Water body adaptive extraction method based on multi-source remote sensing data fusion

Abstract

The invention is suitable for the technical field of multi-source data processing, and provides a water body adaptive extraction method based on multi-source remote sensing data fusion, and the method comprises the steps: obtaining multi-source data of a target region, constructing an improved water body index based on optical remote sensing image data, and enhancing the spectral separation degree of a deep water region and a shallow water region; constructing a scattering feature distinguishing rule set based on synthetic aperture radar image data, and separating a water body from typical interference ground features; generating a terrain gradient constraint condition based on the digital elevation model data; generating a fusion data set according to the improved water body index, the scattering feature distinguishing rule set and the terrain gradient constraint condition; dynamically segmenting the fused data set through an adaptive threshold segmentation algorithm to obtain an initial water body extraction result; and performing mixed pixel decomposition and morphological boundary optimization processing on the initial water body extraction result to generate an optimized result, and outputting a water body spatial distribution diagram, thereby effectively improving the water body extraction precision, the adaptability to complex scenes and the anti-interference capability.

Description

Technical Field

[0001] This invention relates to the field of multi-source data processing technology, and in particular to an adaptive water body extraction method based on multi-source remote sensing data fusion. Background Technology

[0002] Water body extraction is a key technology in water resource management, flood monitoring, and ecological environmental protection. Its core objective is to accurately identify the spatial distribution of surface water bodies through remote sensing. With the development of remote sensing technology, multi-source remote sensing data, such as optical, synthetic aperture radar, and digital elevation model (DEM) data, provides rich information dimensions for water body extraction. The spectral characteristics of optical remote sensing data can effectively distinguish water bodies from vegetation, synthetic aperture radar data has all-weather imaging capabilities and can penetrate clouds and fog to obtain water body information, while DEM can reflect the impact of topographic relief on water body distribution. Currently, existing water body extraction methods are mainly based on single optical or radar data, which are easily affected by cloud cover, wind and wave interference, or topographic shadows, leading to missed or incorrect extractions. In addition, existing technologies lack mixed pixel processing, and mixed pixels in medium and low resolution images are not effectively decomposed, resulting in blurred and fragmented water body boundaries.

[0003] In view of this, an adaptive water body extraction method based on multi-source remote sensing data fusion is proposed. Summary of the Invention

[0004] This invention provides an adaptive water body extraction method based on multi-source remote sensing data fusion to solve the problem of insufficient accuracy in water body extraction.

[0005] This invention provides an adaptive water body extraction method based on multi-source remote sensing data fusion, comprising:

[0006] Acquire optical remote sensing image data, synthetic aperture radar image data, and digital elevation model data for the target area;

[0007] An improved water index is constructed based on the optical remote sensing image data to enhance the spectral separation between deep and shallow water areas.

[0008] A set of scattering feature distinction rules is constructed based on the synthetic aperture radar image data to separate water bodies from typical interfering ground features.

[0009] Topographic slope constraints are generated based on the digital elevation model data.

[0010] Based on the improved water index, scattering feature differentiation rule set, and terrain slope constraint, a multi-source feature fusion dataset is generated.

[0011] The multi-source feature fusion dataset is dynamically segmented using an adaptive threshold segmentation algorithm to obtain initial water body extraction results.

[0012] The initial water extraction results are subjected to hybrid pixel decomposition and morphological boundary optimization to generate optimized water extraction results, and the spatial distribution map of the final water extraction results is output.

[0013] Furthermore, the construction of the improved water index based on the optical remote sensing image data includes:

[0014] Extract the blue-green band reflectance and short-wave infrared band reflectance of the optical remote sensing image data;

[0015] Calculate the standardized difference between the blue-green band reflectance and the short-wave infrared band reflectance;

[0016] An improved water index is generated by enhancing the reflectivity difference gradient between deep and shallow water areas through a preset nonlinear transformation function.

[0017] Furthermore, the preset nonlinear transformation function includes:

[0018] When the reflectivity of the blue-green band is higher than that of the short-wave infrared band, the output value increases exponentially with the increase of the reflectivity difference.

[0019] When the reflectivity of the blue-green band is lower than that of the short-wave infrared band, the output value is set to zero to suppress non-aquatic interference.

