Water body extraction method based on combined water body index frequency

Abstract

The application discloses a water body extraction method based on combined water body index frequency, comprising the following steps: S1, acquiring remote sensing image data and pre-processing the remote sensing image data; S2, constructing a water body index and constructing an instantaneous water body extraction decision tree model according to the water body index; S3, performing instantaneous water body pixel recognition and extraction according to the instantaneous water body extraction decision tree model, discriminating the water body and preliminarily determining the average water body; S4, extracting the water edge line by using a Canny edge detection method, and determining the average water body extraction result according to the output water body pixel value, the edge detection and the boundary elimination result. The application can realize average water body extraction based on dense long-time sequence images, improve the water body extraction precision, improve the effectiveness of water body extraction in the building shadow and terrain shadow noise environment, resist the interference of the cirrus cloud noise in the single image, and make the large-scale and long-time sequence water body extraction more universal.

Description

Technical Field

[0001] This invention relates to remote sensing extraction methods, specifically a water body extraction method based on combined water body index frequencies. Background Technology

[0002] Water resources, as one of the most important resources on Earth, are the essential material foundation for the survival and development of all human beings and living organisms. Extracting water body information is of paramount importance for understanding the ecological environment, water resource utilization, and protection of a study area. Extracting water bodies from remote sensing imagery facilitates a better understanding of existing water resource conditions, aids in more rational water resource planning and management, improves water resource utilization efficiency, and has a significant impact on human life and social activities.

[0003] Water body extraction is mainly achieved through methods such as thresholding, filtering, elevation and gray-level co-occurrence matrices, and modeling by combining optical and radar imagery. In recent years, the continuous launch of numerous optical satellites, such as the Landsat series, Sentinel-2, and Gaofen series, has provided abundant high-resolution, freely available satellite multispectral remote sensing images, making thresholding for water body extraction from spectral images a common approach.

[0004] The single-band thresholding method extracts water bodies based on the specific reflection patterns of near-infrared and short-wave infrared bands. It is simple, easy to implement, and convenient to operate. However, the applicable environment of the single-band thresholding method is limited. In actual study areas, the types of obstructions on both sides of the watershed are complex, and the accuracy of the water extraction results using the single-band thresholding method is not high.

[0005] The interspectral relationship method leverages the varying reflectance of different spectral bands in water bodies and against a water background to measure and analyze the reflectance of water bodies and surrounding topography in the study area. This analysis yields quantitative relationships between band combinations that distinguish water bodies, thus enhancing the extraction of water-land contrast. However, using the interspectral relationship method requires targeted spectral measurements and analyses of the water bodies and relevant typical land features in the study area. Constructing interspectral relationships is also challenging, lacks universality, and is time-consuming and labor-intensive, making it unsuitable for large-scale water body extraction and analysis.

[0006] The water index method extracts water bodies by combining algorithms such as addition, subtraction, and multiplication on spectral bands to highlight the differences between land and water. Different construction methods all involve performing operations on various spectral bands to emphasize water body information and suppress non-water body information, thus acquiring water body data. The water index method utilizes rich spectral information, forming a more universal approach, and has therefore become the mainstream method for water body extraction.

[0007] Current water body extraction studies can use a single water body index to extract local water bodies relatively accurately. However, since water bodies in a region are diverse, using only one water body index for whole-region water body extraction has low accuracy. In addition, current water body extraction mostly relies on single-scene images from a certain period. The instantaneous water surface images obtained cannot reflect the average state of the water body, nor are they suitable for reflecting regional climate and environmental changes. Summary of the Invention

[0008] Purpose of the invention: In order to overcome the shortcomings of the existing technology, the purpose of this invention is to provide a water body extraction method based on the combined water body index frequency, which is suitable for determining the average water body in a dense long-term extraction and improving the extraction accuracy.

