A wetland information extraction method and system based on multi-source remote sensing data fusion

Abstract

This invention relates to the field of remote sensing image processing technology, specifically to a method and system for wetland information extraction based on multi-source remote sensing data fusion. The method includes the following steps: acquiring and fusing wetland optical radar images and water level time-series data; extracting near-infrared and backscattering multi-source feature parameters to distinguish water body pixels; statistically aggregating spatial connectivity to generate water body patch boundaries; calculating a comprehensive model of area, perimeter, and shape index to classify wetland types; adjusting mismatched labels by combining vegetation cover and soil moisture; and forming a wetland information extraction result map. In this invention, the differences in water bodies are enhanced through joint reflection and scattering features; the differences between water bodies and their surroundings are enhanced through joint reflection and scattering characterization; boundary aliasing is suppressed through bidirectional threshold constraints; patch spatial integrity is restored through connectivity analysis; the sensitivity of type classification is enhanced through morphological parameter combinations; and the accuracy and type stability of wetland representation are improved through ecological attribute consistency correction.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing technology, and in particular to a method and system for extracting wetland information by fusing multi-source remote sensing data. Background Technology

[0002] Remote sensing image processing technology involves the analysis and processing of ground information acquired by remote sensing equipment to extract valuable information. It is widely used in environmental monitoring, resource surveys, disaster prediction, and other fields. Core aspects of this field include remote sensing image preprocessing, feature extraction, classification and recognition, change detection, and data fusion. Remote sensing image processing technology achieves accurate description and monitoring of surface targets and phenomena by effectively integrating and analyzing remote sensing data from different types and sources. Its key technologies include image correction, image registration, image fusion, target recognition, and information extraction, providing technical support for applications such as geographic information systems, environmental protection, and agricultural resource management.

[0003] Traditional wetland information extraction methods based on multi-source remote sensing data fusion refer to techniques for extracting wetland information using remote sensing data from different sources. This method primarily enhances the identification and extraction accuracy of wetland areas by fusing multiple types of remote sensing images. Traditional methods employ techniques such as image registration, image fusion, and feature selection to synthesize images with high spatial or temporal resolution from remote sensing data from different sensors or at different times, thereby identifying and classifying wetland areas. Common techniques include pixel-based fusion methods, feature-based fusion methods, and decision-level-based fusion methods. By combining the advantages of different data sources, this approach attempts to address the problems of information loss or noise interference inherent in single-source data sources.

[0004] Traditional multi-source wetland extraction relies on linear registration and pixel-level fusion between images. Optical reflection and radar scattering are prone to spatial misalignment due to scale and temporal differences. The spectral and scattering performance of water body edges and shallow areas is affected by noise disturbances, resulting in aliasing. Threshold discrimination is not stable enough in scenarios with multiple types of water bodies. The water body connectivity structure is often fragmented in complex wetland environments. Patch morphology lacks continuous expression. Wetland type classification relies on single image information and cannot integrate structural features and surrounding ecological attributes. Type labeling is prone to shift in transition areas, leading to fragmentation of wetland range expression and insufficient reliability of type resolution. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of traditional multi-source wetland extraction, which relies on linear registration and pixel-level fusion between images. These shortcomings include the tendency for optical reflection and radar scattering to spatially misalign under scale and temporal differences; the aliasing of spectral and scattering characteristics at water edges and shallow areas due to noise disturbances; insufficient stability of threshold discrimination in scenarios with multiple water body types; the frequent fragmentation of water connectivity structures in complex wetland environments; the lack of continuous expression of patch morphology; wetland type classification relying on single image information, failing to integrate structural features and surrounding ecological attributes; and the tendency for type labeling to shift in transition zones. These shortcomings result in fragmented wetland extent representation and insufficient reliability of type resolution. Therefore, this invention proposes a wetland information extraction method and system based on multi-source remote sensing data fusion.

[0006] To achieve the above objectives, this invention employs a wetland information extraction method based on multi-source remote sensing data fusion, comprising the following steps:

[0007] S1: Acquire optical images, radar images, and water level and height time-series data of wetland areas, align the optical images and radar images spatially, and use a band fusion algorithm to superimpose and fuse multispectral band data and backscattering coefficients according to pixel position to generate a multi-source fused image dataset.

[0008] S2: Call the multispectral reflectance and backscattering coefficient of the pixels in the multi-source fusion image dataset, extract the near-infrared band reflectance and radar backscattering coefficient, identify water body pixels and non-water body pixels based on the dual features of water body combined with the water body feature judgment benchmark value, and generate a wetland water body distribution mask.

[0009] S3: Call the pixel positions marked as water bodies in the wetland water body distribution mask, count the spatial connectivity of water body pixels, calculate the Euclidean distance between adjacent water body pixels, merge water body pixels into water body patches, extract boundary pixel coordinates, and form a wetland water body boundary vector set.

