and 0.4 are set, and all pixels in the image are divided into three categories: pixels with amplitude deviation values ​​lower than 0.2 are regarded as high-amplitude stable pixels, pixels between 0.2 and 0.4 are regarded as medium-amplitude stable pixels, and pixels higher than 0.4 are regarded as low-amplitude stable pixels.

[0036] Step 2: Cluster phase analysis:

[0037] The pixels with high amplitude stability are distributed discretely and non-uniformly in the whole image. The K-means clustering algorithm is used to divide these pixels into K clusters. The average number of pixels with high amplitude stability in each cluster is set to 100. The value of K depends on the total number of pixels with high amplitude stability and their distribution in the image. Assume that all pixels with high amplitude stability constitute a sample set x. Now divide x into K clusters (C 1 ,C 2 ,…,C K ). The sum of squared errors E between clusters can be expressed as

[0038]

[0039] Among them, || || represents the second-order norm, that is, the modulus of the vector. μ i is cluster C i The mean vector of can be expressed as

[0040]

[0041] Among them, | | represents the first-order norm, |C i |That is, the number of points in the cluster. Figure 2 The following is a schematic diagram of clustering division. The smaller circular dots in the figure represent pixels with high stability in amplitude, and the three larger square dots represent the center points of the three clusters.

[0042] Assume that the kth cluster includes N k The complex phase sequence of the nth pixel is expressed as n=1,2,…,N k , m=1,2,…,M. Cluster center C k The complex phase sequence Expressed as

[0043]

[0044] Among them, r n Represents the nth pixel in the cluster and the cluster center point C k The distance between () represents a complex function, and Ang() represents the phase of a complex number. Characterizes the N in the kth cluster k The inverse distance weighted average phase of the nth pixel in the cluster with high amplitude stability can be expressed as

[0045]

[0046] After clustering all the pixels with high amplitude stability, the phase sequence of each cluster center point is first calculated based on formula (3), and then the coherence coefficient of each pixel in the cluster is calculated based on formula (4). The coherence coefficient threshold is set to 0.9, and the pixels with coherence coefficients higher than the threshold are regarded as pixels with high amplitude and phase stability, and those with coherence coefficients lower than the threshold are regarded as pixels with high amplitude stability and low phase stability.

[0047] After the coherence coefficient is screened, the pixels with high amplitude and phase stability are re-clustered, and the average phase sequence of each cluster center is calculated based on formula (3). Then a Delaunay triangulation network is established to connect these cluster centers. The Delaunay triangulation network is a collection of a series of connected but non-overlapping triangles, and the circumscribed circle of any triangle does not contain any other points. Figure 3 The figure shows a schematic diagram of the Delaunay triangulation network, where the circular points represent the cluster centers.

[0048] Step 3: Timing phase analysis:

[0049] Now we classify the stable pixels in the amplitude based on the time series phase analysis. If a stable pixel in the amplitude is in a certain triangle, calculate the distance between it and the three vertices of the triangle, and then use the inverse distance weighted formula shown in formula (3) to calculate an average phase sequence, and calculate the coherence coefficient of the stable pixel in the amplitude based on formula (4). If a stable pixel in the amplitude is outside all triangles, select the cluster center closest to it, and calculate its coherence coefficient based on formula (4).

[0050] After the above processing, the coherence coefficients of all stable pixels in amplitude are obtained, and a coherence coefficient threshold of 0.9 is set. Pixels with coherence coefficients higher than the threshold are stable pixels in amplitude and highly stable pixels in phase, and those with coherence coefficients lower than the threshold are stable pixels in amplitude and low stable pixels in phase.

[0051] Step 4: Regional amplitude and phase analysis:

[0052] After processing from step 1 to step 3, all pixels with high or medium stability in amplitude and low stability in phase are screened out. For these pixels, take any pixel P as an example. With point P as the center, build a 5×5 rectangular window, and then for other pixels of the same type in the rectangular window, take point Q as an example, assume that the amplitude sequence d of point P and point Q P,d Q The probability distribution functions of P ) and F(d Q ). Now based on the KS test, we determine whether the amplitudes of point P and point Q follow the same distribution. The discriminant function of the KS test is expressed as

[0053]

[0054] Among them, D N Represents the maximum vertical distance between two probability distribution functions. N Set a reasonable threshold, the typical value range is 1% to 5%. N When the value of is greater than the threshold, the amplitude distribution of point P and point Q is different; when D N When the value of is less than the threshold, the amplitude distributions of point P and point Q are consistent.

[0055] Based on the above KS test, all the pixels in the rectangular window centered on P that have the same amplitude distribution as point P are judged. If the number of these pixels is not less than 10, they together with point P form a set of statistically homogeneous pixels (SHP). Using the above method, all SHP sets are obtained based on pixels with high or medium stability in amplitude and low stability in phase.

