A high-precision flood inundation area monitoring method based on dual-polarization SAR images

Abstract

The present invention belongs to the technical field of remote sensing image processing, and relates to a high-precision flood inundation area monitoring method based on dual-polarization SAR images. Multi-scale segmentation is performed on a false color image generated by combining pre-processed dual-polarization data; a Gaussian mixture model is used to fit the histogram of SAR images under different polarizations, and the optimal SAR image is selected by comparing the size of the overlapping area of ​​the Gaussian mixture model; the optimal SAR image is classified by combining the gray value fuzzy system and the texture feature fuzzy system to extract water body information; the water body extraction results of SAR before and after the flood are compared to obtain the flood inundation area. The method performs multi-scale segmentation based on the false color image synthesized by dual polarization, which effectively reduces the influence of SAR coherent speckle noise; and the selection of the optimal SAR image and the method of classifying the uncertain area by combining the gray value and texture features effectively reduce the influence of ground object mixing and significantly improve the accuracy of flood inundation area monitoring.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing image processing, and in particular relates to a high-precision flood inundation area monitoring method based on dual-polarization SAR images. Background Art

[0002] Flood disasters are characterized by wide range and suddenness, causing serious losses. Accurate and timely monitoring of flood-affected areas is crucial for early warning, emergency response and reducing disaster losses. Synthetic aperture radar (SAR) has the characteristics of all-day, all-weather and multi-polarization, which can provide high-quality data support for flood inundation area monitoring.

[0003] At present, there are two main methods for extracting water bodies based on SAR images: pixel-based and object-based. The pixel-based method only considers the scattering characteristics of a single pixel, but ignores the relationship between pixels and factors such as texture and shape, so it is easily affected by coherent speckle noise. In comparison, the object-based image analysis method (OBIA) takes multi-pixel uniform targets containing the same information as the basic analysis unit. When extracting targets, it not only considers the pixel backscattering coefficient, but also considers the spatial relationship between adjacent pixels. This method can solve the "salt and pepper" noise phenomenon in traditional pixel-based methods and improve the accuracy of water body extraction. However, most object-based methods are based on one polarization mode and do not make full use of multi-polarization data, which may cause misjudgment or omission when processing some complex ground scene. In addition, SAR images are easily affected by factors such as vegetation, urban buildings, and weather, resulting in ground object mixing, resulting in low monitoring accuracy of flood inundation areas. Therefore, how to effectively remove coherent speckle noise in SAR images and how to effectively reduce the impact of ground object mixing to improve the monitoring accuracy of flood inundation areas are technical problems that technicians in this field need to solve urgently.

[0004] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgment or any form of suggestion that the information constitutes the prior art already known to a person skilled in the art. Summary of the invention

[0005] The purpose of the present invention is to solve the problem in the prior art that when using SAR images to monitor floods, the SAR images are easily affected by coherent speckle noise and mixed ground objects, which leads to low monitoring accuracy of flood inundation areas. A high-precision flood inundation area monitoring method based on dual-polarization SAR images is proposed. The method performs multi-scale segmentation based on dual-polarization synthesized false color images, which not only effectively reduces the influence of SAR coherent speckle noise, but also makes up for the deficiency of single-polarization data; and the selection of the best SAR image and the method of classifying uncertain areas by combining grayscale value and texture features effectively reduce the influence of mixed ground objects and significantly improve the monitoring accuracy of flood inundation areas.

[0006] The technical solution of the present invention is:

[0007] A high-precision flood inundation area monitoring method based on dual-polarization SAR images comprises the following steps:

[0008] (a) Preprocess the dual-polarization SAR images, combine the preprocessed SAR images to generate false color images, and perform multi-scale segmentation based on the false color images. In the segmentation process, digital elevation model (DEM) data is introduced for mask processing;

[0009] (b) The preprocessed SAR images were preliminarily classified using the maximum between-class variance (OTSU) threshold method to obtain the initial parameters of the Gaussian mixture model (GMM). The Gaussian mixture model parameters were estimated using the maximum expectation algorithm (EM). The overlapping area sizes of the SAR Gaussian mixture models under different polarizations were compared to select the best SAR images for monitoring flood disasters.