[0009] Technical solution: The present invention provides a water body extraction method based on combined water body index frequencies, comprising the following steps:

[0010] S1, acquire remote sensing image data and preprocess the remote sensing image data;

[0011] S2, construct the water body index, and construct the instantaneous water body extraction decision tree model based on the water body index;

[0012] S3. Based on the instantaneous water body extraction decision tree model, instantaneous water body pixel identification and extraction are performed to identify the water body and preliminarily determine the average water body.

[0013] S4. The water boundary line is extracted using the Canny edge detection method. The average water extraction result is determined based on the output water pixel value, edge detection and boundary removal results.

[0014] Furthermore, in step S1, the remote sensing image data is dense, long-term, high-resolution remote sensing image data, and the specific preprocessing steps are as follows:

[0015] S11, Filter remote sensing image data to obtain transit image data with cloud cover less than 20%;

[0016] S12, convert the dimensionless DN values ​​in the selected transit image data into atmospheric top reflectance, convert the atmospheric top reflectance in the selected transit image data into surface reflectance, and crop the study area image based on the selected transit image.

[0017] Furthermore, in step S2, the specific steps for extracting the decision tree model are as follows:

[0018] S21. Select and construct five water body indices: NDWI, MNDWI, RNDWI, AWEInsh, AWEIsh, and EVI, and one vegetation index.

[0019] S22. Use the Otsu thresholding algorithm to select the optimal thresholds for various water body indices;

[0020] S23. Construct a decision tree model for instantaneous water body extraction based on the different reflectivities of major ground feature interference factors in the test area. The functional relationship is as follows:

[0021] NDWI.gt(0).and(AWEInsh.gt(-0.375).or(AWEIsh.gt(-0.172)).and(RNDWI.gt(0.004))).orMNDWI.gt(-0.002).and(EVI.lt(0.1)).

[0022] An NDWI index greater than 0 identifies most water pixels. A small number of pixels with an NDWI index less than 0 are misclassified as dark buildings and thus missed. Based on this, the MNDWI index is introduced. When the MNDWI value is greater than or equal to -0.002, building interference is excluded. When the enhanced vegetation index value of a pixel is less than 0.1, vegetation interference is excluded, thus completely extracting water in densely built areas.

[0023] The distinction between silt and water was achieved by combining the AWEInsh and RNDWI indices. When the RNDWI value was greater than 0, a large number of shadow and silt pixels identified as water bodies could be extracted. An AWEInsh threshold of -0.375 was used as the segmentation point to exclude interference from dark buildings and roads in areas less affected by shadows. An AWEIsh threshold of -0.172 was used as the segmentation point to exclude shadow interference. Based on this, RNDWI was used to further filter and extract the water-land interface area.

[0024] In summary, a decision tree model is established to obtain the final complete water body.

[0025] NDWI stands for Normalized Difference Water Index, and its calculation formula is as follows:

[0026]

[0027] MNDWI represents the Improved Normalized Difference Water Index, and its calculation formula is as follows:

[0028]

[0029] RNDWI represents the revised normalized water index, and its calculation formula is as follows:

[0030]

[0031] AWEI comes in two forms: AWEIsh and AWEInsh.

[0032] The formula for calculating AWEIsh is as follows:

[0033] AWEIsh=Blue+2.5×Green-1.5×(NIR+SWIR1)-0.25×SWIR2

[0034] The formula for calculating AWEInsh is:

[0035] AWEInsh=4×(Green-SWIR1)-(0.25×NIR+2.75×SWIR2)

[0036] EVI stands for Vegetation Enhancement Index, and its calculation formula is as follows:

[0037]

[0038] In the formula, Green represents the reflectivity of the green light band; NIR represents the reflectivity of the near-infrared band; SWIR1 represents the reflectivity of the short infrared band one; Red represents the reflectivity of the red light band; Blue represents the reflectivity of the blue light band; and SWIR2 represents the reflectivity of the short infrared band two.

[0039] Further, step S22 includes the following steps:

[0040] S221. Calculate the optimal threshold that can separate water bodies from non-water bodies, so that the intra-class variance of water bodies and non-water bodies is minimized, and perform analysis and statistics on the two classes of pixels after binarization based on the determined threshold.

