A method for selecting homogeneous pixels in DS-InSAR considering topographic features

Abstract

The application discloses a kind of DS-InSAR homogeneous pixel selection methods considering topographic features, comprising: SAR image data is preprocessed;Based on external DEM data, the slope, slope and ground cover type of target area are extracted;Ensure that SAR image data and topographic features data are in the same coordinate system;Combining slope, slope and ground cover type topographic features data, the identification and selection of homogeneous pixel are carried out;Through hard threshold selection algorithm and soft threshold selection algorithm, the selection process of homogeneous pixel is optimized;The selection accuracy of homogeneous pixel is calculated and analyzed;Using the selection point result of homogeneous pixel, in combination with DS-InSAR data processing flow, the deformation rate of target area is solved.The application improves the uniformity and accuracy of DS-InSAR homogeneous pixel selection, effectively eliminates the scattering variation pixel caused by topographic features, and improves the accuracy of deformation rate solution.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of geographic information systems and time series InSAR technology, and particularly relates to a DS-InSAR homogeneous pixel selection method considering topographic features. BACKGROUND

[0002] In distributed scatterer interferometry (DS-InSAR), the selection of homogeneous points is a key step to achieve high-precision ground deformation monitoring. Homogeneous points refer to a set of pixels in time series SAR (Synthetic Aperture Radar) images that have the same backscattering properties and the same phase scattering center. These points usually represent areas on the ground surface with relatively stable physical properties and geometric structures, such as urban buildings, rock outcrops, water bodies, or other man-made surfaces. Existing permanent scatterer interferometry (PS-InSAR) methods rely on the density of stable scatterers, and when applied in large-scale areas, they face challenges such as explosive growth of PS points and uneven distribution of data. In suburban areas and other areas with more natural features, the density of stable scatterers is often low, so DS-InSAR technology is developed to increase the density of measurement scatterers (MS). In deformation analysis, time series InSAR technology with strong noise resistance, long distance, and high resolution for monitoring large-scale scenes is favored. Homogeneous points can provide stable and reliable reference benchmarks, allowing small deformation signals to be separated from noise, which is particularly important for fine deformation monitoring of complex large-scale scenes and can significantly improve the accuracy and reliability of monitoring results. With the development of SAR technology, long-time series deformation monitoring has become a key means. The quality of homogeneous point selection directly affects the accuracy of InSAR time series analysis. If the selected homogeneous points are inaccurate, it may lead to deformation estimation bias, and thus affect the final deformation monitoring results. In addition, the spatial distribution density of homogeneous points also has an important influence on the resolution and coverage of the monitoring results. Therefore, developing high-quality homogeneous point selection algorithms is of great significance to improve the practicality and reliability of DS-InSAR technology.

[0003] Currently, research on homogeneous point selection algorithms focuses on improving their computational efficiency, enhancing their adaptability, and improving their selection accuracy. Existing selection algorithms, such as empirical threshold-based coherence screening, often rely on manually set parameters. This can be inflexible under varying surface conditions and difficult to adapt to complex land cover types. Researchers have proposed a variety of homogeneous point selection algorithms based on statistical models. These algorithms analyze time series of pixel intensities and use hypothesis testing to identify homogeneous points. For example, by constructing a statistical model of pixel intensities, the probability of similarity between pixels can be calculated, thereby assessing homogeneity in a probabilistic manner. Currently, homogeneous point selection algorithms are mainly divided into two categories: parametric and nonparametric. Parametric statistical methods, such as the FaSHPS algorithm, are favored for their fast computational speed and low memory consumption, but may not be robust when dealing with small sample sizes. Nonparametric statistical methods, such as the Baumgartner-Wei-Schindler (BWS) test, exhibit greater robustness when dealing with non-Gaussian distributed data. While existing homogeneous point selection algorithms have made progress in some areas, they still face challenges, such as computational complexity, sensitivity to outliers, and applicability to non-uniform surface conditions. These challenges require further optimization to adapt to diverse monitoring scenarios and data characteristics. Furthermore, these methods are prone to the phenomenon of heterogeneous objects with the same spectrum in complex terrain, misclassifying heterogeneous pixels as homogeneous. Summary of the Invention

