Precise Water Body Extraction Method Combining Active and Passive Remote Sensing Data

Abstract

The present invention relates to a method for large-scale accurate extraction of water bodies by combining active and passive remote sensing data. The method mainly includes three parts: (1) data preparation and preprocessing. (2) Rough extraction of surface water cover based on threshold segmentation. (3) Accurate extraction of water bodies by combining optical remote sensing data. In an urban area, the present invention can reduce SAR missed extraction and suppress false extraction caused by building roofs and shadow areas. In a non-urban area, the present invention can enhance the recognition of turbid water bodies and suppress the interference of sand dunes and bare land. The present invention can improve the spatiotemporal accuracy and extraction precision of large-scale surface water body extraction, which is conducive to the timely investigation and current management of water resources. The present invention makes full use of the all-day and all-weather earth observation capabilities of Sentinel‑1 active radar data remote sensing and the rich spectral information of Sentinel‑2 passive optical remote sensing, and can automatically, quickly and efficiently obtain high-precision water body thematic data in a large area.

Description

Technical Field

[0001] The present invention relates to a method for accurately extracting water bodies by combining active and passive remote sensing data, belonging to the technical field of remote sensing geoscience applications. Technical Background

[0002] Global climate change has intensified the frequency, intensity, and uncertainty of flood and drought natural disasters. At the same time, sustainable development also faces ecological and environmental problems caused by the long-term over-exploitation of water resources. Therefore, developing rapid and timely investigation technologies for surface water bodies at a large scale helps to promote the development of a dynamic water resource management system and can provide quantitative data support and scientific methods for natural disaster risk assessment and early warning.

[0003] Currently, the research on wide-area, rapid, and timely investigation methods for surface water bodies is still insufficient. The main technical problems include: (1) For the task of large-scale extraction of surface water bodies, during the rapid change period of surface water bodies, current observations are required, but the cloudy and rainy weather during the rapid change period makes it difficult to obtain effective observation data. The above is the contradiction between large-scale and timely extraction of water bodies based on optical remote sensing, which restricts the application of optical remote sensing in large-scale water body extraction. (2) Active remote sensing has the ability to observe all-weather and all-time, but due to the strong echo diffusion effect of buildings, a large number of urban water bodies are missed in the extraction, and at the same time, the low scattering characteristics of smooth surfaces, sand dunes, and bare soils are also easily misclassified as water bodies. There are many accuracy problems in water body extraction based on active remote sensing. In view of this, the present invention proposes a method for automatically extracting water bodies by combining Sentinel-1 / 2 data, which can achieve rapid, robust, and automatic extraction of water bodies at a large scale and high temporal resolution. Summary of the Invention

[0004] The problem to be solved by the present invention is to overcome the contradiction between the need for large-scale, timely, and accurate extraction of water bodies and the shortage of optical data, and to propose a method for accurately extracting water bodies by combining active and passive remote sensing data. This method can promote the complementary advantages and integrated application of optical / radar remote sensing data, and provide technical support and quantitative basis for the current management of surface water resources.

[0005] To solve the above technical problems, the present invention proposes a

[0006] Method for accurately extracting water bodies by combining active and passive remote sensing data, comprising the following steps:

[0007] Step 1, Preparation and preprocessing of active and passive remote sensing data: Prepare Sentinel-1 radar remote sensing data and Sentinel-2 optical remote sensing data within the scope of the area of interest; splice and crop the Sentinel-1 radar remote sensing data to obtain a SAR image covering the area of interest; use the QA60 band to remove clouds from the Sentinel-2 optical remote sensing data and construct an image in the same season as the Sentinel-1 radar remote sensing data.

[0008] Step 2, Coarse extraction of water body range SAR image: Extract water body information using the VH cross-polarized image of the GRD data in the Sentinel-1 radar remote sensing data, calculate the water-land segmentation thresholds of multiple typical areas using the Otsu method and average them to obtain the optimal water-land segmentation threshold τ S1 , and extract the pixels with backscattering coefficient less than the optimal water-land segmentation threshold τ S1 as the alternative water bodies in the area of interest.

[0009] Step 3, Removal of mountain shadow noise: Introduce SRTM slope data to remove the shadow noise caused by the undulating terrain in the area, and remove the pixels with slope greater than 10° in the Sentinel-1 radar remote sensing data.