[0041] S222. Obtain the distinction value between the two types of pixels through statistical analysis results;

[0042] S223. Minimize intra-cluster variation by maximizing inter-cluster variance;

[0043] S224. Returns a single intensity value to obtain the optimal threshold for distinguishing the two types of pixels.

[0044] Furthermore, in step S3, the specific steps for initially determining the average water volume are as follows:

[0045] S31. Statistical analysis of water body identification results;

[0046] S32. Calculate the water body frequency based on the statistical water body identification results;

[0047] S33. Remove low-quality extraction results and use the water frequency map to preliminarily determine the average water volume.

[0048] The formula for calculating the frequency of water bodies is:

[0049]

[0050] In the formula, F Water Represents the frequency of water bodies; ∑N Water Represents the number of water pixels on the open surface; ∑N Total ∑N represents the total number of observations within a year; Bad ∑N represents the number of poor-quality observations within a year; Total -∑N Bad This indicates the number of high-quality observations.

[0051] In step S3, the expression for judging the water body is:

[0052]

[0053] In the formula, UNION represents the instantaneous water body extraction decision tree model discrimination method proposed based on the characteristics of different test points in the study area; 0 and 1 values ​​represent non-water bodies and water bodies, respectively; when the pixel value in the study area simultaneously satisfies the enhanced vegetation index EVI being less than 0.1 and meets the UNION characteristics, the pixel can be identified as a water body.

[0054] Furthermore, in step S4, the specific steps for extracting the waterline are as follows:

[0055] S41. Extract water edges that are less affected by background values ​​using the Canny edge detection method;

[0056] S42. Remove the boundaries of artificial water bodies.

[0057] The Canny edge detection method includes the following steps: removing noise from the image using a Gaussian filter; calculating the gradient value and gradient direction; filtering out non-maximum suppression values; and determining edge double thresholds.

[0058] Beneficial effects: Compared with the prior art, the present invention has the following significant features: it can achieve average water body extraction under dense long-term conditions, improve the water body extraction accuracy to 92.75%, enhance the effectiveness of water body extraction under building shadow and terrain shadow noise environment, resist the interference of cirrus cloud noise in single-scene images, make large-scale, long-term water body extraction more universal, and provide technical support for large-scale remote sensing surface water monitoring, and provide technical guarantee for the rational planning and management of water resources and the protection of ecology. Attached Figure Description

[0059] Figure 1 This is a flowchart of the present invention;

[0060] Figure 2 This is a schematic diagram of the instantaneous water body extraction decision tree model of the present invention;

[0061] Figure 3This is a schematic diagram of the instantaneous water body extraction results using the combined water body index of the present invention, where a is test point 1, b is test point 2, and c is test point 3;

[0062] Figure 4 This is a schematic diagram of water body image frequency for the entire Yangtze River Jiangsu section in 2016, based on the present invention.

[0063] Figure 5 This is a schematic diagram of the water body extraction results based on the average water level of this invention;

[0064] Figure 6 This is a schematic diagram of the waterline extraction results of the present invention. Detailed Implementation

[0065] like Figure 1 A water body extraction method based on combined water body index frequencies specifically includes the following steps:

[0066] S1. Acquire remote sensing image data and preprocess the remote sensing image data, including the following steps:

[0067] S11. Acquire dense, long-term, high-resolution remote sensing image data.

[0068] S12. Filter the remote sensing image data to obtain transit image data with cloud cover of less than 20%.

[0069] S13. Perform radiometric calibration, atmospheric correction, and cropping on the selected transit image data.

[0070] S2. Construct a water body index and build a decision tree model for instantaneous water body extraction based on the water body index, including the following steps:

[0071] S21. Select and construct five water body indices: NDWI, MNDWI, RNDWI, AWEInsh, AWEIsh, and EVI, and one vegetation index.

