A method for selecting homogeneous pixels filtering window in farmland area

Abstract

The present invention provides a method for selecting a homogeneous pixel filtering window in a farmland area, comprising Q1: time-series SAR image geocoding; Q2: stable scattering area selection; Q3: filtering window size calculation. The present invention selects a reasonable filtering window method when performing homogeneous pixel filtering in a farmland area, has a certain improvement effect on the decoherence phenomenon caused by seasonal changes in farmland, can better improve the signal-to-noise ratio of homogeneous pixels, and ultimately improves the accuracy of deformation solution; the method is universal, and can be used to estimate a reasonable filtering window size in any area with a large number of farmlands and using any SAR image. The present invention is simple and easy to understand in operation, can calculate a reasonable homogeneous pixel filtering window size, and reduces the technical threshold of data processing.

Description

Technical Field

[0001] The present invention relates to the technical field of big data analysis applications, and in particular to a method for selecting a homogeneous pixel filtering window in a farmland area. Background Art

[0002] In 2011, Ferretti et al. proposed the second-generation permanent scatterer technology SqueeSAR. Since then, the data processing of distributed scatterers (DS) has gradually become one of the research hotspots of time-series InSAR. Distributed scatterer targets are widely distributed in space. They refer to uniform objects that occupy a certain area in space, such as farmland, bare land, and grass. They show similar backscattering characteristics on the pixels corresponding to the radar image, that is, they have high consistency and maintain stable consistency within a certain spatial range. Different from the physical properties of permanent scatterers (PS), DS is easily affected by factors such as spatiotemporal decoherence, with low signal-to-noise ratio and uneven interference quality. Therefore, before deformation settlement, DS targets need to be screened and optimized to improve their signal-to-noise ratio and reduce the probability of error propagation.

[0003] DS data processing is mainly divided into two steps: homogeneous point selection and time-series DS optimization. Homogeneous point selection is to measure the similarity between the field pixel and the central pixel by statistical methods, and then select similar pixels for parameter estimation; DS optimization is to filter the DS target phase after the homogeneous point selection is completed to improve its coherence and signal-to-noise ratio.

[0004] At present, there are many methods for both steps of DS data processing, and each has its own characteristics and advantages. However, in the DS optimization step, any method needs to set a filter window for phase optimization. The size of the filter window is directly related to the effect of phase optimization and ultimately affects the deformation solution result. Therefore, setting a reasonable filter window is particularly important in DS processing. Due to the different distribution of objects in different research areas, there is currently no method for selecting the optimal filter window for a certain object scene. The selection of the filter window mainly depends on the experience and subjective judgment of the data processor, which makes the accuracy of DS processing and deformation solution affected by human subjective factors. Summary of the invention

[0005] In view of the problems described in the above background technology, the purpose of the present invention is to provide a method for selecting a filtering window for homogeneous pixels in a farmland area, so as to improve the signal-to-noise ratio of homogeneous pixels and ultimately achieve the purpose of improving the solution accuracy.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A method for selecting homogeneous pixel filtering windows in a farmland area comprises the following steps:

[0008] Q1: Geocoding of time-series SAR images: Select at least four scenes from the time-series SAR images, representing four seasons, for geocoding to obtain the backscatter coefficient image of the ground objects, i.e., the SAR image, and overlay it on Google Earth or optical remote sensing images to assist in subsequent analysis;

[0009] Q2: Selection of stable scattering area: Select any scene in the SAR image, select a basic farmland unit and save the selected vector file, use the optical remote sensing image as an auxiliary reference, outline the contour vector of the corresponding farmland unit in the remaining SAR images, and finally intersect the farmland vectors. The intersection area is the stable scattering area;

[0010] Q3: Calculation of filter window size: In the stable scattering area, take the longest side in the north-south direction as the rectangle length, and the longest side in the east-west direction as the rectangle width. The rectangle length divided by the azimuth length after multi-viewing of the SAR image is the azimuth size of the filter window; the rectangle width divided by the range length after multi-viewing of the SAR image is the range size of the filter window.