[0010] (c) Preliminary classification of the best SAR image based on the probability threshold, water body probability P( x )The formula is:

[0011] Among them, P( x ) represents the probability that the gray value is water when it is x. are the optimal Gaussian mixture model parameters; where, is the optimal probability that the gray value belongs to the water distribution in the Gaussian mixture model; are the optimal mean and variance of water data in the Gaussian mixture model, respectively; is the optimal probability that the gray value belongs to the non-water distribution in the Gaussian mixture model; are the optimal mean and variance of non-water data in the Gaussian mixture model, respectively;

[0012] The above probability threshold is a set standard value. If the water body probability P( x ) is higher than the set probability threshold, the area is usually judged as a water body; if the water body probability P( x) is lower than the probability threshold, it is judged as a non-water body.

[0013] The optimal probability threshold is determined by evaluating the false positive rate of water bodies and non-water bodies under different thresholds. , and based on the threshold, the best SAR image is divided into water body, uncertain area and non-water body; The smaller the false positive rate is, the higher the accuracy of the water body or non-water body extracted under the threshold condition is. The corresponding evaluation work is carried out by comparing the false positive rate.

[0014] Combining gray value fuzzy system and texture feature fuzzy system to classify uncertain areas and extract water body information;

[0015] (d) By comparing the water body information of SAR images before and after the flood disaster, the flood inundation area is obtained.

[0016] Furthermore, in the step (a), the multi-scale segmentation is performed based on the gray value, texture, color and other characteristics of the pixels of the false color image synthesized from the dual-polarization SAR image, so as to make full use of the multi-polarization data, make up for the deficiency of the single-polarization data, and effectively remove the influence of SAR speckle noise. In addition, the DEM is used for mask processing during the segmentation process to speed up the processing speed and effectively remove the influence of the mountain shadow on the monitoring of the flood inundation area.

[0017] Furthermore, in the step (b), the step of fitting the SAR image histogram under different polarizations using a Gaussian mixture model includes:

[0018] The land-water segmentation problem can be regarded as a binary classification problem, in which the Gaussian mixture model is used to model the mixed distribution of water and land pixels in SAR images as follows: Where x is the gray value of the SAR image pixel, It represents the probability value when the gray value is x. is the Gaussian distribution probability density function of the water body, is the Gaussian distribution probability density function of non-water bodies, represents the probability that x belongs to the water distribution, and represents the probability that x belongs to the non-water distribution, and .

[0019] Furthermore, in step (b), in order to obtain the above-mentioned Gaussian mixture model parameters, the SAR image is first preliminarily classified using the OTSU threshold method to obtain water bodies and non-water bodies; then, based on the above-mentioned preliminary classification results, the mean and variance of the water bodies and non-water bodies are calculated and used as the initial parameters of the Gaussian mixture model; finally, the maximum expectation algorithm is applied to iteratively optimize the Gaussian mixture model to obtain the optimal model parameters.

[0020] Furthermore, in step (b), the Gaussian mixture model overlap region is extracted using a distribution probability threshold. , to extract, the steps are as follows:

[0021] First, based on the Gaussian mixture model of the SAR image histogram under different polarizations calculated in the above steps, the probability P of water distribution under different polarizations is calculated. 水 equal The maximum gray value T H , and when the probability P in the non-water distribution 非水 equal The minimum gray value T L ;

[0022] in, , x is the gray value of the SAR image pixel, is the optimal probability that the gray value belongs to the water distribution in the Gaussian mixture model, are the optimal mean and variance of water data in the Gaussian mixture model, respectively; , x is the gray value of the SAR image pixel, is the optimal probability that the gray value belongs to the non-water distribution in the Gaussian mixture model; are the optimal mean and variance of non-water data in the Gaussian mixture model, respectively;

[0023] Then, calculate the difference between these two thresholds T = T H -T L , determine the grayscale range of the overlapping part;

[0024] Finally, the sizes of overlapping areas of Gaussian mixture models corresponding to SAR images under different polarizations are compared to select the best SAR image.

