A distributed target recognition method based on polarization parameter constraint

Abstract

The embodiment of the application discloses a distributed target identification method based on polarization parameter constraint, comprising the following steps: acquiring N time-series SAR images, obtaining a time-series SAR image stack of a region of interest through image preprocessing; based on the time-series SAR image stack, calculating the amplitude, average amplitude and coherence of each SAR image; based on the time-series SAR image stack, calculating the polarization entropy H and average scattering angle alpha of each pixel point of each SAR image; setting a homogeneous pixel detection window to select homogeneous pixels, and optimizing the selection result of the homogeneous pixels; setting a correlation threshold based on the average amplitude and coherence, and combining a connectivity threshold to complete the selection of DS homogeneous points, so that the SAR data information can be fully utilized, and the use value of the SAR data is greatly improved. Moreover, the DS selection process and quality are further optimized without the aid of other data.

Description

Technical Field

[0001] This invention relates to the field of target data recognition technology, specifically to a distributed target recognition method based on polarization parameter constraints. Background Technology

[0002] The information disclosed in this background section is intended only to enhance understanding of the overall background of the invention and is not necessarily to be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art.

[0003] PS-InSAR technology, as a commonly used geological deformation monitoring technology, can provide stable and reliable deformation monitoring results, but its density is very low, and it cannot provide detailed deformation field information when detecting deformation fields.

[0004] The DS-InSAR method can acquire high-density deformation fields in a study area. Distributed scatterers (DS) refer to scatterers with relatively uniform ground features that remain stable over a certain period of time. In imagery, a DS corresponds to multiple pixels with the same distribution characteristics; these pixels belong to the same ground feature. Approximately 90% of the pixels in an image are DS points, while PS points account for a very small proportion. DS can provide richer deformation details in temporal analysis. DS are widely distributed spatially, including rooftops, bare ground, roads, and idle farmland. In addition to the above-mentioned ground feature types, common DS also include barren mountains, deserts, low shrublands, and concrete surfaces.

[0005] When searching for DS points, it is necessary to find points with the same phase statistical characteristics within a given distance and perform cluster analysis on these points. The reliable acquisition of the DS-InSAR deformation field is highly dependent on the accurate selection of DS points. Conventional DS point selection methods often rely solely on statistical characteristics, neglecting the natural attributes or land use status of the DS points, thus reducing the accuracy of DS selection. Using land use classification as a constraint for DS can appropriately improve the accuracy of DS selection because the physical characteristics of DS points are highly consistent; the same DS point corresponds to the same land cover, therefore its surface roughness, complex permittivity, backscattering coefficient, and other physical characteristics are basically the same. However, the same land cover may not have the same scattering characteristics. Relying solely on land cover classification is not as effective as relying on polarization entropy, and using land use type data increases the amount of data, which in turn affects the accuracy of DS selection. Summary of the Invention

[0006] To address this issue, this invention provides a distributed target recognition method based on polarization parameter constraints, thereby solving the problem of inaccurate distributed target point recognition in the prior art due to insufficient accuracy and correctness of DS point selection.

[0007] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0008] In an embodiment of the present invention, a distributed target recognition method based on polarization parameter constraints is provided, the method comprising the following steps:

[0009] S1: Acquire N temporal SAR images and obtain the temporal SAR image stack of the region of interest through image preprocessing;

[0010] S2: Based on the time-series SAR image stack, calculate the amplitude, average amplitude, and coherence of each SAR image;

[0011] S3: Based on the temporal SAR image stack, calculate the polarization entropy of each pixel in each SAR image. H and average scattering angle α ;

[0012] S4: Set up a homogeneous pixel detection window to select homogeneous pixels and optimize the selection results.

[0013] S5: Based on the average amplitude and coherence, relevant thresholds are set, and the selection of homogeneous points in DS is completed by combining the connectivity threshold.

[0014] Furthermore, in step S1, RSLC images of different polarization types in the region of interest are obtained by performing multi-view and filtering preprocessing on the cropped image.

[0015] Further, in step S2, based on the time-series SAR image stack, the amplitude, average amplitude, and coherence of each SAR image are calculated, which includes the following steps:

[0016] S2.1: Aspect Ratio of Each Image Amp The following formula can be used to calculate:

[0017]

[0018] Among them, abs (RSLC) i To find the absolute value, i = 1, 2, 3, ..., n, where n represents the number of images;

[0019] S2.2: Average amplitude Amp ave and coherence Coh The following formula can be used to calculate:

[0020]

[0021] Among them, E[yy *t ] represents the expectation, y=[y1,y2,...,y n ]T This represents the normalized result of multiple observations of homogeneous points of a distributed target on n SAR images, i.e. .

