A method and device for extracting surface water bodies based on multi-source remote sensing data

Abstract

A method and device for extracting surface water bodies based on multi-source remote sensing data, wherein the method comprises: acquiring synthetic aperture radar images, multi-scene optical remote sensing images and digital elevation model data of a target study area; constructing a first water body confidence of each pixel based on the statistical distribution law of the backscattering characteristics of the synthetic aperture radar images; spatially registering the multi-scene optical remote sensing images and the synthetic aperture radar images, updating the first water body confidence based on water body spectral characteristics, and obtaining a second water body confidence; extracting terrain undulation characteristics based on the digital elevation model data, and constructing a terrain constraint mask; filtering effective samples and statistical samples of the second water body confidence using the terrain constraint mask, performing threshold segmentation based on the confidence distribution characteristics of the filtered statistical samples, and obtaining an optimal threshold; dividing the effective samples by taking the optimal threshold as a water body determination threshold, and obtaining an initial water body mask; and outputting a final water body extraction result after spatial morphological optimization processing.

Description

Technical Field

[0001] This invention relates to the field of remote sensing data processing and information extraction technology, specifically to a method and apparatus for extracting land surface water bodies based on multi-source remote sensing data. Background Technology

[0002] Land surface water information is crucial foundational data for water resource surveys, ecological environment monitoring, flood control and disaster reduction, and global change research. Currently, remote sensing-based methods for extracting land surface water mainly rely on a single data source or a simple combination of multi-source data.

[0003] Water body extraction methods based on optical remote sensing imagery (such as the water index method or pre-trained model method) rely on the spectral characteristics of ground features. Natural water bodies have strong absorption characteristics in the near-infrared and short-wave infrared bands, while their reflection is relatively strong in the green band. Therefore, the water body signal can be enhanced by constructing a normalized water index (NDWI). However, optical imagery is highly susceptible to the effects of clouds, cloud shadows, mountain shadows, and changes in atmospheric conditions. Under cloudy, shadowy, or complex lighting conditions, the spectral characteristics of water bodies are easily confused, resulting in insufficient stability of single-scene image extraction results and making it difficult to meet actual monitoring needs.

[0004] Water body extraction methods based on synthetic aperture radar (SAR) imagery utilize the low backscattering characteristics of water bodies. Because water surfaces are smooth, radar waves undergo specular reflection upon incident, resulting in very little energy returning to the sensor, which appears as "dark pixels" in SAR images; while land surfaces are rough and appear as "bright pixels." However, existing SAR water body extraction methods mostly use fixed or empirical thresholds to segment backscattering coefficients. The scattering characteristics of water bodies and non-water bodies differ significantly across regions, imaging modes, and surface conditions. Fixed thresholds have limited generalization ability and are prone to misclassifying low-scattering non-water bodies such as radar shadows and smooth bare land as water bodies.

[0005] Existing multi-source remote sensing fusion methods typically combine results from different data sources using logical AND / OR rules or simple weighting, lacking an effective quantification and dynamic adjustment mechanism for the contribution of different data sources, and thus failing to fully leverage the complementary advantages of multi-source data.

[0006] Furthermore, in mountainous and plateau regions with significant topographic relief, terrain shadows and overlapping effects can severely impact the accuracy of water body extraction. Existing methods generally lack effective constraints on topographic factors such as slope and terrain relief, leading to a high misjudgment rate in complex terrain areas.

[0007] In summary, existing technologies struggle to balance stability, adaptability, and accuracy under complex surface environments and variable imaging conditions. Therefore, there is an urgent need for a method and apparatus for extracting land surface water bodies based on multi-source remote sensing data. Summary of the Invention

[0008] The present invention aims to solve the problems of existing land surface water extraction methods, such as strong threshold dependence, insufficient stability and high risk of misjudgment under complex surface conditions.

[0009] To achieve the above objectives, in a first aspect, the present invention provides a method for extracting land surface water bodies based on multi-source remote sensing data, comprising: acquiring multi-source remote sensing data covering the target study area and corresponding to each other in spatial location, wherein the multi-source remote sensing data includes at least synthetic aperture radar image data, multi-scene optical remote sensing image data, and digital elevation model data;

[0010] Based on the statistical distribution law of the backscattering characteristics of the synthetic aperture radar image data, the first water body confidence level of each pixel is constructed.

[0011] Spatially register the multi-scene optical remote sensing imagery with the synthetic aperture radar imagery data. Based on the water body spectral characteristics of the multi-scene optical remote sensing imagery, enhance and update the confidence level of the first water body to obtain the confidence level of the second water body.

[0012] Based on the digital elevation model data, terrain undulation features are extracted and a terrain constraint mask is constructed.

[0013] The terrain constraint mask is used to screen the confidence scores of the second water body by selecting effective samples and statistical samples. Based on the confidence score distribution characteristics of the selected statistical samples, adaptive threshold segmentation is performed to obtain the optimal threshold. The optimal threshold is used as the water body determination threshold to classify the effective samples into water body and non-water body types, thus obtaining the initial water body mask.

[0014] The initial water body mask is subjected to spatial morphological optimization to obtain the final land surface water information extraction result.

[0015] Preferably, the step of constructing the first water body confidence score for each pixel based on the statistical distribution law of the backscattering characteristics of the synthetic aperture radar image data includes:

[0016] The synthetic aperture radar image data is preprocessed to obtain a backscattering coefficient image;

[0017] Histogram statistical analysis is performed on the effective pixels in the backscattering coefficient image whose values ​​are within the preset effective scattering range to determine the peak parameters representing the low scattering characteristics of the water body;

[0018] Based on the peak parameters, a synthetic aperture radar water response function is constructed, and the backscattering coefficient of each pixel is mapped to the first water confidence level in the interval from 0 to 1. The output value of the synthetic aperture radar water response function is assigned a preset weight so that its maximum contribution value is controlled.