[0025] That is, after obtaining the Gaussian mixture models of SAR images under different polarizations (VV, VH, VV+VH), the corresponding overlapping areas are calculated, and the sizes of overlapping areas under different polarizations are compared to select the best SAR image for flood inundation area monitoring.

[0026] Furthermore, in step (c), the fuzzy membership function corresponding to the gray value fuzzy system is a Z function, whose parameters are determined by the optimal probability threshold The fuzzy membership function corresponding to the texture feature fuzzy system is the S function, and its parameters are obtained by the mean and standard deviation of the angular second-order moment corresponding to the non-permanent water body.

[0027] Furthermore, the corresponding texture feature (angular second-order moment: ASM) in the texture feature fuzzy system is calculated using the gray-level co-occurrence matrix.

[0028]

[0029] in, Ng The gray level of the image after quantization; , which can be understood as normalizing the gray-level co-occurrence matrix ( n ( i , j ) is the number of times pixel pairs appear in the gray-level co-occurrence matrix).

[0030] The fuzzy membership function of the texture feature fuzzy system corresponds to the S function, and its parameters are obtained by the following steps:

[0031] The permanent water bodies obtained in step (c) are removed from the water bodies using the WorldCover data, a 10-meter global land cover data for 2020 jointly produced by the European Space Agency and a number of scientific research institutions around the world. The angular second-order moment of the non-permanent water bodies after removing the permanent water bodies is then calculated, and the corresponding mean and standard deviation are calculated to obtain the parameters of the S fuzzy membership function.

[0032] The gray value membership and the angular second-order moment membership are combined to classify the uncertain areas, and finally the water body extraction in SAR images is realized.

[0033] Furthermore, in the step (d), the water body extraction result of the reference SAR image before the flood disaster is compared with the water body extraction result of the flood SAR image after the disaster to obtain the spatial distribution of the flooded area.

[0034] Beneficial effects of the present invention:

[0035] The monitoring method provided by the present invention makes full use of dual-polarization data in the image segmentation process, and performs multi-scale segmentation on the pseudo-color image synthesized by dual-polarization, which not only makes up for the shortcomings of single-polarization data, but also effectively reduces the influence of SAR coherent speckle noise. During classification, the SAR image most suitable for monitoring flood and waterlogging inundation areas is first selected based on the size of the overlapping area of ​​the Gaussian mixture model, thereby reducing the influence of ground object mixing. Subsequently, in order to further reduce the influence of ground object mixing, the uncertain area is classified in combination with gray value and texture feature information, which significantly improves the accuracy of flood and waterlogging inundation area monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 The present invention is a flow chart of monitoring flooded areas based on Sentinel-1 SAR dual-polarization data;

[0037] Figure 2 Examples of multi-scale segmentation results under different polarization conditions; specifically, they are multi-scale segmentation results obtained based on VV, VH, and false color images;

[0038] Figure 3 The Gaussian mixture model fitting results of SAR image histogram under different polarizations;

[0039] Figure 4 This is an example of the flood inundation area extraction results. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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.

[0041] In order to further understand the present invention, the present invention will be further described in conjunction with the accompanying drawings and embodiments.

[0042] like Figure 1 As shown, the present invention provides a high-precision flood inundation area monitoring method based on dual-polarization SAR images. Sentinel-1 SAR dual-polarization data is used to monitor flood disasters in the eastern coastal area of ​​Leizhou Peninsula. In a specific embodiment, the IW mode data of Sentinel-1 SAR is used, and this data has two polarization modes, namely VV and VH. After image preprocessing, the data is resampled to a regular grid of 10 m×10 m.