[0010] Step 4, Extraction of built-up area and bare soil area: Use the normalized difference built-up index (NDBI) to extract the built-up area and bare soil from the Sentinel-2 optical remote sensing data. The calculation formula of the normalized difference built-up index (NDBI) is as follows:

[0011]

[0012] In the formula, ρ SWIR1 is the reflectance of the short-wave infrared band 1, and ρ NIR is the reflectance of the near-infrared band;

[0013] For each scene of the Sentinel-2 optical remote sensing data, extract the pixels with normalized difference built-up index (NDBI) ≥ threshold τ NDBI as the built-up area and bare soil area, construct a pixel-level time series of the built-up area and bare soil area, and extract the stable built-up area and bare soil area using the frequency of the target ground objects appearing in the time series.

[0014] Step 5, Extraction of vegetation area: Use the normalized difference vegetation index (NDVI) to extract the vegetation from the Sentinel-2 optical remote sensing data. The calculation formula of the normalized difference vegetation index (NDVI) is as follows:

[0015]

[0016] In the formula, ρ NIR is the reflectance of the near-infrared band, and ρ Redis the reflectance in the red light band;

[0017] For each scene of Sentinel-2 optical remote sensing data, extract the pixels with the normalized difference vegetation index NDVI ≥ threshold τ NDVI as the vegetation area, construct a pixel-level time series of the vegetation area, and extract the stable vegetation area using the frequency of the target object appearing in the time series;

[0018] Sixth step, stable water body extraction: Use the automatic water extraction index AWEI to extract the water bodies in the Sentinel-2 optical remote sensing data. The calculation formula of the automatic water extraction index AWEI is as follows:

[0019] AWEI = ρ Blue + 2.5×ρ Green - 1.5×(ρ NIR + ρ SWIR1 ) - 0.25×ρ SWIR2

[0020] In the formula, ρ Blue is the reflectance in the blue light band, ρ Green is the reflectance in the green light band, ρ NIR is the reflectance in the near-infrared band, ρ SWIR1 is the reflectance in the short-wave infrared 1 band, ρ SWIR2 is the reflectance in the short-wave infrared 2 band;

[0021] For each scene of Sentinel-2 optical remote sensing data, extract the pixels with the automatic water extraction index AWEI ≥ threshold τ AWEI as the water body, construct a pixel-level time series of the water body, and further extract the stable water body using the frequency of the target object appearing in the time series;

[0022] Seventh step, fine correction of the water body: Combine the above steps to perform fine correction on the alternative water bodies obtained in the second step. The calculation expression is as follows:

[0023] W SC = W S1 - BS S2 - V S2 + W S2

[0024] In the formula, W SC is the accurate extraction result of the water body combining active and passive remote sensing, W S1 refers to the alternative water bodies extracted in the second step; BS S2 refers to the built-up area and bare soil extracted in the fourth step; V S2 refers to the stable water body extracted in the fifth step; W S2 refers to the stable water body extracted in the sixth step;

[0025] The calculation formula for the frequency of the target ground object in the above-mentioned fourth, fifth, and sixth steps in the time series is as follows:

[0026]

[0027] In the formula, fre is the frequency at which a pixel is recognized as the target ground object; N target is the number of times a pixel is recognized as the target ground object, and N total is the total number of pixel observations.

[0028] The main benefits of the present invention are as follows:

[0029] 1. For the urban background, compared with the SAR water body threshold extraction method, the present invention can effectively suppress the noise generated by the strong echo signal of buildings and improve the recall rate of water body extraction; compared with the optical water index method extraction, the present invention can significantly control the false extraction of water bodies caused by the strong reflectivity of building roofs and building shadow areas and improve the precision rate of water body extraction.

[0030] 2. For the non-urban background, compared with the SAR extraction threshold extraction method, the present invention can greatly suppress the interference of sand dunes and bare land on water body extraction and improve the precision rate of water body extraction; compared with the optical water index method extraction, the present invention can enhance the perception ability of turbid water bodies and improve the recall rate of water body extraction.

[0031] 3. The large-scale water body extraction method proposed by the present invention by combining Sentinel-1 and Sentinel-2 data can solve the problem of insufficient currency of optical extraction of surface water bodies under cloudy and rainy weather conditions and the accuracy problem of water body SAR extraction, and provide a fast and robust technical solution for high-resolution, high-precision, and high-currency extraction of surface water bodies in large-scale areas. Description of the Drawings

[0032] Figure 1 is a schematic diagram of the process of the present invention.

[0033] Figure 2 is an example of the NDBI calculation result.

[0034] Figure 3 is an example of the NDVI calculation result.

[0035] Figure 4 is an example of the AWEI calculation result.

[0036] Figure 5 is a partial detail of the accurate extraction result of surface water bodies in the middle and lower reaches of the Yangtze River. Detailed Implementation Modes

[0037] The following elaborates on the preferred embodiments of the present invention in conjunction with the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making the protection scope of the present invention more clearly defined.

