CN111259876B - A method and system for extracting water body information from radar data based on surface water body products - Google Patents

Abstract

The invention provides a radar data water body information extraction method and system based on a land surface water body product, wherein the method comprises the steps of obtaining satellite data and the following steps: preprocessing the satellite data to generate backscattering data and calculating a derivative coefficient of the backscattering data; creating a training sample dataset using global water product data, the backscatter data and the derivative coefficients; extracting water body information of the training sample data based on a random forest model to obtain an initial water body extraction result; and carrying out post-treatment on the initial water body extraction result to obtain a final water body product. The method uses the JRC global water product to automatically generate a training data set, screens the data set, then uses a machine learning classification method to automatically extract water information, performs post-processing on an extracted result, and eliminates images of small image spots and mountain shadows, thereby obtaining an accurate water information product.

Description

A method and system for extracting water body information from radar data based on surface water body products

technical field

The invention relates to the technical field of hydraulic simulation, in particular to a method and system for extracting water body information from radar data based on land surface water body products.

Background technique

It is of great significance to quickly obtain the distribution of water body information for flood assessment, ecological environment monitoring and water resources survey. Remote sensing technology is an important means to obtain water body information quickly and accurately. Due to its short wavelength, optical remote sensing is easily affected by the weather, so its application is limited. In microwave remote sensing, spaceborne Synthetic Aperture Radar (SAR) has the characteristics of all-day, all-weather, etc. It is not affected by clouds, rain, and fog, and can also image at night, making it a useful tool for flood monitoring and lake monitoring. A powerful tool for dynamic monitoring. The Sentinel-1A series satellites launched by ESA in April 2014 have a star orbit period of 12 days for a single satellite and 6 days for two satellites, a maximum width of 400km, a maximum resolution of 5m, and a single polarization. (HH/VV) and dual polarization (HH+HV/VV+VH) imaging, and the data is free and can be downloaded by users worldwide for easy access.

At present, there are two main methods for SAR water body information extraction: one is based on threshold segmentation, and the other is based on classification. The most widely used threshold methods include single threshold method, multi-threshold method, bimodal method and global threshold method. However, in the extraction of large-scale water body information, the proportion of water body is relatively low, and the threshold value is relatively difficult to obtain. Moreover, due to the image of the radar imaging mechanism itself, there will be shadows of mountains, terrain distortion, and noise on the surface of the water body caused by wind and waves. The extraction effect of threshold method is often not good. Classification-based identification of large water bodies has been proven to be a successful algorithm. However, classification-based algorithms are often supervised methods and require sufficient training samples, which are traditionally selected manually, which greatly hinders the automation efficiency of water body information extraction. At present, existing datasets, such as MODIS and SRTM-derived water body masks, have been used to train models for automatic extraction of water body information. This method has great potential. However, due to the low spatial resolution of MODIS and SRTM, the accuracy and effect of water extraction are affected. The global surface water products (with a resolution of 30m and a variety of products) produced by the European Union Joint Research Center JRC (Joint Research Centre) based on Landsat data from 1984 to 2018 have an overall accuracy of 99.6%, which can be used as a classification method for fine water body information Extracted training samples.

Wenli Huang et al. proposed an automatic extraction method of water body information based on radar satellites on Remote Sensing in 2018. The method is as follows: (1) Based on the SRTM water body data set SWBD and the composite dynamic surface water range cDSWE to generate water body labels, using radar The backscattering coefficient, derivative index and local incident angle are used as features, and the water body labels form the training samples; (2) Part of the training samples are randomly selected from the sample data according to the ratio of water body samples and non-water body samples; (3) Using randomly sampled The training samples train the random forest, then apply the trained model to all pixels to generate a water body probability map, which is classified using the high probability water body, medium probability water body, low probability water body, and non-water body labels. This method has a fast extraction speed and relatively high accuracy, but it does not consider the significant wrong samples in the training samples due to seasons and time, which affects the extraction accuracy; Partially mountain shaded.