[0011] In the above technical solution, the steps of geocoding are as follows: first, the SAR image is input and multi-viewed to achieve the effect of reducing speckle noise and improving the visual quality of the image; then the multi-view image is filtered to further suppress the image noise; finally, the SAR image is geocoded and radiometrically calibrated using external DEM data to obtain the SAR backscatter coefficient image in the geographic coordinate system.

[0012] In the above technical solution, the basic farmland unit refers to the smallest dark area surrounded by four bright lines in the SAR image.

[0013] The present invention selects a reasonable filtering window method when performing homogeneous pixel filtering in a farmland area, which has a certain improvement effect on the decoherence phenomenon caused by seasonal changes in farmland, can better improve the signal-to-noise ratio of homogeneous pixels, and ultimately improve the accuracy of deformation solution; the method is universal, and can be used to estimate a reasonable filtering window size in any area with a large number of farmlands and using any SAR image; the present invention is simple and easy to understand, can calculate a reasonable homogeneous pixel filtering window size, and reduces the technical threshold of data processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0015] Figure 1 Geocoded Sentinel-1B SAR images;

[0016] Figure 2 Identification of basic farmland units;

[0017] Figure 3 The interference pattern obtained by conventional default range 21 azimuth 21 window filtering;

[0018] Figure 4 The interference pattern calculated by this method is obtained by filtering windows in 23 ranges and 9 azimuths. DETAILED DESCRIPTION

[0019] The technical solution of the present invention will be clearly and completely described below in combination with the embodiments of the present invention and the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] The present invention uses an embodiment to illustrate a method for selecting a homogeneous pixel filtering window in a farmland area, comprising the following steps:

[0021] Q1: Geocoding of time-series SAR images:

[0022] Distributed scatterer interferometry is an extension of permanent scatterer interferometry, which also belongs to time-series SAR. That is, in order to obtain reliable deformation results, generally no less than 20 SAR images are required. Considering the seasonal variation characteristics of farmland, in order to determine the minimum farmland block size that can maintain uniformity in a long time series, at least 4 images representing four seasons are selected from the time-series SAR images for geocoding. Figure 1 As shown, taking one scene as an example, in the embodiment, the geocoded Sentinel-1B SAR image is selected.

[0023] The steps of geocoding are as follows: First, input the SAR image and perform multi-looking on it to reduce speckle noise, improve the visual quality of the image. Subsequently, filter the multi-looked image to further suppress image noise. Finally, use external DEM data to perform geocoding and radiometric calibration on the SAR image to obtain the SAR backscattering coefficient image in the geographic coordinate system.

[0024] Q2: Selection of stable scattering areas:

[0025] The "field" - shaped texture structure with criss - cross farmlands is clearly visible in the SAR image. Select a basic farmland unit in any geocoded SAR image and save the selected vector file. A basic farmland unit refers to the smallest dark area surrounded by four bright lines (similar to a "square") in the SAR image. Using optical remote sensing images as an auxiliary reference, outline the contour vectors of the corresponding farmland units in the remaining SAR images. Finally, intersect the various farmland vectors, and the intersection area is the stable scattering area.

[0026] As Figure 2 shown, within the black square below the river channel, upon observation, most of the farmlands in the test area are in long strips. Select an area near the river channel where the farmland texture shape can be clearly identified as the basic farmland unit (black square). Use this area as a reference for the remaining SAR images. Outline the basic farmland unit vectors along the "bright lines" in their respective geocoded images, and finally take the intersection area for filtering window calculation.

[0027] This step is inspired by the atmospheric filtering idea in Persistent Scatterer Interferometry (PSI). That is, when separating the atmospheric effects in PSI, a 1 km × 1 km filtering window is selected to perform spatial low - pass filtering on the PS targets to separate the atmospheric phase. This is because the atmosphere shows a high degree of spatial autocorrelation, that is, the atmospheric state within 1 km is basically the same, and the atmospheric state changes slowly with the increase of distance. Therefore, the atmospheric phases of each PS within each 1 km × 1 km filtering window are basically the same. Using this filtering window for low - pass filtering can well remove the influence of atmospheric effects.