[0010] S4: Call the boundary coordinates of the water body patches in the wetland water body boundary vector set, calculate the area, perimeter and shape complexity index of the water body patches, classify them into river type, lake type and marsh type wetlands through multi-parameter fusion discrimination model, and output the wetland type classification result map.

[0011] As a further embodiment of the present invention, the multi-source fusion image dataset includes a spectral reflectance feature set, a radar scattering coefficient feature set, and a land cover attribute feature set; the wetland water body distribution mask includes a water body indicator raster set, a water body classification raster set, and a water body spatial region set; the wetland water body boundary vector set includes a boundary node set, a boundary line set, and a boundary topology set; and the wetland type classification result map includes a wetland type attribute layer, a wetland type distribution layer, and a wetland zoning layer.

[0012] As a further aspect of the present invention, the specific steps of S1 are as follows:

[0013] S111: Based on the pixel spatial coordinates of optical and radar images of wetland areas, the row and column indices of the two types of images are called and the corresponding coordinate values ​​are compared with the image coordinate reference. Based on the offset difference, the image coordinates are corrected by raster position and a photoradar alignment matrix is ​​generated.

[0014] The image coordinate reference is set according to the UTM projection coordinate system of the WGS84 ellipsoid;

[0015] S112: Based on the pixel correspondence of the photoradar alignment matrix, the multispectral band values ​​of the optical image and the backscattering coefficient of the radar image are called. For the same pixel, the two types of values ​​are aggregated and compared according to the spectral reference value and serialized to obtain the pixel fusion feature frame.

[0016] The spectral reference value is set based on the statistical average reflectance of typical wetland vegetation in the near-infrared band.

[0017] S113: For the sequence structure of the pixel fusion feature frames, match the row and column indices of the sequence with the regional grid index, and map the feature frames to the corresponding geographic raster cells according to the grid index to generate a multi-source fusion image dataset.

[0018] As a further aspect of the present invention, the specific steps of S2 are as follows:

[0019] S211: Based on the pixel multispectral reflectance in the multi-source fusion image dataset, obtain near-infrared band reflectance data, use pixel coordinates and spectral values ​​as the basis, perform spectral data extraction, and use the image pixel position relationship to extract the near-infrared reflectance value of the corresponding pixel to generate a near-infrared reflectance matrix.

[0020] S212: Based on the near-infrared reflectivity matrix and radar backscattering coefficient, a joint threshold determination operation is performed on the two types of data. Combining the near-infrared reflectivity value and the radar backscattering value, the following formula is used:

[0021] ;

[0022] Calculate the dual-feature decision histogram values, determine water body and non-water body pixels based on the water body feature judgment benchmark value, and generate a pixel water body judgment matrix. The histogram represents the decision value of the dual features of the pixel in the i-th row and j-th column. This represents the value of the k-th item in the near-infrared reflectance sequence of the i-th pixel. This represents the value of the k-th item in the near-infrared reflectance sequence of the j-th pixel. The radar backscattering coefficient represents the pixel in the i-th row and j-th column. This represents the scattering increment of the geometric spatial neighborhood of the pixel in the i-th row and j-th column. represents the local background scattering energy value of the i-th row and j-th column, the summation symbol k represents the index count of the pixel spectral sequence, and n represents the length of the pixel near-infrared reflectance sequence;

[0023] The water body feature determination benchmark value is a determination threshold for distinguishing water body and non-water body pixels, determined by statistical samples based on the combined characteristics of water body's low reflectivity in the near-infrared band and low backscattering in radar images.

[0024] S213: Based on the pixel water body determination matrix, perform a binary mapping operation on each pixel, mark water body pixels as water body codes, mark non-water body pixels as non-water body codes, and generate a wetland water body distribution mask.

[0025] As a further aspect of the present invention, the specific steps of S3 are as follows:

[0026] S311: Based on the pixel positions marked as water bodies in the wetland water body distribution mask, obtain the two-dimensional coordinate matrix of the marked pixels, perform Euclidean distance calculation, determine whether the water body pixels are adjacent, and obtain the pixel adjacency determination coefficient by comparing the adjacency relationship according to the Euclidean distance.

[0027] S312: Call the pixel adjacency determination coefficient, perform connectivity clustering operation on the pixel number, calculate the water body pixel aggregation intensity value based on the spatial position between the pixel and its neighboring pixels, and obtain the water body clustering segment matrix;

[0028] S313: Based on the water body clustering segment matrix, extract the water body connected region pixels, perform neighborhood topology detection, determine the boundary pixels and perform contour tracking in spatial order, close the curve and perform unified vector expression to obtain the wetland water body boundary vector set.

[0029] As a further aspect of the present invention, the specific steps of S4 are as follows:

[0030] S411: Based on the boundary coordinates of water patches in the wetland water body boundary vector, obtain the boundary coordinate sequence of each water patch, calculate the difference vector between adjacent coordinate points, use the cross product method of the difference vectors of coordinate points to calculate the area, combine the difference between the first and last coordinate points to perform closure correction, and establish the water patch area matrix.