[0056] For each SHP set, circular periodic mean filtering is performed to obtain the filtered phase sequence of the center point of the set. Expressed as

[0057]

[0058] in, represents the mean operation, Represents the phase of the i-th pixel in the SHP set. The average vector of the unit phase vector of a SHP set is expressed as

[0059]

[0060] In step 2, based on all the highly stable pixels, the average phase sequence of each cluster center is obtained, and a triangulated network is constructed to connect these cluster centers. According to the temporal phase analysis in step 3, the coherence coefficient of all SHP sets is calculated, and a coherence coefficient threshold of 0.9 is set. The SHP set with a coherence coefficient higher than the threshold is a regionally stable pixel with amplitude and phase.

[0061] Based on the analysis of amplitude-phase differences, the pixels with high or medium amplitude stability and high phase stability are selected as single-point amplitude-phase stable pixels, and the combined amplitude-phase regional stable pixels are all high-quality pixels.

[0062] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for selecting high-quality pixels for ground-based SAR based on amplitude and phase difference analysis. It is characterized in that The steps include: By calculating the amplitude deviation values ​​of multiple ground-based SAR images, an amplitude deviation threshold is set; according to the threshold, pixels with high and medium stability in amplitude are selected from the multiple ground-based SAR images; phase screening is performed on the pixels with high stability in amplitude, and based on cluster phase analysis, pixels with high stability in amplitude and phase are extracted; based on the pixels with high stability in amplitude and phase, pixels with medium stability in amplitude and high stability in phase are extracted based on time series phase analysis; regional amplitude and phase analysis is performed on pixels with high or medium stability in amplitude and low stability in phase, and pixels with regional stability in amplitude and phase are extracted; The amplitude deviation thresholds are 0.2 and 0.4, pixels with amplitude deviation values ​​lower than 0.2 are regarded as high-amplitude stable pixels, pixels between 0.2 and 0.4 are regarded as medium-amplitude stable pixels, and pixels higher than 0.4 are regarded as low-amplitude stable pixels; In the cluster phase analysis, after clustering all the pixels with high amplitude stability, the phase sequence of each cluster center point is first calculated, and then the coherence coefficient of each pixel in the cluster is calculated; The coherence coefficient threshold is set. Pixels with coherence coefficients higher than the threshold are regarded as pixels with high amplitude and phase stability, and those with coherence coefficients lower than the threshold are regarded as pixels with high amplitude stability and low phase stability. After the coherence coefficient is screened, the pixels with high amplitude and phase stability are re-clustered and the average phase sequence of each cluster center is calculated. Then, a Delaunay triangulation network is established to connect the cluster centers. The K-means clustering algorithm is used to divide the pixels into K clusters. The average number of pixels with high amplitude stability in each cluster is set to 100, and the value of K depends on the total number of pixels with high amplitude stability and their distribution in the image. The cluster center point C k The complex phase sequence Expressed as Among them, r n Represents the nth pixel in the cluster and the cluster center point C k The distance between them, e() represents a complex function, and Ang() represents the phase of a complex number; Characterizes the N in the kth cluster k The inverse distance weighted average phase of pixels with high stability amplitude; the coherence coefficient of the nth pixel in this cluster is If a stable pixel point in amplitude is in a certain triangle, calculate the distance between it and the three vertices of the triangle, and then calculate the average phase sequence and the coherence coefficient of the stable pixel point in amplitude; if a stable pixel point in amplitude is outside all triangles, select the cluster center closest to it and calculate its coherence coefficient in the same way; set the coherence coefficient threshold, and the pixel points with coherence coefficients higher than the threshold are stable pixels in amplitude and high phase stability, and those with coherence coefficients lower than the threshold are stable pixels in amplitude and low phase stability; From all the pixels with high or medium stability in amplitude and low stability in phase, all the homogeneous pixel sets are obtained; for each homogeneous pixel set, circular periodic mean filtering is performed to obtain the filtered phase sequence of the center point of the set. Expressed as in, represents the mean operation, Represents the phase of the i-th pixel in the homogeneous pixel set; Represents the average vector of the unit phase vector of a set of homogeneous pixels, expressed as The coherence coefficients of all homogeneous pixel sets are calculated, and a homogeneous pixel set with a coherence coefficient higher than a set threshold is a regionally stable pixel with amplitude and phase.

2. The method according to claim 1, It is characterized in that The homogeneous pixel set is obtained as follows: Based on the KS test, we determine whether the amplitudes of point P and point Q follow the same distribution. The discriminant function of the KS test is expressed as Among them, D N Represents the maximum vertical distance between two probability distribution functions; the amplitude sequence d of point P and point Q P ,d Q The probability distribution functions of P ) and F(d Q ); When D N When the value of is greater than the set threshold, the amplitude distribution of point P and point Q is different; when D N When the value of is less than the set threshold, the amplitude distributions of point P and point Q are consistent; Based on the KS test, all the pixel points in the rectangular window centered on P that have the same amplitude distribution as point P are judged. If the number of these pixel points is not less than 10, they together with point P constitute a homogeneous pixel point set.

Images