[0043] The monitoring method includes the following steps:

[0044] Step (a) Preprocess the Sentinel-1 SAR image, including radiometric calibration, geometric correction, and filtering. Then normalize the preprocessed SAR image to a range of 0-255 to prepare for the subsequent false color image synthesis.

[0045] The processed Sentinel-1 SAR VV polarization and VH polarization data are color combined to form a false color image. In the false color image, the VV polarization data corresponds to the blue channel (b: VV), the VH polarization data corresponds to the green channel (g: VH), and the red channel is replaced by zero value (r: 0). Subsequently, multi-scale segmentation is performed based on the false color image to obtain the segmentation boundary line. During the segmentation process, DEM data is introduced for mask processing. The DEM data used is SRTM30 m data, which needs to be interpolated in advance to keep it consistent with the resolution of the SAR image.

[0046] like Figure 2As shown in Figure 1, the multi-scale segmentation results under different polarization conditions are shown. Among them, (e), (f), and (g) are the image segmentation results under VV, VH, and false color images, respectively. The color axis represents the grayscale value.

[0047] Step (b) Use a Gaussian mixture model to fit the SAR image histogram under different polarizations. The Gaussian mixture model models the mixed distribution of water and land pixels in the SAR image as follows: Where x is the gray value of the SAR image pixel, It represents the probability value when the gray value is x. is the Gaussian distribution probability density function of the water body, is the Gaussian distribution probability density function of non-water bodies, represents the probability that x belongs to the water distribution, and represents the probability that x belongs to the non-water distribution, and .

[0048] The maximum expectation algorithm is used to estimate the parameters in the Gaussian mixture model. First, the preprocessed SAR image is segmented using the OTSU threshold method to obtain water and non-water bodies. Then, based on this segmentation result, the mean and variance of the water and non-water bodies are calculated and used as the initial parameters of the Gaussian mixture model. Subsequently, the maximum expectation algorithm is applied to iteratively optimize the Gaussian mixture model to obtain the optimal model parameters.

[0049] like Figure 3 As shown in the figure, the Gaussian mixture model fitting results of the SAR image histogram under different polarizations. Among them, (h), (i), and (j) are the Gaussian mixture model fitting results under VV, VH, and VV+VH polarizations, respectively. The blue and green solid lines in the figure represent the fitting component probability density of the water and non-water regions, respectively, the red dotted line is the GMM fitting curve, and the blue and green dotted lines represent the fitting component probability density of the water and non-water regions when the fitting component probability density is 10 - 4 , the x-axis vertical line corresponds to the gray value, and the gray value here is also given in the corresponding color.

[0050] Step (c) Calculate the probability P of water distribution in the Gaussian mixture model under different polarizations VV, VH and VV+VH 水 For 10 -4 The maximum gray value T H , and when the probability of non-water distribution P 非水 For 10 -4 The minimum gray value T L Then calculate the difference between these two thresholds T = T H -T L, so as to determine the grayscale range of the overlapping part. Then, the overlapping area sizes of the Gaussian mixture models corresponding to the SAR images under different polarizations are compared to select the best SAR image. By comparison, it can be seen that the overlapping area of ​​the Gaussian mixture model under VV+VH polarization is the smallest, so in this example, the best SAR image is the SAR image with VV+VH polarization.

[0051] The probability of each gray value belonging to a water body is calculated based on the fitted Gaussian mixture model.

[0052] in, It indicates the probability of being a water body when the gray value is x. are the optimal Gaussian mixture model parameters, where is the optimal probability that the gray value belongs to the water distribution in the Gaussian mixture model; are the optimal mean and variance of water data in the Gaussian mixture model, respectively; is the optimal probability that the gray value belongs to the non-water distribution in the Gaussian mixture model; are the optimal mean and variance of non-water data in the Gaussian mixture model, respectively.