[0031] S412: Based on the water patch area matrix, extract the Euclidean distance between any coordinate point in the water patch boundary coordinate sequence and the next coordinate point in the sequence, calculate the ratio of the coordinate sequence length to the distance sequence, and generate a set of morphological complexity parameters.

[0032] S413: Call the aforementioned morphological complexity parameter set and water patch area matrix, and classify the two data items based on the multi-parameter fusion discrimination model, labeling them as riverine wetlands, lake wetlands, and marsh wetlands respectively, to obtain a wetland type classification result map.

[0033] As a further aspect of the present invention, the method further includes step S5:

[0034] S5: Call the type label and spatial location of water patches in the wetland type classification result map, combine vegetation coverage data and soil moisture content data, calculate the vegetation type distribution probability in the surrounding buffer zone, adjust the wetland type label of mismatched areas, and form a wetland information extraction result map.

[0035] The wetland information extraction result map includes a wetland type adjustment layer, a wetland ecological attribute layer, and a wetland spatial consistency layer.

[0036] As a further aspect of the present invention, the specific steps of S5 are as follows:

[0037] S511: Based on the type label and spatial location of water patches in the wetland type classification result map, call the vegetation coverage data and soil moisture content data, perform difference comparison on the coverage value and moisture content value in the buffer and fit the value distribution sequence, and perform segment aggregation to generate segment aggregation sequence;

[0038] S512: Based on the segment aggregation sequence, call the vegetation coverage data and soil moisture data, monitor the values ​​in the buffer zone around the water patch, and perform segment normalization calculation based on the segment values ​​in the segment aggregation sequence to obtain the segment normalization matrix.

[0039] S513: Based on the segment normalization matrix, call the water patch type label in the wetland type classification result map, compare the corresponding label values ​​of the probability values ​​in the matrix with the vegetation coverage data and soil moisture content data, and adjust the mismatched labels to obtain the wetland information extraction result map.

[0040] The wetland information extraction system based on multi-source remote sensing data fusion is used to execute the aforementioned wetland information extraction method based on multi-source remote sensing data fusion. The system includes:

[0041] The data fusion module acquires optical images, radar images, and water level height time series data of wetland areas, aligns the optical images and radar images spatially, and uses a band fusion algorithm to superimpose and fuse multispectral band data and backscattering coefficients according to pixel positions to generate a multi-source fused image dataset, which is then transmitted to the optical radar discrimination module.

[0042] The optical radar discrimination module calls the multispectral reflectance and backscattering coefficient of the pixels in the multi-source fusion image dataset, extracts the near-infrared band reflectance and radar backscattering coefficient, identifies water body pixels and non-water body pixels based on the dual characteristics of water body combined with the water body feature judgment benchmark value, generates a wetland water body distribution mask, and transmits it to the water body connectivity analysis module.

[0043] The water connectivity analysis module calls the pixel positions marked as water bodies in the wetland water body distribution mask, counts the spatial connectivity of water body pixels, calculates the Euclidean distance between adjacent water body pixels, merges water body pixels into water body patches, extracts boundary pixel coordinates, forms a wetland water body boundary vector set, and transmits it to the wetland structure classification module.

[0044] The wetland structure classification module calls the boundary coordinates of water body patches in the wetland water body boundary vector set, calculates the area, perimeter and shape complexity index of the water body patches, classifies them into river type, lake type and marsh type wetlands through a multi-parameter fusion discrimination model, outputs the wetland type classification result map, and transmits it to the wetland label calibration module;

[0045] The wetland label calibration module calls the type label and spatial location of water patches in the wetland type classification result map, combines vegetation coverage data and soil moisture content data, calculates the distribution probability of vegetation type in the surrounding buffer zone, adjusts the wetland type label of mismatched areas, and forms a wetland information extraction result map.

[0046] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0047] In this invention, the differences between water bodies and surrounding land features are enhanced by constructing a joint reflection and scattering description of fused images. Boundary aliasing is suppressed by bidirectional threshold constraints on water body features. The connectivity structure of water bodies is restored in spatial proximity to maintain the integrity of patch morphology. The structural sensitivity of type classification is enhanced by the joint characterization of area perimeter and morphological index. Type deviation is corrected by buffer consistency correction of vegetation cover and soil moisture content. This forms a wetland expression that matches environmental conditions and improves the continuity of range and the stability of type identification. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the workflow of the present invention;

[0049] Figure 2 This is a flowchart illustrating the acquisition process of the multi-source fusion image dataset in this invention.

[0050] Figure 3 This is a flowchart illustrating the process of obtaining the wetland water distribution mask in this invention.

[0051] Figure 4This is a flowchart illustrating the process of obtaining the wetland water body boundary vector set in this invention.

[0052] Figure 5 This is a flowchart illustrating the process of obtaining the wetland type classification result map in this invention.