[0004] The present invention proposes a new homogeneous point selection technology, which optimizes the selection process of homogeneous points by introducing external terrain data, such as surface feature data such as slope, aspect, and land cover type. This technology not only improves the adaptability and selection accuracy of the algorithm, but also can effectively eliminate scattered variation pixels caused by terrain features, thereby improving the accuracy of subsequent deformation rate solutions. Compared with existing technologies, the new technology has shown obvious advantages in ensuring the uniform distribution of homogeneous points in space and improving selection accuracy. Especially in areas with drastic terrain changes, the new technology can significantly reduce the risk of mistakenly judging heterogeneous pixels as homogeneous pixels, thereby confirming its applicability and efficiency in dealing with complex terrain.

[0005] In view of the shortcomings of the prior art, the present invention provides a DS-InSAR homogeneous pixel selection method that takes into account topographic features, comprising the following steps:

[0006] Step 1: Preprocess the SAR image data to ensure data quality and usability;

[0007] Step 2: Extract the slope, aspect, and land cover type of the target area based on the external DEM data using the following formulas;

[0008]

[0009] in, Indicates the rate of change in the horizontal direction starting from the center pixel. Indicates the rate of change in the vertical direction starting from the center pixel, tan -1 is the inverse tangent function, tan2 -1 is a four-quadrant inverse tangent function. The slope calculated by the formula ranges from 0 to 90°, and the aspect ranges from 0 to 360°.

[0010] Step 3: Ensure that the SAR image data and terrain and geomorphological feature data such as slope, aspect, and land cover type are in the same coordinate system.

[0011] Step 4: Combine terrain and geomorphological feature data such as slope, aspect, and land cover type to identify and select homogeneous pixels.

[0012] Step 5: Optimize the selection process of homogeneous pixels through hard threshold selection algorithm and soft threshold selection algorithm. As shown in formula (3) and formula (4):

[0013]

[0014] Where, Indicates p i The sample mean of , P{p} represents the interval estimate of the pixel point p to be estimated, For p i The ground cover type, D P is the coverage type of the central pixel point p to be estimated.

[0015]

[0016] Where CI is the confidence interval, is the weighted sample mean, is the weighted sample variance, N is the number of samples, Represents the standard normal distribution Percentile.

[0017] Step 6: Calculate and analyze the selection quality and accuracy of homogeneous pixels.

[0018] Step 7: Using the point selection results of homogeneous pixels and combining them with the DS-InSAR data processing flow, the deformation rate of the target area is calculated.

[0019] Preferably, step 3 includes the following sub-steps:

[0020] Step 3.1: Check whether the SAR intensity image and the terrain feature data are in the same coordinate system. If not, proceed to step 3.2; if they are, skip step 3.2.

[0021] Step 3.2: Convert both to the same coordinate system.

[0022] Step 3.3: For ease of calculation, convert both to the WGS84 geographic coordinate system.

[0023] Preferably, step 6 and step 7 respectively include the following sub-steps:

[0024] Step 6.1: Obtain the homogeneous point quantity information file and evaluate its distribution uniformity.

[0025] Step 6.2: Obtain the homogeneous point location information file and analyze the point selection accuracy.

[0026] Step 7.1: Based on the point selection results, perform phase optimization on the interferogram.

[0027] Step 7.2: Perform phase unwrapping on the optimized interferogram.

[0028] Step 7.3: Perform deformation calculation to obtain the deformation rate result of the target area.

[0029] The above solution of the present invention has the following beneficial effects:

[0030] 1. This invention improves the uniformity and accuracy of DS-InSAR homogeneous pixel selection by comprehensively considering topographic features. In particular, in areas with large terrain undulations, it can effectively eliminate scattered variation pixels caused by topographic features.