SUMMARY OF THE INVENTION

In order to solve the above technical problems, a method and system for extracting water body information from radar data based on land surface water body products proposed by the present invention, JRC's global water body products are used to automatically generate training data sets, and the data sets are screened, and then The classification method of machine learning is used to automatically extract the water body information, and the extracted results are post-processed, and the images of small patches and mountain shadows are eliminated, so as to obtain accurate water body information products.

The first object of the present invention is to provide a method for extracting water body information from radar data based on surface water products, including acquiring satellite data, and further comprising the following steps:

Step 1: Preprocess the satellite data to generate backscatter data and calculate the derivative coefficient of the backscatter data;

Step 2: Create a training sample dataset using the global water product data, the backscatter data and the derivative coefficients;

Step 3: Extract the water body information from the training sample data based on the random forest model to obtain the initial water body extraction result;

Step 4: post-processing the initial water body extraction result to obtain a final water body product.

Preferably, the preprocessing is to perform orbit correction, radiation correction, speckle filtering, multi-view, terrain correction and db conversion operations on the intensity of the first-level ground distance product after the basic processing of the satellite data, and finally generate a backward Scatter data VH and VV .

。Preferably in any of the above solutions, the derivative coefficient includes the polarization ratio VHrVV , and the calculation formula is

.

。Preferably in any of the above solutions, the derivative coefficient further includes a normalized deviation polarization index NDPI , and the calculation formula is:

.

。Preferably in any of the above solutions, the derivative coefficient also includes a normalized VH index NVHI , and the calculation formula is

.

。Preferably in any of the above solutions, the derivative coefficient further includes a normalized VV index NVVI , and the calculation formula is

.

Preferably in any of the above solutions, the step 2 includes sorting the water body samples and the non-water body samples according to the VV value respectively, and selecting a value with a quantile of N1 % in the water body sample as the maximum threshold value, and deleting Samples larger than the maximum threshold; the non-water samples with a quantile of N2 % are used as the minimum threshold, and samples smaller than the minimum threshold are deleted.

Preferably in any of the above-mentioned schemes, the step 3 includes the following sub-steps:

Step 31: randomly select N3 % water body samples in the training data set, when the number of water body samples is less than the number threshold, select all water body samples;

Step 32: Use the cross-validation grid GridSearch to determine the optimal parameters of the random forest classifier to generate a training model;

Step 33: apply the training model to all pixels of the entire image to obtain the probability that each pixel is a body of water;

Step 34: Obtain the water body distribution according to the threshold value, and obtain the initial water body extraction result.

Preferably in any of the above-mentioned schemes, the step 4 includes the following sub-steps:

Step 41: use the slope map calculated by the digital elevation model DEM of srtm30;

Step 42: Assign the extraction result whose slope corresponding to the initial water body extraction result is greater than the slope threshold value as a non-water body;

Step 43: Delete the water body smaller than the area threshold by the method of seed point diffusion;

Step 44: Replenish the edge part of the complete water body through the regional growth algorithm;

Step 45: Generate the final water body product.

The second object of the present invention is to provide a radar data water body information extraction system based on land surface water body products, including a data acquisition module for acquiring satellite data, and also including the following modules:

Preprocessing module: used to preprocess the satellite data, generate backscatter data and calculate the derivative coefficient of the backscatter data;

Sample training module: used to create a training sample data set using global water product data, the backscatter data and the derivative coefficients;

Information extraction module: used to extract water body information from the training sample data based on the random forest model, and obtain the initial water body extraction results;

Post-processing module: used for post-processing the initial water body extraction result to obtain the final water body product;

Each module in the system extracts water body information from radar data according to the method described in the first objective.

Preferably, the preprocessing is to perform orbit correction, radiation correction, speckle filtering, multi-view, terrain correction and db conversion operations on the intensity of the first-level ground distance product after the basic processing of the satellite data, and finally generate a backward Scatter data VH and VV .

。Preferably in any of the above solutions, the derivative coefficient includes the polarization ratio VHrVV , and the calculation formula is:

.

。Preferably in any of the above solutions, the derivative coefficient further includes a normalized deviation polarization index NDPI , and the calculation formula is:

.

。Preferably in any of the above solutions, the derivative coefficient also includes a normalized VH index NVHI , and the calculation formula is

.