[0053] Figure 6 This is a flowchart illustrating the process of obtaining the wetland information extraction result map in this invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0055] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0056] Example 1:

[0057] Please see Figure 1 This invention provides a method for extracting wetland information from multi-source remote sensing data fusion, comprising the following steps:

[0058] S1: Acquire optical images, radar images, and water level and height time-series data of wetland areas, align the optical images and radar images spatially, and use a band fusion algorithm to superimpose and fuse multispectral band data and backscattering coefficients according to pixel position to generate a multi-source fused image dataset.

[0059] S2: Call the multispectral reflectance and backscattering coefficient of pixels in the multi-source fusion image dataset, extract the near-infrared band reflectance and radar backscattering coefficient, identify water body pixels and non-water body pixels based on the dual features of water body and the water body feature judgment benchmark value, and generate a wetland water body distribution mask.

[0060] S3: Call the pixel positions marked as water bodies in the wetland water body distribution mask, count the spatial connectivity of water body pixels, calculate the Euclidean distance between adjacent water body pixels, merge water body pixels into water body patches, extract the boundary pixel coordinates, and form a wetland water body boundary vector set.

[0061] S4: Call the boundary coordinates of water body patches in the wetland water body boundary vector set, calculate the area, perimeter and shape complexity index of the water body patches, classify them into river type, lake type and marsh type wetland through multi-parameter fusion discrimination model, and output wetland type classification result map;

[0062] S5: Call the type label and spatial location of water patches in the wetland type classification result map, combine vegetation coverage data and soil moisture data, calculate the distribution probability of vegetation type in the surrounding buffer zone, adjust the wetland type label of mismatched areas, and form a wetland information extraction result map.

[0063] The multi-source fusion image dataset includes a spectral reflectance feature set, a radar scattering coefficient feature set, and a land cover attribute feature set. The wetland water body distribution mask includes a water body indicator raster set, a water body classification raster set, and a water body spatial region set. The wetland water body boundary vector set includes a boundary node set, a boundary line set, and a boundary topology set. The wetland type classification result map includes a wetland type attribute layer, a wetland type distribution layer, and a wetland zoning layer. The wetland information extraction result map includes a wetland type adjustment layer, a wetland ecological attribute layer, and a wetland spatial consistency layer.

[0064] Please see Figure 2 The specific steps for obtaining the multi-source fusion image dataset are as follows:

[0065] S111: Based on the pixel spatial coordinates of optical and radar images of wetland areas, the row and column indices of the two types of images are called and the corresponding coordinate values ​​are compared with the image coordinate reference. Based on the offset difference, the image coordinates are corrected by raster position and a photoradar alignment matrix is ​​generated.

[0066] A nature reserve was selected as the monitoring target. Sentinel-2 optical imagery and Sentinel-1 synthetic aperture radar (SAR) imagery of the area in June 2024 were acquired. UTM projection coordinate systems based on the WGS84 ellipsoid were established for both types of imagery. The optical imagery was defined as the reference layer, and the SAR imagery as the layer to be registered. The spatial resolution of all images was set to be resampled to 10m. For any pixel in the optical imagery... Extract its row and column index (r, c). Assuming the index of a certain feature cell in the central region is (5120, 4096), read the geographic coordinate reference of this cell in the metadata. The coordinates are (450230.5, 3205100.0). Simultaneously, based on the orbital and Doppler parameters of the radar imagery, the radar pixel row and column numbers corresponding to the geographical location are obtained through a range-Doppler model. Read the corresponding location in the radar image initial geographic coordinates For (450234.2, 3205102.1), perform a comparison operation between the two types of image coordinate references and calculate the lateral coordinate offset respectively. offset from vertical coordinate Set the offset correction threshold, which is based on the pixel resolution. The offset correction threshold is set to a multiple of 100%. If the calculated Euclidean distance offset Less than the set threshold If the offset is small, it is considered a minor offset. If it exceeds the threshold, polynomial correction based on control points is performed. In this example, the offset is... At the sub-pixel level, calculate its row and column offsets in the raster space, and its lateral raster offset. One pixel, vertical raster offset For each pixel, the radar image coordinates are corrected using the offset difference. The radar pixel values ​​are resampled using bilinear interpolation, aligning the radar pixel center with the optical pixel center. This coordinate comparison and correction process is then performed on all 10000×10000 pixels of the image, constructing a coordinate system with dimensions of... The mapping table, which is the alignment matrix of the LiDAR, records the coordinates of each optical pixel. With the corrected radar cell index A one-to-one mapping relationship.

[0067] S112: Based on the pixel correspondence of the lidar alignment matrix, the multispectral band values ​​of the optical image and the backscattering coefficient of the radar image are called. For the same pixel, the two types of values ​​are aggregated and compared according to the spectral reference value and serialized to obtain the pixel fusion feature frame.