[0019] Specifically, the preset effective backscattering range is -40dB to -10dB; the expression for the synthetic aperture radar water response function is:

[0020]

[0021] in, For pixels The confidence level of the first water body at that location. For pixels Backscattering coefficient at VH polarization For peak parameters, This is the normalized scaling parameter; Preset weights; This is the interval constraint function.

[0022] Preferably, the step of spatially registering the multi-view optical remote sensing imagery with the synthetic aperture radar imagery, and enhancing and updating the confidence level of the first water body based on the water body spectral characteristics of the multi-view optical remote sensing imagery, includes:

[0023] Based on the spatial range and pixel grid of the synthetic aperture radar image, the green band, near-infrared band and scene classification files of each optical remote sensing image are spatially registered with it.

[0024] After removing cloud-polluted pixels from the scene classification file, the normalized water index of each scene image is calculated, and the maximum value of the normalized water index of all images covering the pixel is taken as the composite water index.

[0025] The synthetic water index is subjected to adaptive threshold segmentation to determine the optical discrimination threshold;

[0026] Specifically, the effective pixels in the synthetic water index are filtered by numerical range, and pixels with values ​​between -0.3 and 0.8 are retained for adaptive threshold calculation to determine the optical discrimination threshold.

[0027] For pixels whose synthetic water index is greater than the optical discrimination threshold, their water confidence is updated to a value no lower than a preset high confidence value.

[0028] Specifically, the water body confidence level is updated to a value no lower than a preset high confidence level using the following formula:

[0029]

[0030] in, The confidence level for the second water body. 0.85 represents the confidence level for the first water body, and 0.85 is the preset high confidence level value.

[0031] Preferably, the step of extracting terrain undulation features based on the digital elevation model data and constructing a terrain constraint mask includes:

[0032] Gaussian smoothing filter is applied to the digital elevation model data to obtain the regional background topographic surface;

[0033] The absolute value of the difference between the digital elevation model and the background terrain surface is calculated as the terrain undulation amount of each pixel;

[0034] Pixels with terrain undulation less than a preset terrain undulation threshold are marked as flat terrain areas, and a Boolean terrain constraint mask is generated.

[0035] Specifically, the step of using the terrain constraint mask to screen effective and statistical samples of the confidence level of the second water body, and performing adaptive threshold segmentation based on the confidence level distribution characteristics of the screened statistical samples, includes:

[0036] From the confidence scores of the second water body, pixels that simultaneously meet the conditions of being located within the effective area of ​​the terrain constraint mask and having a confidence score greater than a preset value are selected as valid samples. Among the valid samples, pixels whose confidence scores are within a preset confidence analysis interval are further selected as statistical samples.

[0037] The confidence distribution of the statistical samples is adaptively segmented using the Otsu's method to determine the optimal threshold, which is then used as the water body determination threshold.

[0038] Preferably, the preset value is 0.3, and the preset reliability analysis interval is 0.5 to 1.0.

[0039] Specifically, the morphological optimization of the initial water mask includes performing morphological closing operations and hole filling operations on the initial water mask sequentially.

[0040] Secondly, the present invention provides a land surface water extraction device based on multi-source remote sensing data, comprising:

[0041] The data acquisition module is used to acquire multi-source remote sensing data that covers the target study area and whose spatial locations correspond to each other. The multi-source remote sensing data includes at least synthetic aperture radar image data, multi-scene optical remote sensing image data, and digital elevation model data.

[0042] The synthetic aperture radar confidence construction module is used to construct the first water body confidence of each pixel based on the statistical distribution law of the backscattering characteristics of the synthetic aperture radar image data.

[0043] The optical confidence update module is used to spatially register the multi-scene optical remote sensing images with the synthetic aperture radar images, and enhance and update the first water body confidence based on the water body spectral characteristics of the multi-scene optical remote sensing images to obtain the second water body confidence.

[0044] The terrain constraint module is used to extract terrain undulation features based on the digital elevation model data and construct a terrain constraint mask.

[0045] The threshold adaptive segmentation module is used to use the terrain constraint mask to screen the confidence of the second water body by effective samples and statistical samples, and to perform adaptive threshold segmentation based on the confidence distribution characteristics of the screened statistical samples to determine the optimal threshold. The optimal threshold is used as the water body judgment threshold to classify the effective samples into water body and non-water body types to obtain the initial water body mask.

[0046] The morphological processing module is used to perform spatial morphological optimization processing on the initial water body mask to obtain the final land surface water body information extraction result.

[0047] Compared with existing technologies, this invention has the following advantages: This invention unifies SAR backscattering features, optical water body spectral features, and DEM topographic relief features into water body confidence scores for fusion, and significantly improves the stability and reliability of water body extraction through an adaptive threshold determination mechanism; by introducing DEM topographic constraints, it effectively reduces the probability of misjudgment in complex terrains such as mountainous areas and plateaus; at the same time, it eliminates the need to manually set thresholds for different regions, making it suitable for the engineering processing of large-scale, multi-temporal remote sensing images. Attached Figure Description

[0048] Figure 1 A technical flowchart provided for an embodiment of the present invention;

[0049] Figure 2 This invention provides a preprocessed Sentinel-1 SAR image and its effective data range map.

[0050] Figure 3 This invention provides a set of 13 Sentinel-2 optical remote sensing images covering a research area, as provided in an embodiment of the invention.