[0020] Furthermore, the construction of a scattering feature differentiation rule set based on the synthetic aperture radar image data includes:

[0021] Extract the backscattering intensity and texture features from the synthetic aperture radar image data;

[0022] Based on the specular reflection characteristics of water bodies, a backscattering intensity threshold is set to screen potential water body areas;

[0023] Texture feature discrimination rules are established based on the scattering characteristics of typical interfering ground objects to eliminate interference from building and terrain shadows.

[0024] By fusing the backscattering intensity threshold with the texture feature discrimination rule, a set of scattering feature discrimination rules is generated.

[0025] Furthermore, the step of constructing a scattering feature differentiation rule set based on the synthetic aperture radar image data further includes:

[0026] The backscattering intensity threshold is set to satisfy the following condition: when the backscattering intensity of a pixel is lower than the preset threshold, it is determined to be a candidate region for water body.

[0027] The texture feature discrimination rule includes: calculating the homogeneity index of the gray-level co-occurrence matrix in the neighborhood of a pixel, and when the homogeneity index is higher than a preset threshold, it is determined to be building interference and removed.

[0028] Furthermore, the generation of terrain slope constraints based on the digital elevation model data includes:

[0029] Calculate the slope value of the digital elevation model data;

[0030] Slope thresholds are set based on the topographical patterns of natural water body distribution;

[0031] The area with a slope value greater than the slope threshold is marked as a water-independent zone, and a terrain slope mask is generated.

[0032] Terrain slope constraints are constructed based on the terrain slope mask.

[0033] Furthermore, the dynamic segmentation of the multi-source feature fusion dataset using an adaptive threshold segmentation algorithm includes:

[0034] The target area is divided into multiple geomorphic units; the statistical distribution of water feature values ​​in the multi-source feature fusion dataset of each geomorphic unit is calculated;

[0035] The segmentation thresholds for each landform unit are determined based on the statistical distribution.

[0036] Furthermore, determining the segmentation threshold for each landform unit based on the statistical distribution includes:

[0037] Identify the histogram distribution patterns of water body characteristic values ​​within various topographic units;

[0038] When the histogram shows a bimodal distribution, the feature value corresponding to the trough position is used as the initial segmentation threshold;

[0039] The initial segmentation threshold is dynamically adjusted based on the terrain complexity of each landform unit.

[0040] Furthermore, obtaining the initial water extraction results includes:

[0041] Pixels in the multi-source feature fusion dataset whose water feature values ​​are greater than or equal to the segmentation threshold are marked as water bodies;

[0042] Pixels whose water feature values ​​are less than the segmentation threshold are marked as non-water bodies;

[0043] Aggregate all pixels marked as water bodies to generate the initial water body extraction result.

[0044] Furthermore, the process of performing hybrid pixel decomposition and morphological boundary optimization on the initial water extraction result to generate an optimized water extraction result includes:

[0045] Based on the optical remote sensing image data, the water abundance of each pixel in the initial water extraction result is calculated using a linear spectral decomposition model.

[0046] Pixels with water abundance greater than or equal to a preset abundance threshold are marked as water pixels;

[0047] Morphological closing operations are performed on the marked water pixels to fill the internal holes in the water area.

[0048] The results of morphological closing operations are processed by morphological opening operations to eliminate small noise points and generate optimized water extraction results.

[0049] As can be seen from the above technical solutions, the present invention has the following advantages:

[0050] This invention acquires optical remote sensing image data, synthetic aperture radar (SAR) image data, and digital elevation model (DEM) data for the target area. Based on the optical image data, it constructs an improved water body index to enhance the spectral differentiation between deep and shallow water. Based on the SAR image data, it constructs a scattering feature differentiation rule set to effectively separate water bodies from typical interfering features. Based on the DEM data, it generates terrain slope constraints to suppress false extractions such as those caused by mountain shadows. A multi-source feature fusion dataset is generated, and this fusion dataset is dynamically segmented using an adaptive threshold segmentation algorithm to obtain initial water body extraction results. Finally, hybrid pixel decomposition is performed on the initial results to handle boundary uncertainties, and morphological boundary optimization is applied to obtain an optimized high-precision water body spatial distribution map. This invention achieves complementary advantages and synergistic enhancement of multi-source remote sensing information, significantly improving the accuracy of water body extraction, particularly its adaptability and anti-interference capabilities in complex scenarios such as deep water, shallow water, and shadowed water bodies. Attached Figure Description