。Preferably in any of the above solutions, the derivative coefficient further includes a normalized VV index NVVI , and the calculation formula is

.

Preferably in any of the above solutions, the sample training module is used to sort water samples and non-water samples according to the VV value respectively, and select a value with a quantile of N1 % in the water samples as the maximum threshold value , delete the samples larger than the maximum threshold; the non-water samples whose quantile is N2 % are used as the minimum threshold, and the samples smaller than the minimum threshold are deleted.

Preferably in any of the above solutions, the information extraction includes the following sub-steps:

Step 31: randomly select N3 % water body samples in the training data set, when the number of water body samples is less than the number threshold, select all water body samples;

Step 32: Use the cross-validation grid GridSearch to determine the optimal parameters of the random forest classifier to generate a training model;

Step 33: apply the training model to all pixels of the entire image to obtain the probability that each pixel is a body of water;

Step 34: Obtain the water body distribution according to the threshold value, and obtain the initial water body extraction result.

Preferably in any of the above schemes, the post-processing includes the following sub-steps:

Step 41: use the slope map calculated by the digital elevation model DEM of srtm30;

Step 42: Assign the extraction result whose slope corresponding to the initial water body extraction result is greater than the slope threshold value as a non-water body;

Step 43: Delete the water body smaller than the area threshold by the method of seed point diffusion;

Step 44: Replenish the edge part of the complete water body through the regional growth algorithm;

Step 45: Generate the final water body product.

The invention provides a method and system for extracting water body information from radar data based on land surface water body products, which can quickly and automatically extract large-scale water body information, and is not affected by mountain shadows, and the extracted water bodies are relatively complete and connected.

Seasonity data refers to a kind of JRC global water body product. It records the number of water bodies that appear in each pixel in the global land area in the past year. The highest value is 12, that is, the water body exists in the pixel for one year; the lowest is 12. 0 means non-water body pixels; 1-11 means seasonal water bodies. The latest water body product is Seasonality2018 based on 2018 Landsat data.

Description of drawings

FIG. 1 is a flowchart of a preferred embodiment of a method for extracting water body information from radar data based on land surface water body products according to the present invention.

FIG. 2 is a flow chart of a preferred embodiment of the method for extracting water body information from radar data based on land surface water body products according to the present invention.

3 is a flowchart of a preferred embodiment of the method for extracting water body information from radar data based on land surface water body products according to the present invention.

FIG. 4 is a block diagram of a preferred embodiment of a system for extracting water body information from radar data based on land surface water body products according to the present invention.

5 is a technical roadmap of another preferred embodiment of the method for extracting water body information from radar data based on land surface water body products according to the present invention.

6 is a schematic diagram of cloud shadow displacement of the embodiment shown in FIG. 3 according to the method for extracting water body information from radar data based on land surface water body products according to the present invention.

FIG. 7 is a schematic diagram of a preferred embodiment of the extraction result of wide-channel water body based on the method for extracting water body information from radar data of land surface water body products according to the present invention. FIG. 8 is a schematic diagram showing the result of water body extraction in narrow streams according to the embodiment shown in FIG. 7 according to the method for extracting water body information from radar data based on land surface water body products.

9 is a pre-disaster satellite data map of a preferred embodiment of a flood event based on the method for extracting water body information from radar data of land surface water products according to the present invention.

FIG. 10 is a diagram showing the result of water body extraction before the disaster according to the embodiment shown in FIG. 9 according to the method for extracting water body information from radar data based on land surface water body products.

FIG. 11 is a satellite data map of the disaster in the embodiment shown in FIG. 9 according to the method for extracting water body information from radar data based on land surface water body products according to the present invention.

FIG. 12 is a diagram showing the results of water body extraction in disasters in the embodiment shown in FIG. 9 according to the method for extracting water body information from radar data based on land surface water body products according to the present invention.

FIG. 13 is a post-disaster satellite data diagram of the embodiment shown in FIG. 9 according to the method for extracting water body information from radar data based on land surface water body products according to the present invention.