[0022] Furthermore, in step S3, the specific application of optimizing the selection of homogeneous pixels is as follows:

[0023] The polarization entropy of each image is calculated using the following formula. H and average scattering angle α ;

[0024]

[0025]

[0026] Where S is the scattering matrix, k is the target scattering vector, T denotes matrix transpose, and <·> denotes time averaging or spatial averaging, k i It is an eigenvector. λ represents the complex conjugate transpose. i It is an eigenvalue, P i From the eigenvalue (λ) i The pseudo-probabilities obtained are, where .

[0027] Furthermore, the selection of homogeneous pixels in step S4 is as follows:

[0028] S4.1: Set the size of the homogeneous pixel detection window to m*n, i.e., m rows and n columns, all of which are odd numbers;

[0029] S4.2: Using amplitude information as the data source, set the polarization entropy of each pixel within the window. H and average scattering angle α The constraint conditions are as follows: when the constraint conditions of the center pixel in the window are available and are the same as those of other pixels in the window, then all pixels in the window are subjected to AD test with the center point of the window to determine homogeneous pixels; otherwise, no test is performed.

[0030] Calculate polarization entropy H With average scattering angle α The constraint conditions are formulated as follows:

[0031]

[0032] Among them, Hseg i polarization entropy H Segmentation threshold, αseg i The average scattering angle αThe segmentation threshold, F, is a function of two thresholds. This function can represent all scattering mechanisms and will divide the study area into different scattering mechanism categories based on the segmentation threshold. Each different scattering mechanism category is used as the same constraint to divide the entire study area. Then, a suitable window is selected to test for homogeneous pixels. When the constraint condition of the central pixel in the detection window is available and it is the same as other pixels in the window, then all pixels in the window are tested against the window center point one by one to determine homogeneous pixels; otherwise, no test is performed.

[0033] S4.3: Perform connectivity detection on the central pixel and its homogeneous pixels within the homogeneous pixel detection window, and count the number of homogeneous pixels within the homogeneous pixel detection window.

[0034] Furthermore, in step S4, the selection of homogeneous pixels using the AD test includes a statistical test and the sampling values ​​of two pixels. and The test statistic can be defined as:

[0035] In the formula, and Sampled values and The empirical cumulative distribution function, Let be the empirical cumulative distribution function of the two sample sets.

[0036] Further, in step S5, relevant thresholds are set based on the average amplitude map and coherence, and the selection of homogeneous points in the DS is completed in combination with the connectivity threshold. This includes the following steps:

[0037] S5.1: Set relevant thresholds based on average amplitude map and coherence. Based on the effect of the average amplitude map, and referring to conditions such as land-water boundaries, set an empirical threshold to eliminate non-DS points. Calculate the coherence of the entire DS block. If the overall coherence is less than the set empirical coherence threshold, it is considered a non-DS point.

[0038] S5.2: Based on the number of homogeneous pixels within the statistical homogeneous pixel detection window, set a connectivity empirical threshold to complete the final point selection for the DS block.

[0039] According to embodiments of the present invention, this method has the following advantages: it allows for full utilization of SAR data information, greatly enhancing the usability of SAR data. Furthermore, it further optimizes the DS selection process and quality without requiring additional data assistance. Using precisely matched temporal SAR images, multiple polarization parameters for each image are obtained based on polarization decomposition. These parameters are used as constraints in the DS homogeneous point selection process. Parameter constraint thresholds are set to optimize the distributed target selection results. Finally, a connectivity threshold is set based on coherence or average intensity maps to select the final DS point. Attached Figure Description

[0040] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0041] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0042] Figure 1 A flowchart illustrating a distributed target recognition method based on polarization parameter constraints provided in an embodiment of the present invention;

[0043] Figure 2 A flowchart illustrating the implementation of the distributed target recognition method based on polarization parameter constraints provided in this embodiment of the invention;

[0044] Figure 3 This is a multi-view average magnitude map in the distributed target recognition method based on polarization parameter constraints provided in this embodiment of the invention;

[0045] Figure 4 The polarization constraint condition diagram in the distributed target recognition method based on polarization parameter constraints provided in the embodiments of the present invention;

[0046] Figure 5 This refers to unoptimized DS point images in existing technologies.