[0028] The selection of the filtering window for homogeneous pixels is also inspired by this, that is, by selecting the time - series stable scattering area, the size of the smallest farmland unit that can maintain uniformity at different times is determined, and then a reasonable filtering window size is determined to improve the filtering effect.

[0029] Q3: Calculation of the filtering window size:

[0030] The intersection area of ​​the farmland unit vectors drawn in the SAR images representing the four seasons is often not a regular rectangle, and the filter window calculation needs to determine a regular rectangle before calculation. Therefore, the longest side of the stable scattering area (i.e. the intersection area of ​​the farmland unit vectors drawn in the four SAR images) in the north-south direction is taken as the rectangle length, and the longest side in the east-west direction is taken as the rectangle width. The rectangle length divided by the azimuth length after multi-viewing of the SAR image is the azimuth size of the filter window; the rectangle width divided by the distance length after multi-viewing of the SAR image is the distance size of the filter window.

[0031] In this example, the basic farmland area was measured in Google Earth, and the intersection area rectangle was 900 meters long and 430 meters wide (Note: the so-called basic farmland in this method is not the actual size of a single farmland). The Sentinel-1B single-view complex image, that is, the SLC image is 5 meters in range and 20 meters in azimuth. 10:2 multi-view processing was used during processing, so the size of a single pixel is 50 meters in range and 40 meters in azimuth. According to the above filter window calculation method, the filter window size should be: 23 in range and 9 in azimuth.

[0032] like Figure 3 As shown, the interference map of the farmland area obtained by using the conventional default range window size 21 and azimuth size 21, and Figure 4 As shown, the interference patterns obtained by the range window 23 and the azimuth window 9 calculated by this method are compared as follows. It can be seen that the interference pattern obtained by the filter window calculated by this method has more obvious and smoother interference fringe patterns in the farmland area, and the coherence is significantly improved.

[0033] The core part of distributed scatterer technology, namely homogeneous pixel selection and homogeneous pixel filtering, involves a lot of knowledge of statistics and probability theory, which makes it difficult for general SAR data processing personnel to intuitively and reasonably set parameters when selecting parameters. It often takes many experiments to get the ideal results, which increases the processing threshold. This invention makes the selection of DS phase optimized filter window have a reasonable basis, and at the same time makes the selected filter window reach the approximate optimal size, improves the signal-to-noise ratio of homogeneous pixels, and ultimately achieves the purpose of improving the solution accuracy.

[0034] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for selecting homogeneous pixel filtering windows in farmland areas. Features: The steps include: Q1: Geocoding of time-series SAR images: Select at least four scenes from the time-series SAR images, representing four seasons, for geocoding to obtain the backscatter coefficient image of the ground objects, i.e., the SAR image, and overlay it on Google Earth or optical remote sensing images to assist in subsequent analysis; Q2: Selection of stable scattering area: Select any scene in the SAR image, select a basic farmland unit and save the selected vector file, use the optical remote sensing image as an auxiliary reference, outline the contour vector of the corresponding farmland unit in the remaining SAR images, and finally intersect the farmland vectors. The intersection area is the stable scattering area; Q3: Calculation of filter window size: In the stable scattering area, take the longest side in the north-south direction as the rectangle length, and the longest side in the east-west direction as the rectangle width. The rectangle length divided by the azimuth length after multi-viewing of the SAR image is the azimuth size of the filter window; the rectangle width divided by the range length after multi-viewing of the SAR image is the range size of the filter window.

2. A method for selecting homogeneous pixel filtering windows in farmland areas according to claim 1, Features: The steps of geocoding are as follows: first, the SAR image is input and multi-viewed to reduce the speckle noise and improve the visual quality of the image; then, the multi-view image is filtered to further suppress the image noise; finally, the SAR image is geocoded and radiometrically calibrated using external DEM data to obtain the SAR backscatter coefficient image in the geographic coordinate system.

3. The method for selecting homogeneous pixel filtering windows in farmland area according to claim 1, Features: The basic farmland unit refers to the smallest dark area surrounded by four bright lines in the SAR image.

Images