FIG. 14 is a result diagram of post-disaster water body extraction according to the embodiment shown in FIG. 9 of the method for extracting water body information from radar data based on land surface water body products according to the present invention.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

Example 1

As shown in FIG. 1 , in a method for extracting water body information from radar data based on land surface water body products, step 100 is performed to obtain satellite data. Satellite data refers to the data returned by Sentinel-1A series satellites.

。归一化偏差极化指数NDPI的计算公式为

归一化VH指数NVHI的计算公式为

。归一化VV指数NVVI的计算公式为

。Step 110 is performed to preprocess the satellite data, generate backscatter data, and calculate a derivative coefficient of the backscatter data. The preprocessing is to perform orbit correction, radiation correction, speckle filtering, multi-view, terrain correction and db conversion operations on the intensity of the first-level ground distance product after basic processing of the satellite data, and finally generate backscatter data VH and VV . The derived coefficients include the polarization ratio VHrVV , the normalized deviation polarization index NDPI , the normalized VH index NVHI and the normalized VV index NVVI . The calculation formula of the polarization ratio VHrVV is

. The calculation formula of the normalized deviation polarization index NDPI is:

The normalized VH index NVHI is calculated as

. The normalized VV index NVVI is calculated as

.

Step 120 is executed to create a training sample data set using the global water body product data, the backscatter data and the derivative coefficients. Sort the water samples and non-water samples according to the VV value respectively, select the value with the quantile N1 % in the water sample as the maximum threshold value, and delete the samples larger than the maximum threshold; the non-water sample has a quantile of N2 The value of % is used as the minimum threshold, and samples smaller than the minimum threshold are deleted. In this embodiment, N1 =85 and N2 =15.

Step 130 is executed to extract water body information from the training sample data based on the random forest model to obtain an initial water body extraction result. As shown in FIG. 2 , step 131 is performed to randomly select N3 % of the water body samples in the training data set. When the number of water body samples is less than the number threshold, all water body samples are selected. In this embodiment, N3 =12.5. Step 132 is executed, using the cross-validation grid GridSearch to determine the optimal parameters of the random forest classifier to generate a training model. Step 133 is performed to apply the training model to all pixels of the entire image to obtain the probability that each pixel is a water body. Step 134 is executed to obtain the water body distribution according to the threshold value, and obtain the initial water body extraction result.

Step 140 is executed to perform post-processing on the initial water body extraction result to obtain a final water body product. As shown in FIG. 3 , step 141 is executed, and the gradient map obtained by the digital elevation model DEM of srtm30 is used. Step 142 is executed to assign the extraction result of the initial water body extraction result whose corresponding gradient is greater than the gradient threshold as a non-water body. Step 143 is performed to delete the water body smaller than the area threshold by the method of seed point diffusion. Step 144 is executed to supplement the edge portion of the complete water body through the region growing algorithm. Step 145 is executed to generate the final water body product.

Embodiment 2

As shown in FIG. 4 , a radar data water body information extraction system based on land surface water body products includes a data acquisition module 200 , a preprocessing module 210 , a sample training module 220 , an information extraction module 230 and a post-processing module 240 .

Data acquisition module 200: used to acquire satellite data.

。归一化偏差极化指数NDPI的计算公式为

。归一化VH指数NVHI的计算公式为

。归一化VV指数NVVI的计算公式为

。Preprocessing module 210: used for preprocessing the satellite data, generating backscatter data and calculating a derivative coefficient of the backscatter data. The preprocessing is to perform orbit correction, radiation correction, speckle filtering, multi-view, terrain correction and conversion operations on the intensity of the first-level ground distance product after basic processing of satellite data. The derived coefficients include the polarization ratio VHrVV , the normalized deviation polarization index NDPI , the normalized VH index NVHI and the normalized VV index NVVI . The calculation formula of the polarization ratio VHrVV is

. The calculation formula of the normalized deviation polarization index NDPI is:

. The normalized VH index NVHI is calculated as

. The normalized VV index NVVI is calculated as

.

Sample training module 220: used to create a training sample data set using global water product data, the backscatter data and the derivative coefficients. Sort water samples and non-water samples according to the VV value respectively, select the value with the quantile N1 % in the water sample as the maximum threshold, and delete the samples larger than the maximum threshold; the non-water sample is N2 The value of % is used as the minimum threshold, and samples smaller than the minimum threshold are deleted.