[0072] Specifically, NDWI represents the Normalized Difference Water Index, and its calculation formula is as follows:

[0073]

[0074] In the formula, Green represents the reflectivity in the green light band; NIR represents the reflectivity in the near-infrared band. NDWI can enhance water bodies in satellite images, and its main purpose is to detect and monitor minute changes in water content. The disadvantage of this index is that it is sensitive to building structures and cannot effectively suppress building noise in built-up areas and mountain shadows in mountainous areas, which may lead to an overestimation of water bodies.

[0075] MNDWI represents the Improved Normalized Difference Water Index, and its calculation formula is as follows:

[0076]

[0077] In the formula, Green represents the reflectivity of the green light band; SWIR1 represents the reflectivity of the short infrared band. MNDWI can effectively suppress background noise from buildings, bare land, etc., and has higher accuracy in extracting water bodies in densely built-up urban areas.

[0078] RNDWI represents the revised normalized water index, and its calculation formula is as follows:

[0079]

[0080] In the formula, SWIR1 represents the reflectivity of the short-infrared band; Red represents the reflectivity of the red band. The short-infrared and red bands can eliminate the influence of mountain shadows and are mainly used for extraction in muddy and shallow water bodies.

[0081] Specifically, AWEI stands for Automatic Water Extraction Index, which consistently improves water extraction accuracy under various environmental noise conditions while providing a stable threshold. AWEI comes in two forms: AWEIsh and AWEInsh.

[0082] The formula for calculating AWEIsh is as follows:

[0083] AWEIsh=Blue+2.5×Green-1.5×(NIR+SWIR1)-0.25×SWIR2

[0084] The formula for calculating AWEInsh is:

[0085] AWEInsh=4×(Green-SWIR1)-(0.25×NIR+2.75×SWIR2)

[0086] In the formula, Green represents the reflectivity of the green light band; BLUE represents the reflectivity of the blue light band; NIR represents the reflectivity of the near-infrared band; and SWIR1 and SWIR2 represent the reflectivity of short infrared band one and short infrared band two, respectively.

[0087] Specifically, AWEIsh is primarily designed to remove shadow pixels and makes it easier to distinguish water boundaries; AWEInsh is designed for areas with urban backgrounds and is suitable for shadowy areas.

[0088] Specifically, EVI stands for Vegetation Enhancement Index, and its calculation formula is as follows:

[0089]

[0090] In the formula, NIR represents the reflectance in the near-infrared band; Red represents the reflectance in the red band; and Blue represents the reflectance in the blue band. EVI not only possesses the advantages of the Normalized Difference Vegetation Index (NDVI), but also improves upon its limitations such as saturation in high-vegetation areas, incomplete correction for atmospheric influences, and soil background. EVI can enhance vegetation sensitivity, reduce soil background and atmospheric influences, and has higher sensitivity and superiority in monitoring vegetation changes. It is widely used in studies such as grassland degradation monitoring and quantitative analysis of grassland resources.

[0091] S22. Select the optimal thresholds for various water body indices using the Otsu thresholding method, including the following steps:

[0092] S221. Traverse all possible thresholds and analyze and statistically analyze the two types of pixels after threshold segmentation.

[0093] S222. Obtain the distinction value between the two types of pixels through statistical analysis results.

[0094] S223. Minimize intra-cluster variation by maximizing inter-cluster variance.

[0095] S224. Returns a single intensity value to obtain the optimal threshold for distinguishing the two types of pixels.

[0096] S23. Construct a decision tree model for instantaneous water body extraction based on the different reflectivities of major ground object interference factors in the test area.

[0097] S3. Based on the instantaneous water body extraction decision tree model, instantaneous water body pixel identification and extraction are performed, and the water body index frequency is calculated. This includes the following steps:

[0098] S31. Based on the instantaneous water body extraction decision tree model, instantaneous water body pixel identification and extraction are performed, and water bodies are judged.