[0047] Figure 6 This is an optimized DS point comparison image implemented in an embodiment of the present invention. Detailed Implementation

[0048] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0049] The terms such as "upper," "lower," "left," "right," and "middle" used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Any changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0050] like Figure 1 The diagram illustrates a flowchart of a distributed target recognition method based on polarization parameter constraints provided by an embodiment of the present invention. The method includes the following specific steps:

[0051] S1: Acquire N temporal SAR images and obtain the temporal SAR image stack of the region of interest through image preprocessing;

[0052] S2: Based on the time-series SAR image stack, calculate the amplitude, average amplitude, and coherence of each SAR image;

[0053] S3: Based on the temporal SAR image stack, calculate the polarization entropy of each pixel in each SAR image. H and average scattering angle α ;

[0054] S4: Set up a homogeneous pixel detection window to select homogeneous pixels and optimize the selection results.

[0055] S5: Based on the average amplitude and coherence, relevant thresholds are set, and the selection of homogeneous points in DS is completed by combining the connectivity threshold.

[0056] In the specific implementation process, such as Figure 2 As shown,

[0057] S1: First, acquire N SAR images, generate SLC data through format conversion, then perform image registration and image cropping. By performing multi-look processing and filtering on the cropped images, obtain a CUT_RSLCs image dataset with different polarization types for the region of interest. This embodiment of the invention uses Radarsat-2 fully polarimetric data from January 2020 to December 2021, containing 28 images. The revisit period is 24 days, the range to azimuth multi-look ratio is 2:2, and the study area size is 300*300 pixels. Figure 3The figure shown is a mean amplitude map of the study area.

[0058] S2: Based on the temporal SAR image stack, calculate the amplitude of each SAR image, and calculate the average amplitude and coherence, which includes the following steps:

[0059] S2.1: Aspect Ratio of Each Image Amp The following formula can be used to calculate:

[0060]

[0061] in, abs ( RSLC i To find the absolute value, i = 1, 2, 3, ..., n, n = 28 represents the number of images.

[0062] S2.2: Average amplitude Amp ave and coherence Coh The following formula can be used to calculate:

[0063]

[0064] Among them, E[yy *t [] represents the expected value, y = [y1, y2, ..., yn] T This represents the normalized result of multiple observations of homogeneous points of a distributed target on n SAR images, i.e. .

[0065] Step S3: The specific application is as follows: optimize the selection of homogeneous pixels.

[0066] Calculate and obtain the polarization entropy of each image. H and average scattering angle α .

[0067] 1) Image Reading. Based on the cropped multi-view SLC images, a script is used to input complex files into the MATLAB workspace to generate complex matrices of different polarization types for each image.

[0068] 2) Calculate the polarization coherence matrix T3. Construct the scattering matrix based on the complex matrix in S2.1. For fully polarimetric SAR data, the scattering matrix S can be expressed as:

[0069]

[0070] The scattering matrix can be represented by the target scattering vector k, when the propagation medium is reciprocal. S HV = S VHAt this point, the scattering matrix S is a complex symmetric matrix, and the scattering vector becomes a three-dimensional vector.

[0071]

[0072] Where T represents matrix transpose.

[0073] The scattering process of a distributed target can be understood as a statistical process in time or space. To describe the distributed target, it is necessary to perform a statistical average of the distributed target in time or space. The polarization coherence matrix T3 can be obtained by taking the outer product of the three-dimensional scattering vector and its complex conjugate.

[0074]

[0075] in, <k·k *t > indicates time average or spatial average.

[0076] 3) Calculate polarization entropy H and average scattering angle α For reciprocal media in a single-site scenario, H / α The decomposition method decomposes the polarization coherence matrix T3 into three statistically independent single-objective sums.

[0077]

[0078] in k i It is an eigenvector. λ represents the complex conjugate transpose. i It is an eigenvalue.

[0079] Based on this, the average scattering angle in the average physical scattering mechanism can be obtained. α .

[0080]

[0081] in,

[0082] in P i From the eigenvalue (λ) i The pseudo-probability obtained.

[0083] To define the statistical randomness of target scattering, we use scattering entropy, specifically polarization scattering entropy. H The parameter is defined as follows

[0084]

[0085] At this point, the polarization entropy of each image can be obtained. H and average scattering angle α matrix.

[0086] The data source used in this method is still backscattering information. The same type of land cover may not have the same scattering characteristics. Relying solely on land cover type constraints is not as reliable as relying on polarization entropy constraints.

[0087] S4: Set up a homogeneous pixel detection window to select homogeneous pixels and optimize the selection results.

[0088] Set the size of the homogeneous pixel detection window to m*n, that is, m rows and n columns, all of which are odd numbers. In this embodiment, m=n=11.