[0038] The case study area of the present invention is the middle and lower reaches of the Yangtze River, and the time period for water body extraction is from May 2, 2020 to May 12, 2020. The flow chart of the method for accurately extracting water bodies based on combined active and passive remote sensing data in this embodiment is as Figure 1 shown and includes the following steps:

[0039] First step, preparation and preprocessing of active and passive remote sensing data. Prepare Sentinel-1 radar remote sensing data and Sentinel-2 optical remote sensing data with reference to the scope of the middle and lower reaches of the Yangtze River. Stitch and crop the Sentinel-1 data to obtain a SAR image covering a large area, and use the QA60 band to remove clouds from the Sentinel-2 remote sensing to construct a time-series image in the same season as the Sentinel-1 data.

[0040] The above-mentioned Sentinel-1 radar remote sensing data is the ground range multi-looked image (GRD) in the interferometric wide swath mode (IW), with a pixel size of 10×10m, and the data acquisition time is from May 2, 2020 to May 12, 2020, a total of 49 images. The Sentinel-2 optical remote sensing data is the Level-2A surface reflectance product, and the data acquisition time is from March 1, 2020 to June 30, 2020. The spatial resolution of the visible and near-infrared bands is 10m, and the spatial resolution of the short-wave infrared band is 20m.

[0041] Second step, rough extraction of water body range by SAR. Use the VH cross-polarized image of the Sentinel-1 GRD data to extract water body information, and automatically calculate the optimal water-land segmentation threshold τ S1 of multiple local areas using the Otsu method, and its value is -23dB. Use τ S1 to extract the water body information in the middle and lower reaches of the Yangtze River. Specifically, extract the pixels with a backscattering coefficient less than the optimal water-land segmentation threshold τ S1 as the alternative water bodies in the region of interest. Due to the influence of low backscattering intensity targets such as smooth surfaces (built-up areas), sand dunes, and bare soil, and the double-sided scattering mechanism of urban buildings, there are many accuracy problems with water bodies based on SAR images.

[0042] Third step, removal of mountain shadow noise. Introduce SRTM slope data to remove the shadow noise caused by the undulating terrain in the region, where the slope threshold is determined to be 10°, and remove the pixels with a slope greater than 10° in the Sentinel-1 radar remote sensing data.

[0043] Step 4: Extraction of built-up areas and bare soil areas. The normalized difference built-up index (NDBI) is used to extract built-up areas and bare soil from Sentinel-2 optical remote sensing data. The calculation formula of the normalized difference built-up index NDBI is as follows:

[0044]

[0045] In the formula, ρ SWIR1 is the reflectance of the short-wave infrared band 1, and ρ NIR is the reflectance of the near-infrared band. An example of the NDBI calculation result is shown in Figure 2 as follows.

[0046] The spatio-temporal distribution of built-up areas and bare soil areas is extracted using the threshold τ NDBI , and the value of τ NDBI is 0.1. Specifically, for each scene of Sentinel-2 optical remote sensing data, the pixels with a normalized difference built-up index NDBI ≥ threshold τ NDBI are extracted as built-up areas and bare soil areas, and a pixel-level time series of built-up areas and bare soil areas is constructed. The stable built-up areas and bare soil areas are extracted using the frequency of occurrence of the target ground objects in the time series (temporal frequency feature). The temporal frequency feature formula is as follows:

[0047]

[0048] In the formula, fre B is the frequency of the pixel being identified as buildings and bare soil; N B is the number of times the pixel is identified as buildings and bare soil, and N total is the total number of pixel observations. The temporal feature segmentation method is used to filter the noise of changing ground objects and extract stable built-up areas and bare soil. The frequency threshold τ fre is set to obtain stable buildings and bare soil, and the value of τ fre is 0.8. That is, when the frequency fre of the pixel being identified as the target ground object ≥ 0.8, it is regarded as a stable target ground object.