[0031] 2. When selecting homogeneous pixels, the present invention significantly reduces the misselection of heterogeneous pixels while maintaining a high signal-to-noise ratio, thereby improving the accuracy of deformation rate solution and proving its applicability and effectiveness under complex terrain conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Schematic diagram of the process of selecting homogeneous pixels of DS-InSAR taking into account topographic and geomorphic features of the present invention;

[0033] Figure 2 is the SAR image coverage of the study area in the embodiment of the present invention;

[0034] Figure 3 is the land feature coverage type of the study area of ​​the embodiment of the present invention;

[0035] Figure 4 1 is a map of the local central pixel (red) and its homogeneous pixels (yellow) in the study area of ​​the embodiment;

[0036] Wherein (a), (d) use prior art algorithm; (b), (e) use hard threshold algorithm; (c), (f) use soft threshold algorithm;

[0037] (a), (b), (c) are the selected point results in the average intensity image background;

[0038] (d), (e), (f) are the selected point results in the ground object cover type classification image background;

[0039] Figure 5 is the deformation rate calculation result of the embodiment based on the prior selected point technology, and the red round frame is landslide area A and landslide area B;

[0040] Figure 6 is the deformation rate calculation result of the embodiment based on the new technology of the present application, and the red round frame is landslide area A and landslide area B;

[0041] Figure 7 is the deformation result of the embodiment of landslide area A using the prior art algorithm, and the white frame indicates the approximate range;

[0042] Figure 8 is the deformation result of the embodiment of landslide area A using the method of the present application, and the white frame indicates the approximate range;

[0043] Figure 9 is the time sequence deformation result of the feature points P1 and P2 of the embodiment of landslide area A;

[0044] Figure 10 is the deformation result of the embodiment of landslide area B using the prior art algorithm, and the white frame indicates the approximate range;

[0045] Figure 11 is the deformation result of the embodiment of landslide area B using the method of the present application, and the white frame indicates the approximate range;

[0046] Figure 12 is the time sequence deformation result of the feature points P3 and P4 of the embodiment of landslide area B. DETAILED DESCRIPTION

[0047] The specific embodiments of the present application are described below to facilitate the understanding of the present application by those skilled in the art, but it should be clear that the present application is not limited to the scope of the specific embodiments, and for those skilled in the art, as long as various changes are within the spirit and scope of the present application defined and determined by the appended claims, all the inventions utilizing the concept of the present application are within the scope of protection.

[0048] As Figure 1The process of selecting homogeneous pixels for DS-InSAR considering topographic features includes the following steps:

[0049] Step 1: Perform preprocessing such as denoising and registration on the SAR image data to ensure the quality and availability of the data.

[0050] Step 2: Use formula (1) and formula (2) to extract the slope, aspect and land cover type of the target area based on external DEM data.

[0051]

[0052] in, Indicates the rate of change in the horizontal direction starting from the center pixel. Indicates the rate of change in the vertical direction starting from the center pixel, tan -1 is the inverse tangent function, tan2 -1 is a four-quadrant inverse tangent function. The slope calculated by the formula ranges from 0 to 90°, and the aspect ranges from 0 to 360°.

[0053] Step 3: To minimize the impact of different coordinate bases on the matching results, check the coordinate systems of the SAR image data and the terrain feature data. If they are inconsistent, convert them to the same coordinate system.

[0054] Step 4: Combine terrain and geomorphological feature data such as slope, aspect, and land cover type to identify and select homogeneous pixels.

[0055] Step 5: Optimize the selection process of homogeneous pixels through hard threshold selection algorithm and soft threshold selection algorithm.

[0056] The hard threshold selection algorithm combines the coverage type of the object with the statistical test. By introducing the coverage type information, the algorithm can take into account the scattering characteristics of different objects, thereby achieving more accurate homogeneous point identification. Specifically, with the help of the coverage type of the object, the algorithm can identify and exclude non-homogeneous pixels caused by changes in the coverage of the object, which may be mistakenly included in the process of selecting homogeneous points through statistical tests. In the optimization algorithm, for a pixel point p to be estimated, there are p under a given window size. i , i = 1, 2, ..., k, a total of k pixels, for the remaining k-1 pixels except the central pixel p, the following formula is used to select homogeneous points and screen out heterogeneous points:

[0057]

[0058] Where, Indicates p iThe sample mean of , P{p} represents the interval estimate of the pixel point p to be estimated, For p i The land cover type, D P is the coverage type of the central pixel point p to be estimated.