[0051] Figure 4 A DEM data map of a study area provided in an embodiment of the present invention;

[0052] Figure 5 A comparison image of Sentinel-1 data preprocessing before and after is provided in an embodiment of the present invention;

[0053] Figure 6 A statistical histogram of VH polarization backscattering coefficients is provided in this embodiment of the invention.

[0054] Figure 7 A water body confidence map based on SAR data is provided as an embodiment of the present invention;

[0055] Figure 8 A schematic diagram of the spatial coverage of SAR and optical data provided in an embodiment of the present invention;

[0056] Figure 9 An NDWI maximum value image provided in an embodiment of the present invention;

[0057] Figure 10 A statistical chart for obtaining the NDWI threshold using the Otsu thresholding method is provided in this embodiment of the invention;

[0058] Figure 11 An updated water body confidence map is provided for an embodiment of the present invention;

[0059] Figure 12 A generated DEM terrain mask map provided in an embodiment of the present invention;

[0060] Figure 13 This is a comparison image of water extraction results before and after morphological treatment, provided in an embodiment of the present invention.

[0061] Figure 14 This invention provides a final water extraction result and corresponding confidence level diagram for an embodiment of the invention. Detailed Implementation

[0062] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be described below with reference to the accompanying drawings. It should be noted that similar reference numerals and letters in the following drawings indicate similar items; therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0064] In the description of the embodiments of the present invention, the words "exemplary," "for example," or "for instance" are used to indicate that they are examples, illustrations, or descriptions. Any embodiment or design that is described as "exemplary," "for example," or "for instance" in the embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Rather, the use of the words "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a specific manner.

[0065] Information on land surface water bodies is crucial foundational data for water resource surveys, ecological environment monitoring, flood control and disaster reduction, and global change research. Currently, remote sensing-based methods for extracting land surface water body information mainly include pre-trained model methods and water index methods based on optical remote sensing imagery, threshold segmentation methods based on synthetic aperture radar imagery, and multi-source remote sensing data fusion methods. However, existing technologies still have the following shortcomings.

[0066] (I) Principles and limitations of water body information extraction methods based on SAR data

[0067] Synthetic Aperture Radar (SAR) is an active remote sensing technology that acquires information about ground features by transmitting microwave signals to the Earth's surface and receiving their reflected echoes. Therefore, it is independent of sunlight and can operate stably under complex weather conditions, including day and night, and cloud cover. In SAR observations, different ground features reflect radar signals in significantly different ways, with water bodies being particularly representative: because natural water surfaces are usually smooth, incident radar waves are reflected specularly away from the sensor, with only a small portion returning to the satellite, resulting in low backscattering characteristics in the image, i.e., "dark pixels." Land surfaces (such as soil, vegetation, or buildings) are typically rougher, scattering radar waves in multiple directions, causing some energy to return to the sensor, thus appearing as high backscattering in the image, i.e., "bright pixels." Based on this physical difference of "dark water and bright land," water areas can be identified by analyzing the backscattering intensity of pixels in SAR images.

[0068] Specifically, the process typically begins by statistically analyzing the scattering values ​​of all pixels in the image to identify low-scattering feature ranges representing water bodies. Then, based on the proximity of each pixel's scattering value to this feature, the likelihood of it being a water body is determined, thus extracting water body information. Compared to simply setting a fixed threshold, this statistical feature-based approach better adapts to variations in different regions and environmental conditions, improving the stability and reliability of water body identification. However, it's important to note that in certain special cases (such as radar shadows, extremely smooth bare land, or wetlands), non-water areas may also exhibit low-scattering features. Therefore, in practical applications, it is often combined with other data, such as optical remote sensing data, for comprehensive analysis to further improve the accuracy of water body extraction.

[0069] Existing SAR water body extraction methods mostly use fixed or empirical thresholds to segment backscattering coefficients. The scattering characteristics of water bodies and non-water bodies vary significantly in different regions, imaging modes, and surface conditions. Fixed or empirical thresholds are difficult to apply universally and can easily misclassify low-scattering non-water features or terrain shadows as water bodies.

[0070] (II) Principles and Limitations of Water Body Information Extraction Methods Based on Optical Data

[0071] Water body information extraction based on optical satellite imagery relies primarily on the unique spectral characteristics of water bodies across different wavelengths. Natural water bodies possess a certain reflectivity in the visible light band (especially the green band), while exhibiting strong absorption characteristics in the near-infrared and short-wave infrared bands, with reflectivity significantly lower than that of vegetation, soil, and buildings. Therefore, in optical imagery, water bodies typically appear as "relatively bright in the visible light band and significantly darker in the near-infrared and short-wave infrared bands," and this spectral difference provides the basis for water body identification.

[0072] Based on this principle, studies typically construct water body indices (such as the Normalized Difference Water Index (NDWI) or the modified MNDWI) and utilize the combination relationships between different bands to enhance the contrast between water bodies and non-water bodies, thereby highlighting water body information. For example, by calculating the normalized difference between the green band and the near-infrared (or short-wave infrared) band, water body pixels can exhibit higher index values, while vegetation and soil can show lower or even negative values.

[0073] It should be noted that optical methods are highly sensitive to clouds, shadows, and water turbidity, and may lead to misjudgments under complex meteorological conditions or in areas with highly turbid water. Optical images are easily affected by clouds, fog, cloud shadows, mountain shadows, and changes in atmospheric conditions. Under cloudy, shadowy, or high-reflectivity backgrounds, the spectral characteristics of water bodies are easily confused, making it difficult to unify water body index thresholds, reducing the generalization ability of pre-trained models, and resulting in insufficient stability of water body extraction results from single-scene images, which is difficult to meet actual monitoring needs.