[0089] Using amplitude information as the data source, set the polarization entropy of each pixel within the window. H and average scattering angle α The constraints are as follows: when the constraints of the center pixel within the window are available and are the same as those of other pixels within the window, then all pixels within the window are subjected to an AD test with the window center point to determine homogeneous pixels; otherwise, no test is performed. The polarization entropy is calculated. H With average scattering angle α The constraint conditions are formulated as follows:

[0090]

[0091] Among them, Hseg i polarization entropy H Segmentation threshold, αseg i The average scattering angle α Segmentation threshold, F(Hseg) i αseg i () is a function of two thresholds, with specific constraints as follows: Figure 4 As shown, Hseg i The values ​​are 0.5 and 0.9, respectively, αseg i The values ​​are 40, 42.5, 47.5, 50, and 55. It can be seen that the constraint function... V Will H The constraints are divided into 9 types, each corresponding to a different scattering type. When the constraint conditions of the center pixel within the 11*11 detection window are available and are the same as those of other pixels within the window, then the Anderson-Darling (AD) test is performed on each pixel within the window and compared with the center point of the window to determine homogeneous pixels; otherwise, no test is performed.

[0092] In the above-described selection of homogeneous pixels using the Alternating Distributive (AD) test, the AD test is used to identify statistically homogeneous pixels. Compared to other testing methods, the AD test has proven to be the most effective method, placing more weight on the tails of the distribution, which results in a lower Type II error rate as assumed. For registered and calibrated SAR amplitude images, the sampled values ​​of two pixels... p and q The test statistic can be defined as .

[0093]

[0094] In the formula, N It is the number of images. and Sampled values p and q The empirical cumulative distribution function, Let be the empirical cumulative distribution function of the two sample sets, when If the values ​​are less than a certain threshold, then the two pixels are considered to belong to the same distribution.

[0095] Connectivity detection is performed on the central pixel and its homogeneous pixels within the homogeneous pixel detection window, and the number of homogeneous pixels within the homogeneous pixel detection window is counted.

[0096] S5: Based on the average amplitude plot and coherence setting, relevant thresholds are set, and combined with the connectivity threshold, the selection of homogeneous points in the DS is completed. This includes the following steps:

[0097] S5.1: Set empirical thresholds based on average amplitude map and coherence. Based on the effect of the average amplitude map, set empirical thresholds to eliminate non-DS points, calculate the coherence of the entire DS block. If the overall coherence is less than the set threshold, it is considered a decoherent region instead of a DS. In this embodiment, the average amplitude threshold is set to 0.23, and the coherence threshold is set to 0.3. These parameters are selected according to the actual application range and use universal numerical calibration. The specific parameter threshold settings need to be fine-tuned according to the image.

[0098] S5.2: Set the connectivity threshold. In this embodiment, the connectivity threshold is set to 80. The connectivity threshold specifies the number of DS points counted and is a step in DS technology. Its parameters are determined based on the initial recognition results, completing the final point selection work for the DS block. Please refer to... Figure 6 , Figure 6 This is the result of DS optimization and identification in an embodiment of the present invention. From... Figure 6 It can be seen that the polarization parameter is added during the DS point selection process. H and average scattering angle αThe constraints reduced the number of identified DS points from 6837 to 4778. The optimized DS selection results clearly separated the land and water boundaries and removed some areas that were corrected due to differences in polarization constraints. As can be seen from the examples, the optimized DS point selection results of the present invention are more accurate, proving the effectiveness of the optimization method of the present invention.

[0099] This invention provides a distributed target recognition method based on polarization parameter constraints. First, using precisely matched temporal SAR images, multiple polarization parameters are obtained for each image based on polarization decomposition. These parameters are used as constraints in the DS (Target Selection) homogeneous point selection process. Using polarization parameters as constraints before classification improves computational efficiency. Setting parameter constraint thresholds optimizes the distributed target selection results. Finally, a connectivity threshold is set based on coherence or average intensity maps to select the final DS points. The proposed method fully utilizes SAR data information, greatly enhancing its usability, and optimizes DS point selection results without the need for additional data assistance, resulting in more accurate results.