[0049] Step 5: Extraction of vegetation areas. The normalized difference vegetation index (NDVI) is used to extract vegetation from Sentinel-2 optical remote sensing data. The calculation formula is as follows:

[0050]

[0051] In the formula, ρ NIR is the reflectance of the near-infrared band, and ρ Red is the reflectance of the red light band. An example of the NDVI calculation result is shown in Figure 3as shown

[0052] Using the threshold τ NDVI to extract the spatio-temporal distribution of the vegetation area, where τ NDVI takes the value of 0.3. Specifically, for each scene of Sentinel-2 optical remote sensing data, the pixels with the Normalized Difference Vegetation Index NDVI ≥ threshold τ NDVI are taken as the vegetation area, and a pixel-level time series of the vegetation area is constructed. The stable vegetation area is extracted using the frequency of occurrence of the target object in the time series (temporal frequency feature). The formula for the temporal frequency feature is as follows:

[0053]

[0054] In the formula, fre V is the frequency of the pixel being identified as vegetation; N V is the number of times the pixel is identified as vegetation, and N total is the total number of pixel observations. The temporal feature segmentation method is used to filter noise and extract stable vegetation. The frequency threshold τ fre is set to obtain the extraction of stable vegetation, where τ fre takes the value of 0.8. That is, when the frequency fre of the pixel being identified as the target object is ≥ 0.8, it is regarded as a stable target object.

[0055] Step 6, extraction of stable water bodies. The water bodies in Sentinel-2 optical remote sensing data are extracted using the Automated Water Extraction Index (AWEI). The formula for the Automated Water Extraction Index AWEI is as follows:

[0056] AWEI = ρ Blue + 2.5×ρ Green - 1.5×(ρ NIR + ρ SWIR1 ) - 0.25×ρ SWIR2

[0057] In the formula, ρ Blue is the reflectance of the blue band, ρ Green is the reflectance of the green band, ρ NIR is the reflectance of the near-infrared band, ρ SWIR1 is the reflectance of the short-wave infrared 1 band, ρ SWIR2 is the reflectance of the short-wave infrared 2 band. An example of the AWEI calculation result is as Figure 4 shown

[0058] Using the threshold τ AWEI to extract the spatio-temporal distribution of the water bodies, where τ AWEIThe value of AWEI is 0. Specifically, for each scene of Sentinel-2 optical remote sensing data, pixels with the Automatic Water Extraction Index (AWEI) ≥ threshold τ

[0059]

[0060] are taken as water bodies, and a pixel-level water body time series is constructed. Further, stable water bodies are extracted using the frequency of the target object appearing in the time series (temporal frequency feature). The formula for the temporal frequency feature is as follows: W In the formula, fre W is the frequency of the pixel being identified as a water body; N total is the number of times the pixel is identified as a water body, and N fre is the total number of pixel observations. The temporal feature segmentation method is used to extract stable surface water bodies, and the frequency threshold τ fre is set to obtain stable vegetation extraction, and the value of τ

[0061] is 0.8. That is, when the frequency fre of the pixel being identified as the target object ≥ 0.8, it is regarded as a stable target object.

[0062] Step 7: Fine correction of water bodies. Combine the above steps to perform fine correction on the SAR water body results, and the calculation expression is as follows:

[0063] W SC = W S1 - BS S2 - V S2 + W S2

[0064] In the formula, W SC is the accurate water body extraction result combining active and passive remote sensing. W S1 refers to the water bodies extracted in the second step. BS S2 refers to the built-up areas and bare soils extracted in the fourth step; V S2 refers to the stable water bodies extracted in the fifth step, which are used to remove the mis-extracted SAR water bodies, including low backscattering feature objects such as smooth artificial surfaces, bare soils, sand dunes, paddy fields (after transplanting), and wheat fields (before jointing stage). W S2 refers to the stable water body range extracted in the sixth step, which is used to complement the missed extraction of SAR water bodies, mainly caused by the interference effect of strong echo signals of corner reflector structures in urban areas. The local details of the accurate water body extraction results in the middle and lower reaches of the Yangtze River from May 2, 2020 to May 12, 2020 are asFigure 5 as shown

[0065] The above are only embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied to other related technical fields, shall be similarly included in the patent protection scope of the present invention.

[0066] In addition to the above embodiments, the present invention may also have other implementation manners. Any technical solution formed by equivalent replacement or equivalent transformation falls within the protection scope required by the present invention.