[0074] (III) The role and limitations of digital elevation models (DEM) in water body information extraction

[0075] Digital elevation models (DEMs) play a crucial role in water body information extraction, serving as both topographic constraints and misclassification corrections. Since water bodies are typically controlled by gravity in their natural state, primarily distributed in low-lying areas and continuously along surface runoff paths (such as rivers, lakes, and basins), topographic information provides reliable spatial constraints for water body identification. In practical extraction, relying solely on optical or SAR data is susceptible to interference from factors such as shadows, cloud cover, low-reflectivity surfaces, or wetlands, leading to misclassification of non-water areas as water bodies. DEM data, however, can effectively correct these errors through slope, topographic relief, and runoff characteristics. For example, in high-slope areas, true water bodies are unlikely to exist stably; therefore, slope thresholds can be used to eliminate "pseudo-water bodies" on steep slopes. In areas with topographic shadows, although remote sensing images may exhibit low reflectivity, DEM analysis can identify them as located on the shady side of a mountain rather than in a low-lying catchment area, thus avoiding misclassification. In addition, hydro-topographic parameters such as flow direction and cumulative runoff can be extracted from DEM to determine the spatial connectivity and reasonable distribution range of water bodies, making the extraction results more consistent with the actual hydrological process.

[0076] However, existing methods generally lack effective constraints on topographic factors such as slope, aspect, and terrain shading, which can easily lead to misclassification of terrain shadows or low-lying non-water areas as water bodies, thus limiting the accuracy and reliability of water body extraction results in complex terrain areas.

[0077] (iv) Limitations of existing multi-source fusion methods

[0078] Existing multi-source remote sensing fusion methods mostly use logical AND / OR rules or simple weighting to combine water body identification results from different data sources. If the weights of each data source are not dynamically adjusted according to image quality, imaging conditions, or regional characteristics, it is impossible to effectively quantify and distinguish the information contribution of different data sources in water body identification. This can lead to misjudgments in complex environments and make it difficult to fully leverage the complementary advantages of multi-source remote sensing data.

[0079] To overcome the shortcomings of existing technologies, this invention provides a method for extracting land surface water bodies based on multi-source remote sensing data. The overall technical process is as follows: Figure 1 As shown in the figure. This method constructs pixel-level water body confidence and achieves adaptive threshold determination by fusing SAR, optical, and DEM data.

[0080] Step S1: Acquire multi-source remote sensing data that covers the target study area and whose spatial locations correspond to each other. The multi-source remote sensing data includes at least synthetic aperture radar image data, multi-scene optical remote sensing image data, and digital elevation model data.

[0081] For example, firstly, the geographical scope of the target area for extracting information on land surface water bodies is determined, and multi-source data covering the target study area is acquired. This multi-source data includes at least:

[0082] Synthetic Aperture Radar (SAR) image data, after image preprocessing, is used to generate vector polygon files according to its effective data range for subsequent use in finding spatially corresponding optical remote sensing image data. Figure 2 This invention provides a preprocessed Sentinel-1 SAR image and its effective data range map.

[0083] Multiple optical remote sensing image data that spatially overlap with the aforementioned synthetic aperture radar (SAR) image data, Figure 3 This invention provides a set of 13 Sentinel-2 optical remote sensing images covering a research area, as provided in an embodiment of the invention.

[0084] Digital elevation model (DEM) data for the corresponding spatial range, Figure 4 This is a DEM data map of a study area provided in an embodiment of the present invention.

[0085] SAR imagery is used to characterize surface backscattering properties, optical remote sensing imagery is used to calculate the water spectral index (NDWI), and DEM data is used to characterize topographic relief features. In this embodiment, the SAR imagery is Sentinel-1 imagery, and the optical remote sensing imagery is Sentinel-2 imagery.

[0086] Step S2: Based on the statistical distribution law of the backscattering characteristics of the synthetic aperture radar image data, construct the first water body confidence score for each pixel.

[0087] For example, the SAR image data is preprocessed to obtain a backscattering coefficient image; histogram statistical analysis is performed on the effective pixels in the backscattering coefficient image whose values ​​are within a preset effective scattering range to determine the peak parameter representing the low scattering characteristics of the water body; a synthetic aperture radar water body response function is constructed based on the peak parameter, and the backscattering coefficient of each pixel is mapped to the first water body confidence level in the range of 0 to 1, wherein the output value of the synthetic aperture radar water body response function is assigned a preset weight to control its maximum contribution value.

[0088] In one embodiment, the preset effective scattering range is -40dB to -10dB; the expression for the synthetic aperture radar water response function is:

[0089]

[0090] in, For pixels The confidence level of the first water body at that location. For pixels Backscattering coefficient at VH polarization For peak parameters, For normalized scaling parameters, To preset weights, This is the interval constraint function.

[0091] For example, in the land surface water body information extraction method based on multi-source remote sensing data fusion described in this invention, synthetic aperture radar (SAR) data, as an all-day, all-weather active remote sensing data source, can effectively overcome the limitations of optical remote sensing caused by cloud cover and changes in lighting conditions. Therefore, this invention extracts basic evidence for water body identification from the backscattering characteristics of SAR data, providing a stable and continuous initial basis for subsequent multi-source fusion water body identification.