[0100] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A distributed target recognition method based on polarization parameter constraints, characterized in that... Includes the following steps: S1: Acquire N temporal SAR images and obtain the temporal SAR image stack of the region of interest through image preprocessing; S2: Based on the time-series SAR image stack, calculate the amplitude, average amplitude, and coherence of each SAR image; S3: Based on the temporal SAR image stack, calculate the polarization entropy of each pixel in each SAR image. H and average scattering angle α ; S4: Set up a homogeneous pixel detection window to select homogeneous pixels and optimize the selection results. Among them, the polarization entropy of each pixel within the window is set. H and average scattering angle α The constraint conditions are as follows: when the constraint conditions of the center pixel in the window are available and are the same as those of other pixels in the window, then all pixels in the window are subjected to AD test with the center point of the window to determine the homogeneous pixels; otherwise, no test is performed. S5: Set relevant thresholds based on average amplitude and coherence, and combine them with connectivity thresholds to select homogeneous points in DS; Based on the effect of the average amplitude map, an empirical threshold is set to remove non-DS points, and the coherence of the entire DS block is calculated. If the overall coherence is less than the set threshold, it is regarded as a decoherent region instead of DS. Set the connectivity threshold, the parameters of which depend on the initial recognition results, to complete the final point selection for the DS block.

2. The distributed target recognition method based on polarization parameter constraints according to claim 1, characterized in that, In step S1, RSLC images with different polarization types in the region of interest are obtained by performing multi-view and filtering preprocessing on the cropped image.

3. The distributed target recognition method based on polarization parameter constraints according to claim 2, characterized in that, In step S2, based on the time-series SAR image stack, the amplitude, average amplitude, and coherence of each SAR image are calculated, which includes the following steps: S2.1: Aspect Ratio of Each Image Amp The following formula can be used to calculate: ； Among them, abs (RSLC) i To find the absolute value, i = 1, 2, 3, ..., n, where n represents the number of images; S2.2: Average amplitude Amp ave and coherence Coh The following formula can be used to calculate: ； Among them, E[yy *t ] represents the expectation, y=[y1,y2,...,y n ] T This represents the normalized result of multiple observations of homogeneous points of a distributed target on n SAR images, i.e. .

4. The distributed target recognition method based on polarization parameter constraints according to claim 1, characterized in that, In step S3, the optimization of homogeneous pixel selection is specifically applied in the following steps. The polarization entropy of each image is calculated using the following formula. H and average scattering angle α ; ； Where S is the scattering matrix, k is the target scattering vector, T denotes matrix transpose, T3 is the polarization coherence matrix, and k i It is an eigenvector. λ represents the complex conjugate transpose. i It is an eigenvalue, P i From the eigenvalue (λ) i The pseudo-probabilities obtained are, where .

5. The distributed target recognition method based on polarization parameter constraints according to claim 4, characterized in that, The selection of homogeneous pixels in step S4 is as follows: S4.1: Set the size of the homogeneous pixel detection window to m*n, i.e., m rows and n columns, all of which are odd numbers; S4.2: Using amplitude information as the data source, set the polarization entropy of each pixel within the window. H and average scattering angle α The constraint conditions are as follows: when the constraint conditions of the center pixel in the window are available and are the same as those of other pixels in the window, then all pixels in the window are subjected to AD test with the center point of the window to determine the homogeneous pixels; otherwise, no test is performed. Calculate polarization entropy H With average scattering angle α The constraint conditions are formulated as follows: ； Among them, Hseg i polarization entropy H Segmentation threshold, αseg i The average scattering angle α The segmentation threshold, F, is a function of the two thresholds; S4.3: Perform connectivity detection on the central pixel and its homogeneous pixels within the homogeneous pixel detection window, and count the number of homogeneous pixels within the homogeneous pixel detection window.

6. The distributed target recognition method based on polarization parameter constraints according to claim 5, characterized in that, In step S4, the selection of homogeneous pixels using the AD test includes statistical testing and sampling values ​​of two pixels. and The test statistic can be defined as : ； In the formula, and Sampled values and The empirical cumulative distribution function, Let be the empirical cumulative distribution function of the two sample sets.

7. The distributed target recognition method based on polarization parameter constraints according to claim 1, characterized in that, In step S5, based on the average amplitude map and coherence setting, a relevant threshold is set, and combined with the connectivity threshold, the selection of DS homogeneous points is completed. This includes the following steps: S5.1: Based on the average amplitude map and coherence, set relevant thresholds. Based on the effect of the average amplitude map and referencing the water-land boundary conditions, set an empirical threshold to eliminate non-DS points. Calculate the coherence of the entire DS block. If the overall coherence is less than the set empirical coherence threshold, it is considered a non-DS point. S5.2: Based on the number of homogeneous pixels within the statistical homogeneous pixel detection window, set a connectivity empirical threshold to complete the final point selection for the DS block.

Images