[0099] The expression for instantaneous water body pixel identification and extraction based on the instantaneous water body extraction decision tree model, and for classifying water bodies, is as follows:

[0100]

[0101] In the formula, UNION represents the instantaneous water body extraction decision tree model discrimination method proposed based on the characteristics of different test points in the study area; 0 and 1 values ​​are used to represent non-water bodies and water bodies, respectively; when the pixel value in the study area simultaneously satisfies the Enhanced Vegetation Index (EVI) being less than 0.1 and meets the UNION characteristics, the pixel can be identified as a water body.

[0102] S32. Statistical analysis of the water body identification results.

[0103] S33. Calculate the water body frequency based on the statistical results of water body identification. The formula for calculating the water body frequency is:

[0104]

[0105] In the formula, F Water Represents the frequency of water bodies; ∑N Water Represents the number of water pixels on the open surface; ∑N Total ∑N represents the total number of observations within a year; Bad ∑N represents the number of poor-quality observations within a year; Total -∑N Bad This indicates the number of high-quality observations.

[0106] S34. Remove low-quality extraction results and use the water frequency map to preliminarily determine the average water volume.

[0107] S4. Extract the water boundary line using the Canny edge detection method and determine the average water body extraction result, including the following steps:

[0108] S41. Extract water edges that are less affected by background values ​​using the Canny edge detection method. Specifically, this includes: removing noise from the image using a Gaussian filter; calculating the gradient value and gradient direction; filtering out non-maximum suppression values; and determining edge double thresholds.

[0109] S42. Remove small artificial water boundaries.

[0110] S43. Determine the average water extraction result based on the output water pixel values, edge detection, and boundary removal results.

[0111] The method of this embodiment is applied to the Jiangsu section of the Yangtze River, which has a complex surface environment, as the study area. The Yangtze River shoreline extraction based on the high-precision water body extraction algorithm includes the following steps:

[0112] Based on the Google Earth Engine cloud platform, this paper proposes a high-precision water body extraction algorithm that is operable in a large-scale environment, using a long-term time-series Sentinel-2 MSI image set combined with the "temporal features" of pixels, to extract the Yangtze River shoreline. Sentinel-2 satellite remote sensing image data of the Jiangsu section of the Yangtze River was acquired as experimental data, and ground truth water body images were used as validation data. The Sentinel-2 satellite data was obtained by using the GEE platform to select an annual image dataset from 2016 to 2021, and all transit images with cloud cover less than 20% during 2016-2021 were selected as experimental data. Ground truth water body images were obtained by visually interpreting 83 Sentinel-2 images from 2016 to 2021. Several images were selected for each month, and the water body vector maps and water body boundary vector maps were manually drawn through visual interpretation of the Sentinel-2 images. The water body images at the average water level were obtained using an average value calculation tool as validation data for water body extraction. The images undergo preprocessing such as radiometric calibration, atmospheric correction, and cropping.

[0113] The water bodies in the study area were classified according to their geographical environment into three types: water bodies in densely built-up and shaded water areas, water bodies with abundant vegetation, and undeveloped silty water bodies. Five water body indices (NDWI, MNDWI, RNDWI, AWEInsh, AWEIsh, and EVI) and one vegetation index were constructed. The Otsu thresholding method was used to select the optimal thresholds for each water body index, and segmented studies were conducted within different threshold ranges. Figure 2 A decision tree model for instantaneous water body extraction was constructed, and the decision tree was built based on the different reflectivities of major land cover interference factors in the experimental area, such as shadows, buildings, dark roads, and vegetation. The functional relationship is as follows:

[0114] NDWI.gt(0).and(AWEInsh.gt(-0.375).or(AWEIsh.gt(-0.172)).and(RNDWI.gt(0.004))).or MNDWI.gt(-0.002).and(EVI.lt(0.1)).

[0115] An NDWI index greater than 0 identifies most water pixels. A small number of pixels with an NDWI index less than 0 are misclassified as dark buildings and thus missed. Based on this, the MNDWI index is introduced. When the MNDWI value is greater than or equal to -0.002, building interference is excluded. When the enhanced vegetation index value of a pixel is less than 0.1, vegetation interference is excluded, thus completely extracting water in densely built areas.