[0068] For a specific pixel location in the wetland water and vegetation transition zone Multispectral band values ​​are retrieved from the optical image band set, primarily extracting the green band, which is sensitive to wetland characteristics. Red band and near-infrared band The reflectivity value, which is quantized after radiometric calibration, is... to The integer data is retrieved, and the same polarization backscattering coefficient is retrieved from the radar imagery. With cross-polarization backscattering coefficient The unit of measurement is decibels. For the same pixel, perform aggregation operations on two types of values ​​to construct an initial feature vector. Set spectral reference value This benchmark value is set based on the statistical average reflectance of typical wetland vegetation in the near-infrared band, for example, setting... And set the radar water scattering reference value. Specific examples of aggregated numerical values ​​are shown in Table 1. Sample point 1 is selected as the current calculation object. The value is , The value is The extracted values ​​are compared with the baseline values, and an arithmetic difference calculation is performed to calculate the spectral difference. Calculate the scattering difference Based on the comparison results, the difference values ​​generated are arranged in a predetermined order with the original band values. A serialization operation is then performed to convert the values ​​into a binary byte stream or a standardized JSON string format, for example, generating a string sequence.

[0069] Band:3200|Scat:-12.5|Diff:200,2.5;

[0070] This sequence is the feature combination frame. The above extraction, comparison and serialization process is repeated for pixels in the entire map to obtain a set of pixel fusion feature frames containing location information and multi-dimensional attribute information.

[0071] Table 1: Example Data Table of Light Radar Feature Aggregation at Wetland Monitoring Points

[0072] ;

[0073] As shown in Table 1, the aggregated values ​​of three typical sample points are listed. Sample 1 represents an area with high vegetation cover, and Sample 2 represents a water body area. The above calculation process is based on the data in the table to generate the corresponding feature combination frame content.

[0074] S113: For the sequence structure of pixel fusion feature frames, match the row and column indices of the sequence with the regional grid index, and map the feature frames to the corresponding geographic raster cells according to the grid index to generate a multi-source fusion image dataset.

[0075] Read the sequence data of sample 1 stored in memory and parse the row and column index information contained in its header. Construct a regional grid indexing system and set the grid size. for Pixel, calculate the grid number to which the current pixel belongs, horizontal grid index. Vertical grid index Generate a region grid index key value "Grid_40_32", assign the pixel feature frame to the corresponding grid data bucket, and perform region grid aggregation based on the matching content. When the number of accumulated pixel feature frames in grid "Grid_40_32" reaches a certain threshold... At this time, initiate data reorganization within the grid, perform raster reconstruction on the aggregated grid, and create a dimension of... ( The serialized feature combination frames are deserialized into numerical matrices based on the local coordinates of the pixels within the grid, using matrix blocks (within the feature dimension). ,in , The feature vectors are filled into the corresponding positions in the matrix, and the grid blocks are processed sequentially. The reconstructed matrix blocks are then arranged according to... and The data are written sequentially to the target file, and the geographic transformation parameters and projection information in the file header are defined. Finally, a multi-source fused image dataset containing complete geospatial information and multi-source feature attributes is generated.

[0076] Please see Figure 3 The specific steps for obtaining the wetland water distribution mask are as follows:

[0077] S211: Based on the pixel multispectral reflectance in the multi-source fusion image dataset, obtain near-infrared band reflectance data, use pixel coordinates and spectral values ​​as the basis, perform spectral data extraction, and use the image pixel position relationship to extract the near-infrared reflectance value of the corresponding pixel to generate a near-infrared reflectance matrix.

[0078] Read the generated multi-source fused image dataset, locate the near-infrared band data layer in the dataset, assuming this band corresponds to the 8th band of the Sentinel-2 image, with a pixel depth of 16-bit unsigned integer. For the entire 10000×10000 raster matrix, traverse each pixel coordinate (r, c) in row-major order, select the pixel with coordinates (5120, 4096) as the target object, retrieve its spectral vector data, and extract the near-infrared band reflectance value. Simultaneously, the reflectance values ​​of pixels within its 3×3 neighborhood are extracted to construct a local spectral matrix. If multi-temporal data is used, the reflectance values ​​at that location are read. , , Near-infrared reflectance sequences at three time phases The above extraction operation is performed on all image elements. The extracted values ​​are written into a new memory block according to the original row and column positions. For invalid values ​​or cloud-covered areas, a specific mask value of -9999 is filled. After checking the data integrity, the memory block data is solidified and stored to generate a near-infrared reflectance matrix with the same dimensions as the original image.

[0079] S212: Based on the near-infrared reflectivity matrix and radar backscattering coefficient, a joint threshold determination operation is performed on the two types of data. Combining the near-infrared reflectivity value and the radar backscattering value, the following formula is used:

[0080] ;

[0081] Calculate the dual-feature decision bar value, determine water body and non-water body pixels based on the water body feature judgment benchmark value, and generate a pixel water body judgment matrix.