[0051] Figure 1 This is a schematic flowchart of an embodiment of an adaptive water body extraction method based on multi-source remote sensing data fusion in this invention;

[0052] Figure 2 This is a schematic flowchart of another embodiment of the adaptive water extraction method based on multi-source remote sensing data fusion in this invention;

[0053] Figure 3 This is a schematic flowchart of another embodiment of the adaptive water extraction method based on multi-source remote sensing data fusion in this invention;

[0054] Figure 4This is a schematic flowchart of another embodiment of the adaptive water extraction method based on multi-source remote sensing data fusion in this invention;

[0055] Figure 5 This is a schematic flowchart of another embodiment of the adaptive water extraction method based on multi-source remote sensing data fusion in this invention. Detailed Implementation

[0056] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0057] Example 1

[0058] The implementation method in this embodiment can be implemented in a system, on a server, or on a terminal; no specific limitation is made. The following section will describe the adaptive water extraction method based on multi-source remote sensing data fusion in this application from the perspective of system implementation. Please refer to... Figure 1 The method provided in this application includes the following steps:

[0059] S11. Acquire optical remote sensing image data, synthetic aperture radar image data, and digital elevation model data of the target area;

[0060] In this embodiment, all acquired data must cover the same geographical area and have similar imaging time windows to avoid feature distortion due to environmental changes. The spatial resolution difference between optical and synthetic aperture radar (SAR) image data must be less than 5 times; for example, 10m optical data should be matched with ≤50m SAR image data, and the resolution of digital elevation model (DEM) data should not be lower than that of optical data.

[0061] First, after acquiring the raw optical remote sensing image of the target area, the apparent reflectance is converted to surface reflectance using an atmospheric correction module to eliminate the effects of atmospheric scattering and water vapor absorption. Second, a cloud mask is generated using quality assessment bands to remove contaminated areas with cloud cover >10%. Geometric precision correction is performed using ground control points to ensure a positioning error <0.5 pixels. Reflectance data are extracted from the blue band (B2: 490nm), green band (B3: 560nm), and shortwave infrared band (B11: 1610nm) and resampled to a uniform spatial resolution. This yields preprocessed optical data: a surface reflectance image with controllable cloud cover and accurate radiation.

[0062] After acquiring the raw synthetic aperture radar (SAR) image data, the digital values ​​were converted into backscattering coefficients through radiometric calibration to establish a quantitative relationship with the physical properties of ground features. Next, a Refined Lee filter (5×5 window) was used to suppress speckle noise and increase the equivalent number of views (ENL) to ≥4.0. Terrain correction was performed based on a digital elevation model and a Range-Doppler model to eliminate overlay and shadow distortion. Finally, the image data was spatially aligned with the optical image using mutual information registration, resampled to the same coordinate system and resolution, and the backscattering intensity and texture features of the VV polarization channel were extracted. The resulting preprocessed SAR image data is a backscattering coefficient image with accurate radiometric calibration and eliminated terrain distortion.

[0063] Raw digital elevation model (DEM) data is acquired, and an irregular triangular mesh (ITM) interpolation algorithm is used to fill data gaps, ensuring full coverage of elevation values. Slope maps are calculated based on the D8 algorithm to eliminate local terrain anomalies, with a threshold of slope > 80° considered anomaly. Finally, georegistration is performed using optical imagery as a reference, and bilinear interpolation is used to resample to the same spatial resolution as the optical data, generating a continuous terrain slope field. The resulting preprocessed DEM data is a seamlessly stitched, reliable slope raster map.

[0064] All three types of data have been spatiotemporally registered, forming a standardized dataset with spatial overlay and radial matching, which can be directly used for multi-source feature fusion.