[0092] First, the acquired SAR data for the target area is preprocessed, including necessary processing such as orbit correction, radiometric calibration, speckle noise suppression, and terrain correction, to obtain a backscattering coefficient image with physical meaning. In this embodiment, SNAP software is used to perform normalization processing such as orbit correction, radiometric calibration, speckle noise suppression, and terrain correction on the vertically transmitted and horizontally received (VH) polarimetric SAR data to obtain the backscattering coefficient image. Figure 5 This is a comparison image of Sentinel-1 data before and after preprocessing, provided as an embodiment of the present invention. Subsequently, the valid VH polarization backscattering data is filtered, retaining only image data with values ​​in the range of -40dB to -10dB to remove noise, outliers, and atypical scattering information, thus constructing an effective sample set for statistical analysis.

[0093] Subsequently, histogram statistical analysis was performed on the effective sample set. The VH polarization backscattering coefficients were graded at 0.5 dB intervals, and the number of pixels in each scattering intensity interval was counted to obtain the frequency distribution characteristics of the VH polarization backscattering coefficients. Figure 6 This invention provides a statistical histogram of VH-polarized backscattering coefficients. Based on this, by identifying the scattering intensity range with the largest number of pixels in the frequency distribution, the main peak range of the VH-polarized backscattering coefficients is determined, and the median of the scattering intensity range corresponding to this main peak range is used as the peak characteristic parameter of the VH-polarized backscattering. .

[0094] Then, a confidence matrix with the same spatial size as the SAR image is constructed to store the water confidence values ​​calculated for each pixel based on SAR information, with its initial value set to zero. A SAR response function is constructed, and using the previously determined VH polarization backscattering peak characteristic parameters, the SAR water response for each pixel is calculated using formula (1):

[0095] Formula (1)

[0096] in, For pixels The confidence level of the water body is calculated based on SAR data. For pixels Backscattering coefficient at VH polarization These are the characteristic parameters of the VH polarization backscattering peak obtained through statistical analysis. This is a normalized scale parameter used to adjust the sensitivity to scattering differences; in this embodiment, it is set to 6 dB. This is the weighting coefficient for SAR information, used to limit its contribution in multi-source fusion. In this embodiment, the value is 0.6. This is an interval constraint function used to restrict the calculation results to the range [0,1]. Its definition is given in formula (2):

[0097] Formula (2)

[0098] The above steps normalize the difference between the VH polarization backscattering intensity and the peak characteristic parameter to construct the SAR response function, calculate the SAR water body confidence score for each pixel, and limit the calculation results to a preset range to avoid excessive influence of anomalous scattering values. Simultaneously, preset weights are assigned to the SAR response results, ensuring that the water body confidence score obtained from SAR data acts as a weak constraint factor in the overall confidence score system, and its maximum contribution is controlled.

[0099] Based on this, the confidence matrix is ​​updated only for valid pixel regions, assigning the SAR water response score of the corresponding pixel to the confidence matrix, while invalid pixels retain their default values, thus ensuring spatial consistency and data reliability of the confidence results. See the water confidence map based on SAR data. Figure 7 , Figure 7 This invention provides a water body confidence map based on SAR data.

[0100] Step S3: Spatial registration is performed between the multi-scene optical remote sensing images and the synthetic aperture radar image data. Based on the water body spectral characteristics of the multi-scene optical remote sensing images, the confidence level of the first water body is enhanced and updated to obtain the confidence level of the second water body.

[0101] For example, based on the spatial range and pixel grid of the synthetic aperture radar image, spatial registration is performed on the green band, near-infrared band, and scene classification files of each optical remote sensing image. After removing cloud-polluted pixels according to the scene classification files, the normalized water index (NDWI) of each image is calculated. On a pixel-by-pixel basis, the maximum value of the NDWI of all images covering the pixel is taken as the synthetic water index. The effective pixels in the synthetic water index are filtered by numerical range, and pixels with values ​​between -0.3 and 0.8 are retained for adaptive threshold calculation to determine the optical discrimination threshold. For pixels whose synthetic water index is greater than the optical discrimination threshold, their water confidence is updated to a value not lower than a preset high confidence value, where the high confidence value is greater than the maximum possible value of the initial water confidence.

[0102] In one embodiment, the update of the water body confidence level to a preset high confidence value is performed using the following formula:

[0103]

[0104] in, The confidence level for the second water body. 0.85 represents the confidence level for the first water body, and 0.85 is the preset high confidence level value.

[0105] For example, in the case of multiple Sentinel-2 optical images spatially covering the Sentinel-1 coverage area, a multi-image spatial joint processing method is used to construct optical evidence of water bodies. The specific implementation method is as follows. Figure 8 This is a schematic diagram of the spatial coverage of SAR and optical data provided in an embodiment of the present invention. Figure 8 The spatial coverage of Sentinel-1 and Sentinel-2 data is shown.

[0106] (1) Spatial Registration. Using the spatial extent, pixel grid, and coordinate system of the Sentinel-1 SAR image as the target reference, the green band (B03), near-infrared band (B08), and scene classification file (SCL) of each Sentinel-2 optical image were spatially registered with it. During spatial registration, continuous spectral data (B03, B08) were resampled using bilinear interpolation; discrete classification data (SCL) were resampled using nearest neighbor to maintain category integrity. Simultaneously, to reduce memory consumption, the image was processed block by block according to a fixed block size, and the affine transformation relationship was determined for each block in the reference space. This ensures a one-to-one correspondence between pixels of multiple optical images and the SAR grid, and facilitates joint calculation of multiple NDWI images in space.

[0107] (2) Calculate NDWI for multiple scenes. First, select effective pixels in the optical image as the basis for subsequent processing. Based on the SCL file after reprojection, filter the optical data, remove pixels with a high probability of cloud cover (SCL values ​​of 4, 5, and 6), require that the values ​​of green band and near-infrared band pixels be greater than zero, remove abnormal or invalid pixels, and mark the pixels that meet the conditions as the set of effective pixels. Second, calculate NDWI for the effective pixels of each scene of optical image, see formula (3):

[0108] Formula (3)

[0109] Where Green and NIR are the green band and near-infrared band reflectance of the current block, respectively. Invalid pixels are assigned a value of -999 to avoid participation in subsequent calculations.