[0059] The soft threshold selection algorithm combines data on slope, aspect, and surface cover type with SAR image intensity information, providing richer information for homogeneous pixel selection. It also optimizes statistical tests using external data and adjusts test parameters to adapt to scattering characteristics under different terrain and surface conditions. In the prior art, interval estimates are typically determined based on hypothesis testing. The present invention employs a weighted statistical approach, adjusting sample weights based on terrain and surface features to achieve more accurate interval estimates. For each pixel p, a weight factor is determined based on its slope, aspect, and surface cover type attributes. The surface cover type weight factor can be determined based on prior knowledge, while the slope and aspect weight factors can be determined based on statistical analysis to reflect the influence of different surface features on SAR backscattering characteristics. The sample mean in the prior art is calculated by summing the intensity values ​​of all samples and dividing it by the number of samples. With the introduction of weighting factors, the sample mean calculation becomes a weighted average: the intensity value of each sample is multiplied by its corresponding weight factor and then divided by the sum of all weighting factors. Similarly, the sample variance also needs to be weighted. The weighted variance takes into account the weight of each sample to reflect the contribution of different surface features to the scattering variability. Assuming that the distribution of the sample mean is close to a normal distribution, the interval estimate with a confidence level of 1-α can be expressed as follows:

[0060]

[0061] Where CI is the confidence interval, is the weighted sample mean, is the weighted sample variance, N is the number of samples, Represents the standard normal distribution Percentile.

[0062] Step 6: Calculate and analyze the selection quality and accuracy of homogeneous pixels.

[0063] Step 7: Using the point selection results of homogeneous pixels and combining them with the DS-InSAR data processing flow, the deformation rate of the target area is calculated.

[0064] This invention effectively addresses the key issues with conventional DS-InSAR homogeneous pixel selection, providing a DS-InSAR homogeneous pixel selection method that takes into account topographical features. This method fully considers the impact of topographical features on homogeneous pixel selection, making it suitable for homogeneous pixel selection in complex terrain. The mathematical model is clear, and the selection results can be used for a variety of applications based on practical needs.

[0065] When the present invention is specifically implemented, the implementation steps based on the example data are as follows:

[0066] Data preparation: SAR image data uses Sentinel-1 data products; external DEM data uses SRTM3 DEM products; external land cover type products use GlobeLand30 datasets.

[0067] Step 1: Perform denoising and registration operations on Sentinel-1 SAR image data to ensure data quality and usability.

[0068] Step 2: Using formula (1) and formula (2), extract the slope, aspect and land cover type of the target area based on SRTM3 DEM data and GlobeLand30 dataset. The scope of the study area is as follows: Figure 2 , the ground cover type is as follows Figure 3 shown.

[0069] Step 3: Check whether the coordinate systems of the SAR image data and surface feature data such as slope, aspect, and land cover type are consistent. The SAR image coordinate system is the radar coordinate system, and the surface feature data is the WGS84 geographic coordinate system. Unify the two into the WGS84 geographic coordinate system.

[0070] Step 4: Combine terrain and geomorphological feature data such as slope, aspect, and land cover type to identify and select homogeneous pixels.

[0071] Step 5: Use formulas (3) and (4) to optimize the selection process of homogeneous pixels through hard threshold selection algorithm and soft threshold selection algorithm. First, use external surface feature data to distinguish different types of objects in the study area. Then, based on the slope, aspect, and object cover type information of each pixel, the algorithm adjusts the parameters of the statistical test, such as threshold and weight, to adapt to the specific scattering characteristics of the type, construct a new interval estimate, and finally select high-quality homogeneous pixels.

[0072] Step 6: Calculate and analyze the selection accuracy of homogeneous pixels to ensure that the selected homogeneous pixels have high uniformity and accuracy.

[0073] Step 7: Using the selected results of the homogeneous pixels, combined with the DS-InSAR data processing flow, the deformation rate of the target area is calculated.

[0074] As shown in Figure 4 , the selected results of the example data using the existing selection technology and the new technology of the application are shown. The existing technology selects 51 pixels as homogeneous samples in the window, the hard threshold algorithm of the application selects 49 pixels as homogeneous samples in the same window, and the soft threshold algorithm selects 44 pixels as homogeneous samples in the same window. As can be seen from (d), (e) and (f) in Figure 4 , the hard threshold algorithm eliminates 2 pixels belonging to the forest type, which is different from the shrub type of the center pixel, and the soft threshold algorithm further eliminates 5 pixels on this basis. This is to select homogeneous pixels by statistical test means, refitting SAR image intensity information, slope, slope direction and land cover information interval estimation, and more accurately eliminate heterogeneous pixels from the center pixel.