[0065] S12. An improved water index is constructed based on optical remote sensing image data to enhance the spectral separation between deep and shallow water areas;

[0066] This step involves constructing a nonlinearly enhanced improved water index. While preserving the commonalities of water spectral characteristics, it amplifies the reflectance difference gradient between deep and shallow water areas, while forcibly suppressing the response of non-water pixels. This improves the integrity and anti-interference capability of water boundary extraction. The specific implementation process is as follows:

[0067] 1. Extract the blue-green band reflectance and short-wave infrared band reflectance from optical remote sensing image data;

[0068] 2. Calculate the standardized difference between the reflectance of the blue-green band and the reflectance of the short-wave infrared band.

[0069] The surface reflectance in the blue-green band and shortwave infrared band is extracted from the preprocessed optical image and denoted as ρ. bg and ρ swir ; Calculate the standardized outliers between bands: Theoretically, this value ranges from [-1, 1], and the water body is affected by ρ. bg >ρ swir It shows a positive value, but the difference between shallow and deep water areas is insufficient. Typical values ​​are: 0.1-0.3 for shallow water vs. 0.4-0.6 for deep water. S123. An improved water index is generated by enhancing the reflectance difference gradient between deep and shallow water areas through a preset nonlinear transformation function.

[0070] The preset nonlinear transformation functions include:

[0071] 1. When the reflectivity of the blue-green band is higher than that of the short-wave infrared band, the output value increases exponentially with the increase of the reflectivity difference;

[0072] 2. When the reflectivity of the blue-green band is lower than that of the short-wave infrared band, the output value is set to zero to suppress non-aquatic interference.

[0073] Specifically, piecewise functions enhance spectral separation: Where α = 5.0, controlling the exponential growth rate and expanding the deep / shallow water gradient; k = 100, the normalization coefficient, making the output value range [0, 100]. When ρ bg >ρ swir At that time, exponential growth causes a sharp increase in the MWI value in deep water and a slow increase in shallow water, thus improving spectral separation; when ρ bg ≤ρ swir At that time, a forced reset to zero is used to eliminate erroneous responses.

[0074] S13. Construct a set of scattering feature distinction rules based on synthetic aperture radar image data to separate water bodies from typical interfering ground features;

[0075] This step, by coupling physical scattering mechanisms with texture statistical properties, utilizes the inherent physical differences between water specular reflection, building angular reflection, and terrain shadows; it eliminates false detections of buildings in urban areas and false detections of shadows in mountainous areas; and ensures that low-feature characteristics of shallow and turbid water bodies are not filtered out by rules. Please refer to [link to relevant documentation]. Figure 2 The specific implementation process is as follows:

[0076] S131. Extract backscattering intensity and texture features from synthetic aperture radar image data;

[0077] S132. Based on the specular reflection characteristics of water bodies, a backscattering intensity threshold is set to screen potential water body areas;

[0078] S133. Establish texture feature discrimination rules based on the scattering characteristics of typical interfering ground objects to eliminate interference from building and terrain shadows;

[0079] S134. Fuse the backscattering intensity threshold with the texture feature discrimination rule to generate a scattering feature discrimination rule set.

[0080] Among them, the setting of the backscatter intensity threshold satisfies the following: when the backscatter intensity of a pixel is lower than the preset threshold, it is determined to be a candidate area of ​​water body; the texture feature discrimination rule includes: calculating the homogeneity index of the gray-level co-occurrence matrix in the neighborhood of the pixel, and when the homogeneity index is higher than the preset threshold, it is determined to be building interference and removed.

[0081] Specifically, the backscattering coefficient of the VV polarization channel is first extracted. Water exhibits extremely low values ​​due to ingress surface reflection (typical range: -22 to -15 dB), while buildings show >-10 dB due to angular reflection effects. The homogeneity of the 5×5 window is then calculated based on the gray-level co-occurrence matrix. Where i and j are the gray values ​​of two adjacent pixels in the image, P(i,j) is the probability value of position (i,j) in the gray-level co-occurrence matrix, representing the frequency of a pair of pixels with gray values ​​i and j appearing together in a specific direction and at a specific distance, and N is the total number of gray levels in the image; if the pixel pair is concentrated on the diagonal (i≈j), then |i―j| approaches 0, the denominator 1+|i―j|≈1, and the homogeneity value is high, such as the uniform area of ​​water / shadow. It should be noted that microwave specular reflection in water bodies (Hom≈0.4~0.6) results in weak grayscale fluctuations, with P(i,j) concentrated near the diagonal, but slightly diffused due to noise. The regular structure of buildings (Hom<0.35) causes grayscale jumps, with P(i,j) dispersed and |i―j| relatively large. In the terrain shadow (Hom>0.85), the grayscale is highly consistent in the no-echo area, with P(i,j) almost entirely located on the diagonal i=j. This formula, by quantifying local texture characteristics, becomes the core criterion for distinguishing "low backscattering intensity interference objects" by the regular concentration of scattering features.