[0110] Subsequently, for each pixel location in Sentinel-1, the NDWI values ​​of all scene images covering that location are collected, and the maximum value is selected to obtain the joint NDWI, as shown in formula (4):

[0111] Formula (4)

[0112] By selecting the maximum NDWI value, the NDWI response with the best optical conditions covering the location is retained, ensuring that the water signal is fully expressed throughout the Sentinel-1 space, while suppressing interference from clouds, shadows, and invalid observations, and weakening the impact of local missing or low-quality optical data. Figure 9 This invention provides an NDWI maximum value image as an embodiment of the invention. Figure 9 The image showing the maximum NDWI value corresponding to Sentinel-1 satellite data is displayed.

[0113] (3) Water body confidence update based on NDWI.

[0114] The effective pixels in the NDWI after the above processing are filtered by numerical range, retaining NDWI values ​​between -0.3 and 0.8 for adaptive threshold calculation. Based on the distribution of the filtered NDWI samples, the optical water body discrimination threshold is automatically calculated using the Otsu's method of maximum inter-class variance, as shown in the following formula:

[0115] Formula (5)

[0116]

[0117] in, To obtain the optimal threshold, For candidate thresholds, The independent variable that corresponds to the value of this variable; When the threshold is set to When, the variance between the two classes (between-class variance); , When the threshold is set to At that time, the proportions (probabilities) of the two categories respectively; , When the threshold is set to At that time, the average value of each of the two categories; : The average value of all pixels.

[0118] Figure 10 This invention provides a statistical chart for obtaining the NDWI threshold using the Otsu thresholding method. Figure 10 The Otsu thresholding method was used to obtain an NDWI threshold of 0.024.

[0119] Subsequently, for pixels with NDWI greater than the adaptive threshold, their water confidence score is updated by taking the maximum value between the original confidence score and the preset high confidence score value of 0.85, as shown in the following formula:

[0120] Formula (6)

[0121] The confidence level of the water body after updating by formula (6) is given.

[0122] Figure 11 An updated water body confidence map is provided as an embodiment of the present invention. Figure 11 exist Water confidence map updated using NDWI thresholds (Based on) ).

[0123] This step involves using adaptive thresholding to identify water information in optical data, updating and improving the confidence of pixels that strongly support the optical water signal, thus providing strong evidence for multi-source data fusion.

[0124] Step S4: Extract terrain undulation features based on the digital elevation model data and construct a terrain constraint mask.

[0125] For example, Gaussian smoothing filtering is applied to digital elevation model data to obtain the regional background topographic surface;

[0126] The absolute value of the difference between the digital elevation model and the background terrain surface is calculated as the terrain undulation amount of each pixel;

[0127] Pixels with terrain undulation less than a preset terrain undulation threshold are marked as flat terrain areas, and a Boolean terrain constraint mask is generated.

[0128] For example, in the land surface water body information extraction method based on multi-source remote sensing data fusion described in this invention, to reduce misjudgment of water bodies caused by factors such as ridges and shadows in complex terrain areas, digital elevation model (DEM) data is used to apply topographic prior constraints to the study area during the water body information extraction process. By analyzing the topographic relief characteristics reflected by the DEM, areas with excessive topographic relief are identified and eliminated, thereby limiting water body extraction to areas with reasonable topographic conditions.

[0129] First, considering the high memory consumption during large-scale remote sensing data processing, this invention employs a block-based processing method to traverse the DEM data. During processing, the DEM data is read block by block according to a preset block size; and an expansion boundary with a preset number of pixels is added around each block to avoid calculation errors at the block edges during subsequent filtering processing; simultaneously, calculations are performed only on the DEM data within the current block range, and intermediate variables are released promptly after processing to ensure the stability of the program.

[0130] Secondly, Gaussian smoothing filtering is applied to the original DEM data to eliminate high-frequency local undulations and obtain a smooth terrain surface reflecting the background topographic trend. The smoothing scale is controlled by the Gaussian filter parameters and can be set according to the topographic characteristics of the study area. Then, the original DEM data and the smoothed DEM data are subtracted, and the absolute value of the difference is taken to obtain the terrain undulation amount reflecting the degree to which the local elevation deviates from the background topography. The greater the intensity of terrain change at a certain pixel location, the larger the terrain undulation amount, indicating that the terrain in that area is steeper and more undulating.

[0131] Finally, a terrain feasibility threshold is determined. Based on the terrain undulation, a terrain undulation threshold is set, and terrain feasibility is determined for each pixel. When the terrain undulation of a pixel is less than the preset threshold, the pixel is determined to have gentle terrain, meeting the terrain conditions for the presence of water. When the terrain undulation is greater than or equal to the threshold, the pixel is determined to have large terrain undulation, not meeting the terrain conditions for the presence of water. The DEM terrain filtering result is output as a two-dimensional Boolean raster mask: areas with a mask value of True represent gentle terrain conditions and are allowed to participate in subsequent water body determination; areas with a mask value of False represent large terrain undulation and are directly set as non-water areas. This output result is an important prior constraint for subsequent water body determination. Figure 12 This invention provides a generated DEM terrain mask image. Figure 12 The generated DEM terrain mask is shown.