[0082] in, The histogram represents the decision value of the dual features of the pixel in the i-th row and j-th column. This represents the value of the k-th item in the near-infrared reflectance sequence of the i-th pixel. This represents the value of the k-th item in the near-infrared reflectance sequence of the j-th pixel. The radar backscattering coefficient represents the pixel in the i-th row and j-th column. This represents the scattering increment of the geometric spatial neighborhood of the pixel in the i-th row and j-th column. represents the local background scattering energy value of the pixel in the i-th row and j-th column, the summation symbol k represents the index count of the pixel spectral sequence, and n represents the length of the pixel near-infrared reflectance sequence.

[0083] The water body characteristic judgment benchmark value is a judgment threshold determined by statistical samples or threshold algorithm based on the combined characteristics of water body's low reflectivity in the near-infrared band and low backscattering in radar images.

[0084] The near-infrared reflectance matrix and the radar backscattering coefficient numerical matrix are used. The pixel at coordinates (5120, 4096) is selected as a calculation example. The near-infrared reflectance sequence length is set to n=3, corresponding to three monitoring phases. The near-infrared sequence for this pixel is... The values ​​are [3200, 3150, 3220]. A known water sample point (200, 200) is selected as the reference pixel j, and its near-infrared sequence... The values ​​are [450, 460, 440]. Obtain the radar backscattering coefficient for pixel (5120, 4096). for Convert it to a linear intensity value Calculate the average scattering intensity within the 5×5 neighborhood of this pixel. Calculate the scattering increment in the geometric space neighborhood Calculate the background scattering energy value in the neighborhood. Take the variance of the scattering intensity in the neighborhood, and let... Substitute the above parameters into the formula to obtain the dual-feature decision bar chart. The calculation. In the formula, Representing the Line number The columnar bar value of dual-feature decision is the core indicator for determining water bodies; and These represent the target pixel and the reference pixel at the 1st... The near-infrared reflectance at each time phase is summed, along with the absolute values ​​of the differences between them. Used to quantify the deviation of target pixels from standard water bodies in spectral time series; the summation sign reflects the cumulative difference over the entire time period. For radar scattering intensity, combined with neighborhood increment Divide by background energy The square root of the formula is used to weight the spectral differences using radar texture features. The advantage of this formula lies in weighting the spectral distance using the texture signal-to-noise ratio of radar scattering intensity, thus suppressing low-reflectivity shadow noise while highlighting the low-scattering smoothness of the water body. The numerical calculation is as follows: First, calculate the cumulative spectral difference term:

[0085] ;

[0086] ;

[0087] The second step is to calculate the radar weighting term:

[0088] ;

[0089] The third step is to calculate the final eigenvalues:

[0090] ;

[0091] Setting benchmark values ​​for water body characteristics This threshold is obtained from the statistical layer. A random water sample Value distribution is obtained, and the mean is calculated. Standard deviation ,set up The calculated result 132506.4 was compared with the threshold 7400. The result showed that the dual feature decision value of the target pixel (5120, 4096) was much higher than the water body threshold. The pixel attribute was determined to be non-water body (such as vegetation or building). The pixel water body determination matrix was generated by traversing the entire map.

[0092] Table 2: Parameters for Joint Calculation of Multi-Source Characteristics of Wetlands

[0093] ;

[0094] Table 2 lists the specific parameter values ​​used in the above formula calculations, comparing the differences in multiple parameters between non-water body pixels and typical water body pixels, and the final... The value of non-water bodies is much greater than that of water bodies.

[0095] S213: Based on the pixel water body determination matrix, perform a binary mapping operation on each pixel, mark water body pixels as water body codes, mark non-water body pixels as non-water body codes, and generate a wetland water body distribution mask.

[0096] Based on the pixel-based water body determination matrix, a binarization mapping rule is defined, and water body codes are set. Non-water body coding Iterate through each element in the decision matrix and read the value calculated above. The value for the pixel at coordinates (5120, 4096) is 132506.4, and the comparison condition is... (Assuming the corrected logic is that the smaller the difference from the water body reference value, the more it leans towards the water body, here because...) The value represents the difference distance from the water body reference point; therefore, the smaller the value, the more likely it is a water body. The previous judgment logic should be adjusted to state that values ​​less than a threshold indicate a water body. The test cell (200, 201) was determined not to meet the water quality requirements and was marked as 0. The calculated value for another test cell was... ,satisfy Marked as 1, a single-band bitmap matrix with the same dimension as the original image is constructed in memory. The 0 or 1 after judgment is written to the corresponding row and column positions. Morphological opening operation is used to filter noise in the edge area. The structural element is defined as a 3×3 rectangle. Isolated patches with an area of ​​less than 5 pixels are removed. Finally, a standard GeoTIFF format file is output to generate a wetland water body distribution mask.