Information extraction module 230: used for extracting water body information from the training sample data based on the random forest model to obtain an initial water body extraction result. The information extraction includes the following sub-steps: Step 31: randomly select N3 % water body samples in the training data set, when the number of water body samples is less than the number threshold, select all water body samples; Step 32: use the cross-validation grid GridSearch to determine The best parameters of the random forest classifier are used to generate a training model; Step 33: Apply the training model to all pixels of the entire image to obtain the probability that each pixel is a water body; Step 34: Obtain the water body distribution according to the threshold, and obtain the initial water body Extract results.

Post-processing module 240: used for post-processing the initial water body extraction result to obtain a final water body product. The post-processing includes the following sub-steps: Step 41: use the slope map calculated by the digital elevation model DEM of srtm30; Step 42: assign the extraction result whose slope is greater than the slope threshold in the initial water body extraction result as a non-water body; Step 43 : delete the water body smaller than the area threshold by the method of seed point diffusion; step 44: supplement the edge part of the complete water body by the area growth algorithm; step 45: generate the final water body product.

Embodiment 3

The automatic extraction of water body information proposed by the present invention mainly includes four steps:

(1) Preprocess Sentinel-1 SAR data to generate backscatter data and calculate its derivative coefficients;

(2) Use the seasonality2018 product and backscatter data and derivative coefficients to create a training sample dataset;

(3) Extract water body information based on random forest model;

(4) Post-processing: firstly remove significantly misclassified water bodies; then use slope data to remove hill shadows; then remove small connected areas, use regional growth to supplement water bodies, and obtain final water body products; finally, use auxiliary images to evaluate the accuracy of the products.

The technical route is shown in Figure 5.

1. sar data preprocessing and index calculation

Orbit correction, radiation correction, speckle filtering, multi-view, terrain correction, and conversion to dB are performed on the intensity of the high-resolution Sentinel-1 Level-1 ground detection product GRDH (High-resolution Ground Range Detected, 5mx20m). Among them, the speckle filter adopts Refine-Lee filtering, and it is better to convert the image into 20m square pixels in multi-view (10m has more speckle noise, and 30m has lower resolution); using SRTM 1 arc second (about 30m) digital Elevation model for terrain correction.

At the same time, the correlation index is calculated and added to the classification as characteristic data. These include the polarization ratio VHrVV, the normalized deviation polarization index NDPI, the normalized VH index NVHI, and the normalized VV index NVVI. The index calculation is shown in Table 1.

Table 1 List of polarization indices

2. Creation of water body training samples

The sample data comes from the seasonality data of JRC Global Surface Water. This data records the number of months that each water body pixel appears in the global inland area in the last year. The highest value is 12, that is, the water body has always existed in one year; the lowest value is 0, that is, non-water pixels; 1-11 are seasons body of water. The latest product is seasonality2018. Due to the problems of season and accuracy, there will be fewer misclassified samples in water samples and non-water samples. Therefore, this paper combines the characteristics of water in radar images to first sort water samples and non-water samples by VV value, and then select The value in the 85% quantile in the water samples is used as the maximum threshold, and the samples larger than the maximum threshold are deleted; the value in the non-water samples with the 15% quantile is used as the minimum threshold, and the samples smaller than the minimum threshold are deleted.

3. Regional water body information extraction

A random forest model was used to classify water bodies from Sentinel-1 data. Random forest has excellent accuracy among all current algorithms. The classification method of machine learning is affected by the imbalance of training data. For categories with few samples, the prediction effect is often not good. Therefore, when the ratio of water body to non-water body is greater than 1:20, this paper limits it to 1:20. In order to simplify the training samples, this paper randomly samples 1/8 of the water body samples in the training data set. When the water body samples are less than 10,000, all water body samples are selected, and non-water bodies are randomly selected in proportion. Also use a cross-validation grid to determine the best parameters for the random forest classifier, such as the number of trees used and the maximum number of features. The trained model is then applied to all pixels of the entire image to get the probability that each pixel is a body of water. Finally, the water body distribution is obtained according to the threshold value. In this paper, the water body is selected to be greater than 0.5, so as to obtain the initial water body extraction result.