[0132] Step S5: Use the terrain constraint mask to screen the confidence scores of the second water body using effective and statistical samples, and perform adaptive threshold segmentation based on the confidence distribution characteristics of the screened statistical samples to determine the optimal threshold. Use the optimal threshold as the water body determination threshold to classify the effective samples into water and non-water body types to obtain the initial water body mask.

[0133] For example, from the second water body confidence scores, pixels that simultaneously meet the conditions of being located within the effective area of ​​the terrain constraint mask and having a confidence score greater than a preset value are selected as valid samples, and pixels located within a preset confidence analysis interval are selected as statistical samples; the confidence distribution of the statistical samples is adaptively segmented using the maximum inter-class variance method to obtain the optimal threshold as the water body determination threshold.

[0134] In one embodiment, the preset value is 0.3, and the preset reliability analysis interval is 0.5 to 1.0.

[0135] For example, under terrain constraints, the current water body confidence image (i.e., the image updated with NDWI water body confidence) Effective pixel screening is performed, retaining only pixels with a confidence level greater than 0.3, satisfying terrain constraints and having limited values ​​as valid samples, in order to eliminate noise, shadows, false water body responses caused by terrain and outliers, and to ensure the stability and reliability of subsequent threshold statistics.

[0136] Secondly, to improve the robustness of threshold determination, an interval constraint is further applied to the effective confidence samples before adaptive threshold calculation. Only pixels with confidence levels between 0.5 and 1.0 are selected as statistical samples for statistical analysis. This interval corresponds to pixels suspected of being water bodies with evidence of moderate to strong water bodies, thus avoiding the dominant influence of non-water body pixels on threshold estimation in terms of quantity. Based on this, the Otsu method is used to adaptively segment the confidence histogram of the above statistical samples. The optimal threshold is automatically determined by maximizing inter-class variance (equivalent to minimizing intra-class variance). This allows the threshold to fall naturally at the boundary between high-confidence water bodies and transitional or non-water body responses, effectively eliminating interference from shadows and weak responses, avoiding the need for human experience-based threshold settings, and improving the objectivity and reliability of water body determination.

[0137] Finally, an initial water mask is generated. With an optimal threshold ( Using the water body determination threshold as a basis, the valid samples are segmented by thresholding. Pixels in the valid samples with a confidence level greater than or equal to the water body determination threshold are classified as water bodies, while the remaining pixels are classified as non-water bodies, thus obtaining the initial water body mask.

[0138] Step S6: Perform spatial morphological optimization processing on the initial water body mask to obtain the final land surface water body information extraction result.

[0139] For example, morphological closure operations and hole filling are performed on the initial water mask to output the final water extraction results and a continuous water confidence map.

[0140] For example, based on the optimal threshold Tc, the initial water mask can be obtained using formula (7). :

[0141] Formula (7)

[0142] Finally, morphological closure operations and hole-filling were performed on the initial water mask to ensure the integrity of the water object, resulting in the final water extraction results and corresponding confidence maps. Comparison images before and after morphological processing are shown below. Figure 13 As shown (a is before processing, b is after processing), Figure 13 This invention provides a comparison chart of water extraction results before and after morphological treatment, and the final water extraction results and corresponding confidence level chart are shown below. Figure 14 As shown (a is the final water body extraction result, b is the final water body confidence map), Figure 14 This invention provides a final water extraction result and corresponding confidence level diagram for an embodiment of the invention.

[0143] Through the above steps, this invention achieves unified fusion of SAR, optical, and DEM multi-source remote sensing data, and performs adaptive threshold determination based on the statistical distribution of water body confidence. This avoids the reliance on fixed thresholds and simple logical fusion in existing technologies, and significantly improves the stability, reliability, and engineering application capability of water body extraction results under complex surface conditions.

[0144] According to another embodiment, a land surface water body information extraction device based on multi-source remote sensing data is provided, the device comprising:

[0145] The data acquisition module is used to acquire multi-source remote sensing data that covers the target study area and whose spatial locations correspond to each other. The multi-source remote sensing data includes at least synthetic aperture radar image data, multi-scene optical remote sensing image data, and digital elevation model data.

[0146] The synthetic aperture radar confidence construction module is used to construct the first water body confidence of each pixel based on the statistical distribution law of the backscattering characteristics of the synthetic aperture radar image data.

[0147] The optical confidence update module is used to spatially register the multi-scene optical remote sensing images with the synthetic aperture radar images, and enhance and update the first water body confidence based on the water body spectral characteristics of the multi-scene optical remote sensing images to obtain the second water body confidence.

[0148] The terrain constraint module is used to extract terrain undulation features based on the digital elevation model data and construct a terrain constraint mask.

[0149] The threshold adaptive segmentation module is used to use the terrain constraint mask to screen the confidence of the second water body into effective samples and statistical samples, and to perform adaptive threshold segmentation based on the confidence distribution characteristics of the screened statistical samples to obtain the optimal threshold. The optimal threshold is used as the water body determination threshold to classify the effective samples into water body and non-water body types to obtain the initial water body mask.

[0150] The morphological processing module is used to perform spatial morphological optimization processing on the initial water body mask to obtain the final land surface water body information extraction result.