[0082] S14. Generate terrain slope constraints based on digital elevation model data;

[0083] This step utilizes the physical topographical laws governing the distribution of natural water bodies—that water flows to low-lying areas under gravity and cannot exist stably on steep slopes—to construct topographic slope constraints. This eliminates the possibility of false detections in areas with slopes greater than 20°, thus preserving the true water body areas. Please refer to [link / reference]. Figure 3 The specific implementation process is as follows:

[0084] S141. Calculate the slope value of the digital elevation model data;

[0085] S142. Set the slope threshold according to the topographical laws of natural water body distribution;

[0086] S143. Mark areas with slope values ​​greater than the slope threshold as areas where water cannot exist, and generate a terrain slope mask;

[0087] S144. Construct terrain slope constraints based on terrain slope mask.

[0088] Specifically, the slope value θ of each pixel is calculated using the eight-direction steepest slope method based on the preprocessed digital elevation model data: in and The elevation change rate in the x / y directions is calculated using a 3x3 Sobel convolution kernel. Secondly, a global slope threshold T is set based on hydrogeomorphological statistical laws. slope =20°; where the maximum stable slope of natural still water bodies (lakes, reservoirs) is ≤15°; the maximum existing slope of flowing water bodies (rivers) in mountainous areas is ≤20°; the slope calculation deviation caused by vertical error in digital elevation model data is ≤±3°. Threshold formula: T slope =15°+Δ error =20°; Regions with slope values ​​θ > 20° are marked as areas where water cannot exist, generating a binary terrain slope mask M. slope : Finally, the slope mask is integrated as a Boolean condition into the multi-source fusion rule to form the terrain slope constraint condition.

[0089] S15. Generate a multi-source feature fusion dataset based on the improved water index, scattering feature distinction rule set, and terrain slope constraint.

[0090] Specifically, based on an improved water body index, a scattering feature discrimination rule set, and terrain slope constraints, a fused dataset is constructed through spatial logical operations and feature layer stacking. The optical feature layer normalizes the improved water body index to continuous feature values ​​from 0 to 100; the scattering feature layer binarizes the scattering rule set output; and the terrain constraint layer directly uses a slope mask to determine the presence of water bodies. Finally, a raster dataset with consistent spatial resolution and pixel values ​​ranging from [0, 100] is output, where higher values ​​indicate a greater probability of water bodies.

[0091] S16. The multi-source feature fusion dataset is dynamically segmented using an adaptive threshold segmentation algorithm to obtain the initial water body extraction results;

[0092] This step improves the accuracy of initial water body extraction by using dynamic thresholds driven by geomorphic unit zoning and statistical distribution. Please refer to [link / reference]. Figure 4 The specific implementation process is as follows:

[0093] S161. Divide the target area into multiple geomorphic units; calculate the statistical distribution of water feature values ​​in the multi-source feature fusion dataset of each geomorphic unit;

[0094] S162. Determine the segmentation threshold for each landform unit based on the statistical distribution.

[0095] 1. Identify the histogram distribution patterns of water body characteristic values ​​within each topographic unit;

[0096] 2. When the histogram shows a bimodal distribution, the feature value corresponding to the trough position is used as the initial segmentation threshold;

[0097] 3. The initial segmentation threshold is dynamically adjusted based on the topographic complexity of each landform unit.