[0116] The distinction between silt and water was achieved by combining the AWEInsh and RNDWI indices. When the RNDWI value was greater than 0, a large number of shadow and silt pixels identified as water bodies could be extracted. An AWEInsh threshold of -0.375 was used as the segmentation point to exclude interference from dark buildings and roads in areas less affected by shadows. An AWEIsh threshold of -0.172 was used as the segmentation point to exclude shadow interference. Based on this, RNDWI was used to further filter and extract the water-land interface area.

[0117] In summary, a decision tree model is established to obtain the final complete water body. Here, "water" represents a water body, and "no water" represents a non-water body.

[0118] like Figure 3 The instantaneous water body pixel extraction and identification is performed according to the constructed decision tree extraction rules. Test point a is test point 1, which contains a large number of buildings and shadow interference along the shore, and is an artificial shoreline with concentrated buildings; test point b is test point 2, which is a silty shoreline with a large amount of beach and silt; test point c includes tributaries of the Yangtze River and vegetated land, and is a natural shoreline with rich vegetation.

[0119] The number of water body pixels in dense long-term imagery can be counted using the following formula:

[0120]

[0121] Calculate the water body frequency using the following formula:

[0122]

[0123] Pixels with a mask value greater than 1 are identified as low-quality observations and are either masked or excluded from the study.

[0124] The frequency of water body images for the entire Jiangsu section of the Yangtze River in 2016 is as follows: Figure 4 As shown in the figure. A combined water body index frequency method was used to conduct water body extraction experiments at three test points. It was ultimately determined that the water body extraction result was closest to the water surface image under the average water level when the water body frequency was 0.63, with an extraction accuracy of 92.75%. The water body image extraction results are shown in the figure. Figure 5 As shown.

[0125] Based on the average water quality of the Jiangsu section of the Yangtze River from 2016 to 2021, Canny edge detection was used to extract water edges that were less affected by background values. A Python script was used to remove small fishpond boundaries; the specific code is shown below.

[0126]

[0127]

[0128] like Figure 6 By filtering by shape, the waterline extraction results are obtained.

[0129] In summary, by utilizing the above-mentioned technical solution of this invention, the optimal threshold for water body indices can be determined using the Otsu thresholding method. A decision tree model for instantaneous water body extraction is constructed using five water body indices and one vegetation index. Instantaneous water body pixel identification is performed based on this decision tree model. Water body frequency is calculated by statistically analyzing the water body discrimination results of dense long-term imagery. Water edges are extracted using the Canny edge detection method, ultimately determining the average water body under dense long-term conditions. This solves the problems of current water body extraction methods, which mostly rely on single-scene images from a specific period, resulting in low extraction accuracy and limited research scope. This invention enables the extraction of average water bodies under dense long-term conditions, improves extraction accuracy, enhances the effectiveness of water body extraction in environments with building shadows and terrain shadow noise, resists interference from cirrus noise in single-scene images, and makes large-scale, long-term water body extraction more universal. Furthermore, it provides technical support for large-scale remote sensing surface water monitoring and offers technical guarantees for the rational planning and management of water resources and ecological protection.