[0151] It is understood that the method steps in the embodiments of the present invention can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0152] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0153] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for extracting land surface water bodies based on multi-source remote sensing data, comprising: Acquire multi-source remote sensing data that covers the target study area and whose spatial locations correspond to each other. The multi-source remote sensing data includes at least synthetic aperture radar image data, multi-scene optical remote sensing image data, and digital elevation model data. Based on the statistical distribution law of the backscattering characteristics of the synthetic aperture radar image data, the first water body confidence level of each pixel is constructed. Spatially register the multi-scene optical remote sensing imagery with the synthetic aperture radar imagery data. Based on the water body spectral characteristics of the multi-scene optical remote sensing imagery, enhance and update the confidence level of the first water body to obtain the confidence level of the second water body. Based on the digital elevation model data, terrain undulation features are extracted and a terrain constraint mask is constructed. The terrain constraint mask is used to screen the confidence scores of the second water body for effective and statistical samples, and adaptive threshold segmentation is performed based on the confidence score distribution characteristics of the screened statistical samples to obtain the optimal threshold. Using the optimal threshold as the water body determination threshold, the effective samples are classified into water body and non-water body types to obtain the initial water body mask; The initial water body mask is subjected to spatial morphological optimization to obtain the final land surface water information extraction result.

2. The method according to claim 1, wherein, The statistical distribution law of backscattering characteristics based on the synthetic aperture radar image data is used to construct the first water body confidence score for each pixel, including: The synthetic aperture radar image data is preprocessed to obtain a backscattering coefficient image; Histogram statistical analysis is performed on the effective pixels in the backscattering coefficient image whose values ​​are within the preset effective scattering range to determine the peak parameters representing the low scattering characteristics of the water body; Based on the peak parameters, a synthetic aperture radar water response function is constructed, and the backscattering coefficient of each pixel is mapped to the first water confidence level in the interval from 0 to 1. The output value of the synthetic aperture radar water response function is assigned a preset weight so that its maximum contribution value is controlled.

3. The method according to claim 2, wherein, The preset effective scattering range is -40dB to -10dB; the expression for the synthetic aperture radar water response function is: in, For pixels The confidence level of the first water body at that location. For pixels Backscattering coefficient at VH polarization For peak parameters, For normalized scaling parameters, To preset weights, This is the interval constraint function.

4. The method according to claim 1, wherein, The step of spatially registering the multi-scene optical remote sensing images with the synthetic aperture radar images, and enhancing and updating the confidence level of the first water body based on the water body spectral characteristics of the multi-scene optical remote sensing images, includes: Based on the spatial range and pixel grid of the synthetic aperture radar image, the green band, near-infrared band and scene classification files of each optical remote sensing image are spatially registered with it. After removing cloud-polluted pixels from the scene classification file, the normalized water index of each scene image is calculated, and the maximum value of the normalized water index of all images covering the pixel is taken as the composite water index. The synthetic water index is subjected to adaptive threshold segmentation to determine the optical discrimination threshold; For pixels whose synthetic water index is greater than the optical discrimination threshold, their water confidence is updated to a value no lower than a preset high confidence value.

5. The method according to claim 4, wherein, The following formula is used to update the confidence level of the water body to a value no lower than the preset high confidence level: in, The confidence level for the second water body. 0.85 represents the confidence level for the first water body, and 0.85 is the preset high confidence level value.

6. The method according to claim 1, wherein, The step of extracting terrain undulation features based on the digital elevation model data and constructing a terrain constraint mask includes: Gaussian smoothing filter is applied to the digital elevation model data to obtain the regional background topographic surface; The absolute value of the difference between the digital elevation model and the background terrain surface is calculated as the terrain undulation amount of each pixel; Pixels with terrain undulation less than a preset terrain undulation threshold are marked as flat terrain areas, and a Boolean terrain constraint mask is generated.

7. The method according to claim 1, wherein, The step of using the terrain-constrained mask to screen effective and statistical samples of the confidence level of the second water body, and performing adaptive threshold segmentation based on the confidence level distribution characteristics of the screened statistical samples, includes: From the confidence scores of the second water body, pixels that simultaneously meet the conditions of being located within the effective area of ​​the terrain constraint mask and having a confidence score greater than a preset value are selected as valid samples, and pixels within the preset confidence analysis interval among the valid samples are selected as statistical samples. The confidence distribution of the statistical samples is adaptively segmented using the maximum inter-class variance method to determine the optimal threshold; Using the optimal threshold as the water body determination threshold, the effective samples are classified into water body and non-water body types to obtain the initial water body mask.

8. The method according to claim 7, wherein, The preset value is 0.3, and the preset reliability analysis interval is 0.5 to 1.

0.

9. The method according to claim 1, wherein, The morphological optimization process for the initial water mask includes performing morphological closing operations and hole filling operations on the initial water mask sequentially.

10. A device for extracting land surface water information based on multi-source remote sensing data, comprising: The data acquisition module is used to acquire multi-source remote sensing data that covers the target study area and whose spatial locations correspond to each other. The multi-source remote sensing data includes at least synthetic aperture radar image data, multi-scene optical remote sensing image data, and digital elevation model data. The synthetic aperture radar confidence construction module is used to construct the first water body confidence of each pixel based on the statistical distribution law of the backscattering characteristics of the synthetic aperture radar image data. The optical confidence update module is used to spatially register the multi-scene optical remote sensing images with the synthetic aperture radar images, and enhance and update the first water body confidence based on the water body spectral characteristics of the multi-scene optical remote sensing images to obtain the second water body confidence. The terrain constraint module is used to extract terrain undulation features based on the digital elevation model data and construct a terrain constraint mask. The threshold adaptive segmentation module is used to use the terrain constraint mask to screen the confidence of the second water body into effective samples and statistical samples, and to perform adaptive threshold segmentation based on the confidence distribution characteristics of the screened statistical samples to obtain the optimal threshold. The optimal threshold is used as the water body determination threshold to classify the effective samples into water body and non-water body types to obtain the initial water body mask. The morphological processing module is used to perform spatial morphological optimization processing on the initial water body mask to obtain the final land surface water body information extraction result.

Images