4. Post-processing

After information extraction, a small number of abnormal high values were misclassified as water bodies. In order to eliminate the influence of these values, water bodies with VV values greater than the maximum threshold of water bodies were assigned as non-water bodies. After the data after the random forest is classified, most of the water bodies have been extracted, but due to the characteristics of the radar images, the shadows of the hills also show the same characteristics as the water bodies. This paper uses the digital elevation model DEM of srtm30 to calculate the slope. Figure, for all water bodies, when the corresponding slope is greater than 5°, it is assigned as a non-water body, that is, when it is greater than a certain slope, we believe that it cannot retain the water body. There are many scattered pixel points in the water body products generated here. We use the method of seed point diffusion to delete water bodies smaller than a certain area. Due to the accuracy and error of DEM, the edge part of the complete water body is removed, so this part of the water body is supplemented by the regional growth algorithm to obtain the final water body product.

Embodiment 4

This example selects the Linhai flood event in Taizhou City, Zhejiang Province on August 10, 2019 as the research object. Linhai City, Zhejiang Province includes 3 districts and counties, with an average altitude of 300m in the area, and the terrain is relatively complex, including both mountainous areas and plain areas; There are various types of water bodies within the scope, with a catchment area of about 254 square kilometers.

The data used are Sentinel-1 SAR, Landsat8, JRCGlobal Surface Water, SRTM DEM data.

The Sentinel-1 satellite is the first Copernicus project satellite constellation launched by ESA. It consists of two satellites A and B. The combination of the two satellites can achieve image acquisition of the same location every 6 days. It is a C-band satellite and has four imaging modes: Strip Map Mode (SM), Interferometric WideSwath (IW), Extra-Wide Swath Mode (EW), and microwave mode. (Wave-Mode, Wave). Among them, the IW mode is mainly used to cover the earth's land, the SM mode is mainly used for emergency events, and the EW and Wave are mainly used for ocean monitoring. Sentinel-1 satellite data can be obtained free of charge. The width of the IW mode is 250km and the resolution is 5x20m. It only takes about 3-6 hours from satellite shooting to data distribution to the database. Earth observation can penetrate clouds and fog, and is not affected by weather images. , these characteristics make it very suitable for remote sensing monitoring of flood disasters. The Sentinel-1 data in this paper is downloaded from the ESA website (https://scihub.copernicus.eu/ ), and can also be downloaded from the website forwarded by NASA (https://search.asf.alaska.edu).

The JRC global suface water product is mainly used to make training samples for calibrating the random forest model. Produced by the European Joint Research Centre, this water product contains a map of the location and time distribution of surface water from 1984 to 2018, and provides extent and statistics of the water surface, using data from 16 March 1984 to 2018 Generated on 31 December from 3,865,618 scenes collected from Landsat 5, 7 and 8. Each pixel was separately classified as water/non-water using an expert system, and the results were collated into monthly history for the entire time period and two periods (1984-1999, 2000-2018) for change detection. The data mainly used in this article is seasonality2018, which mainly records the number of months of water bodies in 2018, 12 months are permanent water bodies, and less than 12 months are seasonal water bodies.

Embodiment 5

As shown in Figure 6, the results of water body extraction in Example 4 are shown. Most of the water body areas have been extracted, and the phenomenon of missing extraction is less, which is basically not affected by the shadow of mountains. In Figure 7, the wide channel water body in local area 1 is relatively completely extracted, and is not affected by the shadow of the mountains in the upper right corner; in Figure 8, the narrower streams and ponds in local area 2 are also extracted, and the river channel is relatively complete .

Embodiment 6

As shown in Figures 9-14, the satellite images and water body extraction results of the present invention before, during and after the flood in Zhejiang Linhai flood event are shown respectively. The results show that the water body extraction is relatively complete and can be better applied to flood disasters. Remote sensing monitoring.