[0053] By evaluating the false positive rates of water bodies and non-water bodies at different thresholds, the optimal probability threshold is determined, and the optimal SAR images are divided into three categories based on the threshold. After evaluation, it is found that, as shown in Table 1, when the water body probability threshold exceeds 80%, the false positive rates of water bodies and non-water bodies are basically below 1%. Therefore, in this example, the optimal probability threshold is 80%.

[0054] Table 1 False positive rates of water bodies and non-water bodies at different probability thresholds

[0055]

[0056] The gray-level co-occurrence matrix is ​​used to calculate the texture feature information and the angular second moment.

[0057]

[0058] in, Ng The gray level of the image after quantization; , which can be understood as normalizing the gray-level co-occurrence matrix ( n ( i , j ) is the number of times pixel pairs appear in the gray-level co-occurrence matrix).

[0059] The gray values ​​x1 and x2 when the probability thresholds are 80% and 20% are used to define the parameters of the fuzzy membership function Z of the fuzzy system corresponding to the gray values ​​in the SAR image.

[0060] Using WorldCover data, the permanent water bodies in the obtained water bodies are removed to obtain the non-permanent water bodies. Then the mean and standard deviation of the angular second-order moment corresponding to this part of the water body are calculated to obtain the parameters of the fuzzy membership function S of the fuzzy system corresponding to the angular second-order moment.

[0061] The gray value fuzzy membership function Z is used to calculate the membership F1 of each object's gray value belonging to the water body, and the angular second-order moment fuzzy membership function S is used to calculate the membership F2 of each object's angular second-order moment belonging to the water body. Then calculate F = 1 / 2 * F1 + 1 / 2 * F2, and use F to determine the classification of the uncertain area. When F ≥ 1 / 2, it is determined to be a water body, otherwise it is a non-water body, so as to achieve water body extraction in SAR images.

[0062] Step (d) extracting the water bodies in the SAR images before and after the flood disaster, that is, the water bodies in the flood SAR image and the water bodies in the reference SAR, and comparing them to obtain the flood disaster scope. Figure 4 As shown in the figure, it is an example of the extraction result of flood inundation area. Since the resolution of SAR image is 10 m×10 m, if the number of pixels covered by the obtained flood disaster range is N, then N*10*10 is the area of ​​the flood zone. Figure 4 The flooded area in = 26.58 km 2 It is obtained based on this calculation method.

[0063] The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still modify the technical solutions described in the above embodiments, or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, modification, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A high-precision flood inundation area monitoring method based on dual-polarization SAR images, characterized in that: The following steps are involved: (a) Preprocessing the dual-polarization SAR images, combining the preprocessed SAR images to generate a false color image, and performing multi-scale segmentation based on the false color image. In the segmentation process, digital elevation model data is introduced for mask processing; (b) The preprocessed SAR images were preliminarily classified using the OTSU threshold method to obtain the initial parameters of the Gaussian mixture model. The Gaussian mixture model parameters were estimated using the maximum expectation algorithm. The overlapping area sizes of the SAR Gaussian mixture models under different polarizations were compared to select the best SAR images for monitoring flood disasters. (c) Preliminary classification of the best SAR image based on the probability threshold, water body probability P( x )The formula is: (2) Among them, P( x ) represents the probability that the gray value is water when it is x. are the optimal Gaussian mixture model parameters; where, is the optimal probability that the gray value belongs to the water distribution in the Gaussian mixture model; are the optimal mean and variance of water data in the Gaussian mixture model, respectively; is the optimal probability that the gray value belongs to the non-water distribution in the Gaussian mixture model; are the optimal mean and variance of non-water data in the Gaussian mixture model, respectively; The optimal probability threshold is determined by evaluating the false positive rate of water bodies and non-water bodies under different thresholds. , and based on the threshold, the best SAR image is divided into water body, uncertain area and non-water body; Combining gray value fuzzy system and texture feature fuzzy system to classify uncertain areas and extract water body information; (d) By comparing the water body information of SAR images before and after the flood disaster, the flood inundation area is obtained.