[0075] The subsequent deformation calculation results are shown in Figure 5 and Figure 6 , the new algorithm of the application shows higher accuracy and reliability in the calculation of deformation rate, especially in landslide area A and landslide area B, the new algorithm can provide more continuous and smooth deformation rate distribution, which shows that the new algorithm can effectively capture the small changes of the ground surface, especially in the area with large terrain undulations.

[0076] Further analysis of the deformation rate results of landslide area A and landslide area B shows that Figure 7 , Figure 8 and Figure 10 , Figure 11 , it can be seen that the deformation rate graph of the existing technology algorithm in the landslide area is relatively chaotic, and part of the deformation characteristics is covered by noise, which is caused by the misselection of heterogeneous points due to only considering SAR image intensity information. In contrast, the deformation rate graph calculated by the new algorithm of the application is basically consistent with the landslide trend and range shown by the optical image, which further verifies the applicability and effectiveness of the new algorithm in complex terrain conditions.

[0077] Figure 9 and Figure 12 respectively show the time series deformation results of four feature points in the landslide area based on the existing technology algorithm and the new algorithm of the application. The new algorithm can provide more stable and accurate deformation trend at these key feature points, which further proves the superiority of the new algorithm in deformation monitoring, especially in the mutation area of deformation rate, the new algorithm can more accurately capture the start and end of the deformation signal, which is crucial for early warning and risk assessment of landslide disasters.

Claims

1. A DS-InSAR homogeneous pixel selection method taking into account topographic features, characterized by: The following steps are involved: Step 1: Preprocess the SAR image data to ensure data quality and usability; Step 2: Extract the slope, aspect, and land cover type of the target area based on the external DEM data using the following formulas; in, Indicates the rate of change in the horizontal direction starting from the center pixel. Indicates the rate of change in the vertical direction starting from the center pixel, tan -1 is the inverse tangent function, tan2 -1 is a four-quadrant inverse tangent function; the slope calculated by the formula ranges from 0 to 90°, and the aspect ranges from 0 to 360°; Step 3: Ensure that the SAR image data and the terrain and geomorphological feature data (slope, aspect, and land cover type) are in the same coordinate system; Step 4: Combine the topographic and geomorphic feature data of slope, aspect, and land cover type to identify and select homogeneous pixels; Step 5: Optimize the selection process of homogeneous pixels through hard threshold selection algorithm and soft threshold selection algorithm; as shown in formula (3) and formula (4): Where, Indicates p i The sample mean of , P{p} represents the interval estimate of the pixel point p to be estimated, For p i The land cover type, D P is the ground object coverage type of the central pixel point p to be estimated; Where CI is the confidence interval, is the weighted sample mean, is the weighted sample variance, N is the number of samples, Represents the standard normal distribution quantiles; Step 6: Calculate and analyze the selection quality and accuracy of homogeneous pixels; Step 7: Using the point selection results of homogeneous pixels and combining them with the DS-InSAR data processing flow, the deformation rate of the target area is calculated.

2. The DS-InSAR homogeneous pixel selection method taking into account topographic features according to claim 1, characterized in that: Step 3 includes the following sub-steps: Step 3.1: Check whether the SAR intensity image and the terrain feature data are in the same coordinate system; if not, proceed to step 3.2; if they are consistent, skip step 3.2; Step 3.2: Convert both to the same coordinate system; Step 3.3: For ease of calculation, convert both to the WGS84 geographic coordinate system.

3. The DS-InSAR homogeneous pixel selection method taking into account topographic features according to claim 1, characterized in that: Step 6 includes the following sub-steps: Step 6.1: Obtain the homogeneous point quantity information file and evaluate its distribution uniformity; Step 6.2: Obtain the homogeneous point location information file and analyze the point selection accuracy.

4. The DS-InSAR homogeneous pixel selection method taking into account topographic features according to claim 1, characterized in that: The step 7 includes the following sub-steps: Step 7.1: Based on the point selection results, perform phase optimization on the interferogram; Step 7.2: Perform phase unwrapping on the optimized interferogram; Step 7.3: Perform deformation calculation to obtain the deformation rate result of the target area.

Images