[0098] Specifically, based on the principle of consistent landform, the target area is divided into 500m x 500m grid cells, and histogram statistics are performed on the fused dataset within each cell. The frequency distribution of all pixel feature values ​​within the cell is calculated based on topographic relief and land cover type, generating a probability density histogram. A segmentation threshold is adaptively set based on the histogram shape; when the histogram shows a clear bimodal distribution, the trough position V is selected. trongh Set as the initial threshold T base If it is a unimodal distribution, then follow T. base =μ + σ, where μ is the mean and σ is the standard deviation; based on the terrain correction coefficient β = 1 + 0.05 × Ruggedness, the terrain undulation degree Ruggedness = standard deviation of slope within the unit, and the final threshold: T final =T base ×β; Finally, perform segmentation: pixel value ≥ T final Marked as a water body; pixel value <T final , marked as non-water bodies.

[0099] S17. Perform hybrid pixel decomposition and morphological boundary optimization on the initial water extraction results to generate optimized water extraction results and output the final water spatial distribution map of the water extraction results.

[0100] This step separates mixed pixels using spectral decomposition technology and combines morphological operations to optimize the boundary topology, improving the integrity of small water bodies while ensuring boundary smoothness. Please refer to [link to relevant documentation]. Figure 5 The implementation process is as follows:

[0101] S171. Based on optical remote sensing image data, the water abundance of each pixel in the initial water body extraction result is calculated using a linear spectral decomposition model;

[0102] S172. Mark the pixels with water abundance greater than or equal to the preset abundance threshold as water pixels;

[0103] S173. Perform morphological closing operations on the marked water pixels to fill the holes inside the water area;

[0104] S174. Perform morphological opening operation on the result after morphological closing operation to eliminate small noise points and generate optimized water extraction results.

[0105] Specifically, based on the band reflectance of optical images, the water abundance A of each pixel is calculated using a linear spectral decomposition model (LSMM). w : (λ = blue, green, red, near-infrared) Extracting pure water end-members from deep water areas Background soil end-members were selected from adjacent bare soil areas. Least squares fitting A w ∈[0,1](A w =1 represents pure water). Let A be an example. w ≥0.8 is the threshold for water body identification; when A w At a value ≥0.8, the pixel spectral characteristics are dominated by water bodies, preserving fragmented water bodies while suppressing false detections of soil moisture. A morphological closing operation of 3×3 structuring elements is performed on the binary water body map: Closing operation = dilation·erosion (dilation first, then erosion), filling internal holes ≤3 pixel widths and connecting water body fractures ≤2 pixel spacing; a morphological opening operation of 3×3 structuring elements is performed: Opening operation = erosion·dilation (erosion first, then dilation), eliminating small isolated points (such as false detection ripples caused by wind and waves) and smoothing jagged boundaries (such as irregular burrs on lake shores), ultimately outputting a water body spatial distribution map with a well-structured topology.

[0106] The above embodiments achieve complementary advantages and synergistic enhancement of multi-source remote sensing information, including spectral features, radar scattering features, and topographic features. This improves the accuracy of water body extraction, particularly its adaptability and anti-interference capabilities in complex scenarios such as deep water, shallow water, and shaded water bodies.

[0107] It is understood that those skilled in the art can combine various implementation methods in the above embodiments under the guidance of the above examples to obtain technical solutions with multiple implementation methods.

[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A water body adaptive extraction method based on multi-source remote sensing data fusion, characterized in that, include: Acquire optical remote sensing image data, synthetic aperture radar image data, and digital elevation model data for the target area; An improved water index is constructed based on the optical remote sensing image data to enhance the spectral separation between deep and shallow water areas. A set of scattering feature distinction rules is constructed based on the synthetic aperture radar image data to separate water bodies from typical interfering ground features. Topographic slope constraints are generated based on the digital elevation model data. Based on the improved water index, scattering feature differentiation rule set, and terrain slope constraint, a multi-source feature fusion dataset is generated. The multi-source feature fusion dataset is dynamically segmented using an adaptive threshold segmentation algorithm to obtain initial water body extraction results. The initial water extraction results are subjected to hybrid pixel decomposition and morphological boundary optimization to generate optimized water extraction results, and the spatial distribution map of the final water extraction results is output.

2. The adaptive water body extraction method based on multi-source remote sensing data fusion according to claim 1, characterized in that, The construction of the improved water index based on the optical remote sensing image data includes: Extract the blue-green band reflectance and short-wave infrared band reflectance of the optical remote sensing image data; Calculate the standardized difference between the blue-green band reflectance and the short-wave infrared band reflectance; An improved water index is generated by enhancing the reflectivity difference gradient between deep and shallow water areas through a preset nonlinear transformation function.