Claims

1. A method for water body extraction based on combined water body index frequencies, characterized in that, Includes the following steps: S1, acquire remote sensing image data and preprocess the remote sensing image data; S2, construct the water body index, and construct the instantaneous water body extraction decision tree model based on the water body index; S3. Based on the instantaneous water body extraction decision tree model, instantaneous water body pixel identification and extraction are performed to identify the water body and preliminarily determine the average water body. S4. The water boundary line is extracted using the Canny edge detection method. The average water extraction result is determined based on the output water pixel value, edge detection and boundary removal results. In step S2, the specific steps for extracting the decision tree model are as follows: S21. Select and construct five water body indices: NDWI, MNDWI, RNDWI, AWEInsh, AWEIsh, and EVI, and one vegetation index. S22. Use the Otsu algorithm to select the optimal threshold for various water body indices; S23. Construct a decision tree model for instantaneous water body extraction based on the different reflectivities of major ground feature interference factors in the test area. The functional relationship is as follows: (NDWI.gt(0).and(AWEInsh.gte(-0.375).or(AWEIsh.gte(-0.172).and(RNDWI.gte(0.004)))).or(NDWI.lte(0).and(MNDWI.gte(-0.002)).and(EVI.lt(0.1))); In step S3, the specific steps for initially determining the average water volume are as follows: S31. Statistical analysis of water body identification results; S32. Calculate the water body frequency based on the statistical water body identification results; S33. Remove low-quality extraction results and use the water body frequency map to preliminarily determine the average water body. The formula for calculating the frequency of the water body is: In the formula, Indicates the frequency of water bodies; This indicates the number of pixels representing water bodies on the open surface of the earth. This represents the total number of observations within a year; This indicates the number of poor-quality observations within a year; This indicates the number of high-quality observations; In step S4, the specific steps for extracting the waterline are as follows: S41. Extract water edges that are less affected by background values ​​using the Canny edge detection method; S42. Remove the boundaries of artificial water areas; The Canny edge detection method includes the following steps: removing noise from the image using a Gaussian filter; calculating the gradient value and gradient direction; and filtering out non-maximum suppression values. Determine the edge double threshold.

2. The water body extraction method based on combined water body index frequency according to claim 1, characterized in that: In step S1, the remote sensing image data is dense, long-term, high-resolution remote sensing image data. The specific preprocessing steps are as follows: S11, Filter remote sensing image data to obtain transit image data with cloud cover less than 20%; S12, convert the dimensionless DN values ​​in the selected transit image data into top atmospheric reflectance, convert the top atmospheric reflectance in the selected transit image data into surface reflectance, and crop the study area image based on the selected transit image.

3. The water body extraction method based on combined water body index frequencies according to claim 1, characterized in that: The NDWI stands for Normalized Difference Water Index, and its calculation formula is as follows: The MNDWI represents the Improved Normalized Difference Water Index, and its calculation formula is as follows: The RNDWI stands for Revised Normalized Difference Water Index, and its calculation formula is as follows: The AWEI comes in two forms: AWEIsh and AWEInsh. The formula for calculating AWEIsh is as follows: The formula for calculating AWEInsh is: The EVI stands for Vegetation Enhancement Index, and its calculation formula is as follows: In the formula, Green represents the reflectivity of the green light band; NIR represents the reflectivity of the near-infrared band; SWIR1 represents the reflectivity of the short infrared band one; Red represents the reflectivity of the red light band; Blue represents the reflectivity of the blue light band; and SWIR2 represents the reflectivity of the short infrared band two.

4. The water body extraction method based on combined water body index frequencies according to claim 1, characterized in that: S22 includes the following steps: S221. Calculate the optimal threshold that can separate water bodies from non-water bodies, so that the intra-class variance of water bodies and non-water bodies is minimized, and perform analysis and statistics on the two classes of pixels after binarization based on the determined threshold. S222. Obtain the distinction value between the two types of pixels through statistical analysis results; S223. Minimize intra-cluster variation by maximizing inter-cluster variance; S224. Returns a single intensity value to obtain the optimal threshold for distinguishing the two types of pixels.

5. The water body extraction method based on combined water body index frequency according to claim 1, characterized in that: In step S3, the expression for identifying the water body is: In the formula, UNION represents the instantaneous water body extraction decision tree model discrimination method proposed based on the characteristics of different test points in the study area; 0 and 1 values ​​represent non-water bodies and water bodies, respectively; when the pixel value in the study area simultaneously satisfies the enhanced vegetation index EVI being less than 0.1 and meets the UNION characteristics, the pixel can be identified as a water body.

Images