For a better understanding of the present invention, the above description is made in detail with reference to the specific embodiments of the present invention, but it is not intended to limit the present invention. Any simple modifications made to the above embodiments according to the technical essence of the present invention still belong to the scope of the technical solutions of the present invention. Each embodiment in this specification focuses on the points that are different from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. As for the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and for related parts, please refer to the partial description of the method embodiment.

Claims (9)

1. A radar data water body information extraction method based on a land surface water body product comprises the steps of obtaining satellite data, and is characterized by further comprising the following steps:

step 1: preprocessing the satellite data to generate backscattering data and calculating a derivative coefficient of the backscattering data;

step 2: creating a training sample data set using seasconity data, the backscatter data and the derivation coefficients; the seacontinity data is one of JRC global water products and is used for manufacturing a training sample;

and step 3: extracting water body information of the training sample data based on a random forest model to obtain an initial water body extraction result;

and 4, step 4: post-processing the initial water body extraction result to obtain a final water body product, wherein the step 4 comprises the following substeps:

step 41: calculating a gradient map by using a digital elevation model DEM of srtm 30;

step 42: assigning the corresponding extraction result with the gradient larger than the gradient threshold value in the initial water body extraction result as a non-water body;

step 43: deleting the water body smaller than the area threshold value by a seed point diffusion method;

step 44: supplementing the edge part of the complete water body through a region growing algorithm;

step 45: and generating a final water body product.

2. The method as claimed in claim 1, wherein the preprocessing comprises performing orbit correction, radiation correction, speckle filtering, multi-vision, terrain correction and db conversion on the intensity of the level 1 product after the satellite data is subjected to basic processing, and finally generating the back scattering dataVHAndVV。

3. the method of claim 2, wherein the derivative coefficients comprise polarization ratiosVHrVVThe calculation formula is

。

4. The method of claim 2, wherein the derivative coefficients further comprise normalized deviation polarization indexNDPIThe calculation formula is

。

5. Land water based product mine of claim 2The method for extracting the data water body information is characterized in that the derivative coefficient further comprises a normalized VH indexNVHIThe calculation formula is

。

6. The method of claim 2, wherein the derivation coefficients further comprise a normalized VV indexNVVIThe calculation formula is

。

7. The method of claim 2, wherein the step 2 comprises selecting the water body sample and the non-water body sample according to the water body informationVVSorting the values, and selecting the quantiles in the water body sample asN1% as the maximum threshold, deleting samples greater than the maximum threshold; the water body sample has the following componentsN2% value as the minimum threshold, samples less than the minimum threshold are deleted.

8. The radar-data water body information extraction method based on land-based water body products according to claim 7, wherein the step 3 comprises the following sub-steps:

step 31: randomly selecting from training data setsN3% water samples, when the number of the water samples is less than a number threshold, selecting all the water samples;

step 32: determining the optimal parameters of a random forest classifier by using a cross validation grid GridSearch, and generating a training model;

step 33: applying the training model to all pixels of the whole image to obtain the probability that each pixel is a water body;

step 34: and obtaining water body distribution according to the threshold value to obtain an initial water body extraction result.

9. A radar data water body information extraction system based on a land surface water body product comprises a data acquisition module for acquiring satellite data, and is characterized by further comprising the following modules:

a preprocessing module: the satellite data are preprocessed to generate backscattering data, and a derivative coefficient of the backscattering data is calculated;

a sample training module: for creating a training sample data set using seasconity data, the backscatter data and the derivative coefficients; the seacontinity data is one of JRC global water products and is used for manufacturing a training sample;

the information extraction module: the method is used for extracting water body information of training sample data based on a random forest model to obtain an initial water body extraction result;

a post-processing module: the water body extraction device is used for carrying out post-treatment on the initial water body extraction result to obtain a final water body product; the post-processing comprises the following sub-steps:

step 41: calculating a gradient map by using a digital elevation model DEM of srtm 30;

step 42: assigning the corresponding extraction result with the gradient larger than the gradient threshold value in the initial water body extraction result as a non-water body;

step 43: deleting the water body smaller than the area threshold value by a seed point diffusion method;

step 44: supplementing the edge part of the complete water body through a region growing algorithm;

step 45: generating a final water product;

modules in the system perform radar data water body information extraction according to the method of claim 1.

Images