3. The adaptive water body extraction method based on multi-source remote sensing data fusion according to claim 2, characterized in that, The preset nonlinear transformation function includes: When the reflectivity of the blue-green band is higher than that of the short-wave infrared band, the output value increases exponentially with the increase of the reflectivity difference. When the reflectivity of the blue-green band is lower than that of the short-wave infrared band, the output value is set to zero to suppress non-aquatic interference.

4. The adaptive water body extraction method based on multi-source remote sensing data fusion according to claim 1, characterized in that, The construction of a scattering feature differentiation rule set based on the synthetic aperture radar image data includes: Extract the backscattering intensity and texture features from the synthetic aperture radar image data; Based on the specular reflection characteristics of water bodies, a backscattering intensity threshold is set to screen potential water body areas; Texture feature discrimination rules are established based on the scattering characteristics of typical interfering ground objects to eliminate interference from building and terrain shadows. By fusing the backscattering intensity threshold with the texture feature discrimination rule, a set of scattering feature discrimination rules is generated.

5. The adaptive water body extraction method based on multi-source remote sensing data fusion according to claim 4, characterized in that, The step of constructing a scattering feature differentiation rule set based on the synthetic aperture radar image data further includes: The backscattering intensity threshold is set to satisfy the following condition: when the backscattering intensity of a pixel is lower than the preset threshold, it is determined to be a candidate region for water body. The texture feature discrimination rule includes: calculating the homogeneity index of the gray-level co-occurrence matrix in the neighborhood of a pixel, and when the homogeneity index is higher than a preset threshold, it is determined to be building interference and removed.

6. The adaptive water body extraction method based on multi-source remote sensing data fusion according to claim 1, characterized in that, The generation of terrain slope constraints based on the digital elevation model data includes: Calculate the slope value of the digital elevation model data; Slope thresholds are set based on the topographical patterns of natural water body distribution; The area with a slope value greater than the slope threshold is marked as a water-independent zone, and a terrain slope mask is generated. Terrain slope constraints are constructed based on the terrain slope mask.

7. The adaptive water body extraction method based on multi-source remote sensing data fusion according to claim 1, characterized in that, The step of dynamically segmenting the multi-source feature fusion dataset using an adaptive threshold segmentation algorithm includes: The target area is divided into multiple geomorphic units; the statistical distribution of water feature values ​​in the multi-source feature fusion dataset of each geomorphic unit is calculated; The segmentation thresholds for each landform unit are determined based on the statistical distribution.

8. The adaptive water body extraction method based on multi-source remote sensing data fusion according to claim 7, characterized in that, The step of determining the segmentation threshold for each landform unit based on the statistical distribution includes: Identify the histogram distribution patterns of water body characteristic values ​​within various topographic units; When the histogram shows a bimodal distribution, the feature value corresponding to the trough position is used as the initial segmentation threshold; The initial segmentation threshold is dynamically adjusted based on the terrain complexity of each landform unit.

9. The adaptive water body extraction method based on multi-source remote sensing data fusion according to claim 7 or 8, characterized in that, The process of obtaining the initial water extraction results includes: Pixels in the multi-source feature fusion dataset whose water feature values ​​are greater than or equal to the segmentation threshold are marked as water bodies; Pixels whose water feature values ​​are less than the segmentation threshold are marked as non-water bodies; Aggregate all pixels marked as water bodies to generate the initial water body extraction result.

10. The adaptive water body extraction method based on multi-source remote sensing data fusion according to claim 1, characterized in that, The process of performing hybrid pixel decomposition and morphological boundary optimization on the initial water extraction results to generate optimized water extraction results includes: Based on the optical remote sensing image data, the water abundance of each pixel in the initial water extraction result is calculated using a linear spectral decomposition model. Pixels with water abundance greater than or equal to a preset abundance threshold are marked as water pixels; Morphological closing operations are performed on the marked water pixels to fill the internal holes in the water area. The results of morphological closing operations are processed by morphological opening operations to eliminate small noise points and generate optimized water extraction results.

Images