[0097] Please see Figure 4 The specific steps for obtaining the wetland water body boundary vector set are as follows:

[0098] S311: Based on the pixel positions marked as water bodies in the wetland water body distribution mask, obtain the two-dimensional coordinate matrix of the marked pixels, perform Euclidean distance calculation, determine whether water body pixels are adjacent, and obtain the pixel adjacency determination coefficient by comparing the adjacency relationship based on the Euclidean distance.

[0099] The binary mask data stored in memory is called, and a raster coordinate system is established with the top left corner of the image as the origin (0, 0). The entire image of 10000×10000 pixels is traversed to retrieve the target set with a pixel value of 1. Assuming that the center pixel located at row and column number (200, 201) is retrieved, the system will retrieve the target set. At the same time, identify its There are two adjacent cells labeled as water bodies in the neighborhood; the cell on the right is the one on the right. and the lower right pixel Perform Euclidean distance calculation, using coordinate differences to calculate the spatial distance between the center pixel and adjacent pixels. ,distance ,for ,distance Adjacency determination threshold This threshold is set based on the diagonal length of the raster cell to determine whether water pixels are adjacent. and ,determination and All are related to the central pixel To establish effective spatial connectivity, adjacency relationships are compared based on Euclidean distance, and adjacency weight coefficients are constructed. The coefficient for directly adjacent (distance 1) is set to 1.0, and the coefficient for diagonally adjacent (distance 1.414) is set to 0.707 (i.e., ...). The above calculation results are stored in a sparse matrix. For pixel pairs that do not meet the threshold condition, the coefficients are set to zero. The pixel pairs of water bodies are traversed to finally obtain the pixel adjacency determination coefficient that records the spatial connection strength of water bodies in the whole map.

[0100] S312: Call the cell adjacency determination coefficient, perform connectivity clustering operation on the cell number, and use the formula based on the spatial position between the cell and its neighboring cells:

[0101] ;

[0102] Calculate the pixel aggregation intensity value of the water body to obtain the water body clustering segment matrix;

[0103] in, This represents the numerical representation of the adjacency determination state of the i-th cell participating in the clustering of the k-th segment. This represents the Euclidean distance difference between the i-th pixel participating in the k-th segment cluster and its neighboring pixels. This indicates the number of pixels in the k-th segment. This represents the spatial sequence index difference of the i-th cell. This represents the aggregate intensity value of the water cell in the k-th segment. This represents the upper bound of the number of cells participating in the clustering. In the summation symbol, i is the cell index identifier and k is the paragraph index identifier.

[0104] Call the cell adjacency determination coefficient to select a connected candidate region containing n=3 cells as the first... The analysis is performed on several clustered segments with cell indices i=1, 2, and 3. Connectivity clustering is performed on the cell numbers, and the required parameters are obtained based on the spatial positions of the cell and its neighbors. Represents the adjacency state of a cell; the value is 1 if the cell is connected to at least one cell in its neighborhood, and 0 otherwise; Parameter This represents the Euclidean distance difference, which is the difference between the distance from the pixel to the cluster center and the standard raster step size (1.0); Parameter The parameter is 3, which is a constant. is the index jump value of a cell in the storage sequence, used to measure the continuity of spatial storage. In the formula, Indicates the first The water pixel aggregation intensity value of a segment is used to quantitatively assess the compactness and confidence level of the water patch; summation term It reflects the total number of connections within a polygon; squaring amplifies the weight of connectivity. This reflects the regularity of the pattern shape; taking the absolute value prevents positive and negative values ​​from canceling each other out; in the denominator... and The combination of these factors normalizes the constraints on patch size and storage dispersion. The advantage of this formula is that, by calculating the ratio of the squared connectivity term in the numerator to the spatial dispersion in the denominator, it effectively enhances the signal strength of high-compact water patches while suppressing noise pixels with discrete linear distributions (such as fine artifacts), thus significantly improving the accuracy of water body extraction.

[0105] Table 3: Examples of Water Body Pixel Clustering Parameter Calculation

[0106] ;

[0107] Referring to Table 3, substitute the experimental data listed in Table 3 into the formula for calculation: First step, calculate the molecule:

[0108] ;

[0109] molecular ;

[0110] The second step is to calculate the denominator:

[0111] ;

[0112] denominator ;

[0113] The third step is to calculate the final result:

[0114] ;

[0115] Set strength screening benchmark value This benchmark value is adjusted downwards after calculating the average intensity of known stable water body patches. The settings are to compare the calculation results, because This result indicates that the first Each cluster segment has high internal aggregation and spatial compactness, meets the water body patch feature requirements, and is identified as a valid water body object. Conversely, if the calculated value is lower than the threshold, it is regarded as noise and is removed. Candidate segments are processed in this way to obtain the water body cluster segment matrix.