Claims

1. A method for accurately extracting water bodies by combining active and passive remote sensing data, comprising the following steps: Step 1, Preparation and preprocessing of active and passive remote sensing data: Prepare Sentinel-1 radar remote sensing data and Sentinel-2 optical remote sensing data within the scope of the area of interest; And splice and crop the Sentinel-1 radar remote sensing data to obtain a SAR image covering the area of interest; Use the QA60 band to perform cloud removal on the Sentinel-2 optical remote sensing data to construct an image in the same season as the Sentinel-1 radar remote sensing data; Step 2, Coarse extraction of SAR images of water body range: Extract water body information using the VH cross-polarized image of the GRD data in Sentinel-1 radar remote sensing data, calculate the water-land segmentation thresholds of multiple typical regions using the Otsu method and take the average to obtain the optimal water-land segmentation threshold τ S1 , extract the pixels with backscattering coefficient less than the optimal water-land segmentation threshold τ S1 as the alternative water bodies of the region of interest; Step 3, Removal of mountain shadow noise: Introduce SRTM slope data to remove the shadow noise caused by the undulating terrain in the area, and remove the pixels with a slope greater than 10° in the Sentinel-1 radar remote sensing data; Step 4, Extraction of built-up areas and bare soil areas: Use the Normalized Difference Built-up Index (NDBI) to extract the built-up areas and bare soil in the Sentinel-2 optical remote sensing data. The calculation formula of the Normalized Difference Built-up Index (NDBI) is as follows: where ρ SWIR1 is the reflectance in the short-wave infrared band 1, and ρ NIR is the reflectance in the near-infrared band; For each scene of Sentinel-2 optical remote sensing data, extract the pixels with the Normalized Difference Built-up Index (NDBI) ≥ threshold τ NDBI as the built-up areas and bare soil areas, construct the time series of built-up areas and bare soil areas at the pixel level, and extract the stable built-up areas and bare soil areas by using the frequency of the target ground objects appearing in the time series; Step 5, Extraction of vegetation areas: Use the Normalized Difference Vegetation Index (NDVI) to extract the vegetation in the Sentinel-2 optical remote sensing data. The calculation formula of the Normalized Difference Vegetation Index (NDVI) is as follows: where ρ NIR is the reflectance in the near-infrared band, and ρ Red is the reflectance in the red band; For each scene of Sentinel-2 optical remote sensing data, extract the pixels with the normalized difference vegetation index (NDVI) ≥ threshold τ NDVI as the vegetation area, construct a pixel-level time series of the vegetation area, and extract the stable vegetation area by using the frequency of the target ground object appearing in the time series; Step 6, Extraction of stable water bodies: Use the Automated Water Extraction Index (AWEI) to extract the water bodies in the Sentinel-2 optical remote sensing data. The calculation formula of the Automated Water Extraction Index (AWEI) is as follows: AWEI = ρ Blue + 2.5×ρ Green - 1.5×(ρ NIR + ρ SWIR1 ) - 0.25×ρ SWIR2 where ρ Blue is the reflectance in the blue light band, ρ Green is the reflectance in the green light band, ρ NIR is the reflectance in the near-infrared band, ρ SWIR1 is the reflectance in the short-wave infrared 1 band, ρ SWIR2 is the reflectance in the short-wave infrared 2 band; For each scene of Sentinel-2 optical remote sensing data, extract the pixels with the Automated Water Extraction Index (AWEI) ≥ threshold τ as water bodies, construct a pixel-level water body time series, and further extract stable water bodies by using the frequency of the target ground object appearing in the time series; AWEI ​ Step 7, Fine correction of water bodies: Combine the above steps to perform fine correction on the candidate water bodies obtained in Step 2. The calculation expression is as follows: W SC = W S1 - BS S2 - V S2 + W S2 In the formula, W SC is the accurate water body extraction result of combined active and passive remote sensing, and W S1 refers to the alternative water bodies extracted in the second step; BS S2 refers to the built-up areas and bare soils extracted in the fourth step; V S2 refers to the stable vegetation areas extracted in the fifth step; W S2 refers to the stable water bodies extracted in the sixth step. The calculation formula for the frequency of occurrence of the target ground objects in the above Step 4, Step 5 and Step 6 in the time series is as follows: where fre is the frequency at which a pixel is recognized as the target ground object; N target is the number of times a pixel is recognized as the target ground object, and N total is the total number of times the pixel is observed.

2. The method for accurately extracting water bodies by combining active and passive remote sensing data according to claim 1, characterized in that: The Sentinel-1 radar remote sensing data is a ground-range multi-looked image (GRD) in the Interferometric Wide Swath (IW) mode, with a pixel size of 10m × 10m; The Sentinel-2 optical remote sensing data is a Level-2A surface reflectance product, with a spatial resolution of 10m in the visible and near-infrared bands and a spatial resolution of 20m in the short-wave infrared band.

3. The method for accurately extracting water bodies by combining active and passive remote sensing data according to claim 1, characterized in that: Threshold τ S1 , τ NDBI , τ NDVI , τ AWEI The values are -23 dB, 0.1, 0.3, and 0 respectively.

4. The method for accurately extracting water bodies by combining active and passive remote sensing data according to claim 1, characterized in that: When the frequency fre of a pixel being identified as a target ground object is ≥ 0.8, it is regarded as a stable target ground object.

Images