2. The high-precision flood inundation area monitoring method based on dual-polarization SAR images according to claim 1 is characterized in that: In the step (a), the multi-scale segmentation is performed based on the pixel features including gray value, texture and color of the false color image synthesized from the dual-polarization SAR image.

3. The high-precision flood inundation area monitoring method based on dual-polarization SAR images according to claim 1 is characterized in that: In the step (b), a Gaussian mixture model is used to model the mixed distribution of water and land pixels in SAR images under different polarizations, and the formula is as follows: (1) Where x is the gray value of the SAR image pixel, p ( x ) represents the probability value when the gray value is x, is the Gaussian distribution probability density function of the water body, is the Gaussian distribution probability density function of non-water bodies, represents the probability that x belongs to the water distribution, and represents the probability that x belongs to the non-water distribution, and .

4. The high-precision flood inundation area monitoring method based on dual-polarization SAR images according to claim 1 is characterized in that: In the step (b), the Gaussian mixture model parameters are estimated using the maximum expectation algorithm. First, the SAR image is preliminarily classified using the OTSU threshold method to obtain water bodies and non-water bodies. Then, based on the preliminary classification results, the means and variances of the water bodies and non-water bodies are calculated as the initial parameters of the Gaussian mixture model. Finally, the Gaussian mixture model is iteratively optimized using the maximum expectation algorithm to obtain the optimal model parameters.

5. The high-precision flood inundation area monitoring method based on dual-polarization SAR images according to claim 1 is characterized in that: In step (b), the Gaussian mixture model overlap region is extracted using a distribution probability threshold To extract, follow these steps: First, calculate the probability P of water distribution under different polarizations 水 equal The maximum gray value T H , and when the probability P in the non-water distribution 非水 equal The minimum gray value T L ; in, , x is the gray value of the SAR image pixel, is the optimal probability that the gray value belongs to the water distribution in the Gaussian mixture model, are the optimal mean and variance of water data in the Gaussian mixture model, respectively; , x is the gray value of the SAR image pixel, is the optimal probability that the gray value belongs to the non-water distribution in the Gaussian mixture model; are the optimal mean and variance of non-water data in the Gaussian mixture model, respectively; Then, calculate the difference between these two thresholds T = T H -T L , determine the grayscale range of the overlapping part; Finally, the sizes of overlapping areas of Gaussian mixture models corresponding to SAR images under different polarizations are compared to select the best SAR image.

6. The high-precision flood inundation area monitoring method based on dual-polarization SAR images according to claim 1 is characterized in that: In step (c), the fuzzy membership function corresponding to the gray value fuzzy system is a Z function, whose parameters are determined by the optimal probability threshold The fuzzy membership function corresponding to the texture feature fuzzy system is the S function, and its parameters are obtained by the mean and standard deviation of the angular second-order moment corresponding to the non-permanent water body.

7. The high-precision flood inundation area monitoring method based on dual-polarization SAR images according to claim 6 is characterized in that: The fuzzy membership function of the texture feature fuzzy system corresponds to the S function, and its parameters are obtained by the following steps: The permanent water bodies in the water bodies obtained in step (c) are removed using WorldCover data; then the angular second-order moment of the water bodies after the permanent water bodies are removed is calculated, and the corresponding mean and standard deviation are calculated to obtain the parameters of the S fuzzy membership function.

8. The high-precision flood inundation area monitoring method based on dual-polarization SAR images according to claim 1 is characterized in that: In the step (d), the water body extraction result of the reference SAR image before the flood disaster is compared with the water body extraction result of the flood SAR image after the disaster to obtain the spatial distribution of the flooded area.

Images