[0116] S313: Based on the water body clustering segment matrix, extract the connected region pixels of the water body, perform neighborhood topology detection, determine the boundary pixels and perform contour tracking in spatial order, close the curve and then perform unified vector expression to obtain the wetland water body boundary vector set;

[0117] Read the first after strength filtering For each water body patch data point, corresponding water body connected region pixels are extracted based on the water body connected region segment matrix. Morphological erosion operations are performed on the patches to identify the set of pixels located at the region boundaries, and their raster center point coordinate sequences are extracted. It is assumed that an ordered boundary coordinate sequence is obtained. After extracting the ordered boundary cell sequence, a contour tracing operation is performed, connecting the boundary cells sequentially according to their spatial adjacency to automatically form closed curves. Then, the geographic information system's vectorization engine is invoked to convert the raster center point coordinates of the boundary cells into vector polygon vertices, and polygon boundary segments are drawn sequentially. The projected coordinate system is defined as UTMZone50N for unified vector representation. The area attribute of the generated polygons is calculated, and polygons with areas smaller than a certain threshold are discarded. The tiny, fragmented patches are ultimately output as an EsriShapefile file, yielding a vector set of wetland water body boundaries.

[0118] Please see Figure 5 The specific steps for obtaining the wetland type classification result map are as follows:

[0119] S411: Based on the boundary coordinates of water patches in the wetland water body boundary vector, obtain the boundary coordinate sequence of each water patch, calculate the difference vector between adjacent coordinate points, use the cross product method of the difference vectors of coordinate points to calculate the area, combine the difference between the first and last coordinate points to perform closure correction, and establish the water patch area matrix.

[0120] Access the vector dataset stored in the geographic information database, read the feature IDs from the layer attribute table, and select the feature ID with the specified number. Taking independent water patches as the processing object, the boundary node coordinate sequence stored in their geometric fields is parsed. Assume this sequence contains N=5 key node coordinates (x, y), which are respectively... , , , and closure point The unit is meters. Obtain the boundary coordinate sequence of each water patch and initialize the surface accumulation variable. The algorithm performs the calculation and accumulation of the difference vector between adjacent coordinate points, using the logic of the Shoelace Algorithm. It sequentially calculates the cross product of the coordinates of adjacent nodes. The specific calculation process is as follows: First step: Calculate... The second step is to calculate. The third step is to calculate. Fourth step calculation The summation of the above step-by-step results yields the following result: By combining the difference between the first and last coordinate points for closed-loop correction, the absolute value of the cumulative value is taken and divided by a constant 2, i.e. This value is the geographic projection area of ​​the target patch. Traverse 5000 patch features in the vector layer, repeat the above coordinate analysis and arithmetic accumulation process, write the calculated area value into a double-precision floating-point array, and establish an index mapping with the patch ID to build a water patch area matrix containing the geometric size attributes of all water features in the map.

[0121] S412: Based on the water patch area matrix, extract the Euclidean distance between any coordinate point in the water patch boundary coordinate sequence and the next coordinate point in the sequence, calculate the ratio of the coordinate sequence length to the distance sequence, and generate a set of morphological complexity parameters.

[0122] Read number Area of ​​the patch And the corresponding boundary coordinate sequence, extract the Euclidean distance between the boundary coordinates of the water patch and the adjacent coordinate points, and perform a pairing analysis on the adjacent points in the sequence. and Calculate its spatial straight-line distance Substitute the aforementioned coordinate data and calculate , , , The length of the coordinate sequence (i.e., the perimeter of the patch) is obtained by summing the segmented distances. To quantify the complexity of the morphology, a distance sequence reference value is constructed, and the circumference of a standard circle with the same area as the patch is calculated. Calculate the ratio of the coordinate sequence length to the distance sequence, i.e., the morphological complexity index. The closer the ratio is to The closer the shape is to a circle or a regular polygon, the larger the ratio indicates that the shape is more elongated or the boundaries are more fragmented. For another number... Linear water patches, their area But perimeter Calculate its complexity The results show significant morphological differences. The ratio is calculated by traversing the patches, and the results are serialized and stored to generate a set of morphological complexity parameters.

[0123] S413: Call the morphological complexity parameter set and the water patch area matrix, and classify the two data items based on the multi-parameter fusion discrimination model, labeling them as river wetlands, lake wetlands and marsh wetlands respectively, to obtain the wetland type classification result map;

[0124] The system calls the morphological complexity parameter set and the water patch area matrix, loads a pre-defined multi-parameter fusion discrimination model, which is constructed based on decision tree logic, and sets a morphological threshold. With area threshold Based on wetland geographical characteristics, a threshold for the morphological complexity of riverine wetlands is set. Set an area threshold for lake-type wetlands. And the upper limit of morphological complexity Regarding the aforementioned plaque, its , Execution logic judgment: First, judge ,Right now If the result is false, exclude river type; then determine... ,Right now The result is false, excluding large lake types; finally, it is determined to fall into the middle range, and based on its high regularity (low complexity), it is marked as a pond or small lake type wetland (classified here as a general lake type). plaque, its If it meets the morphological characteristics of a river, it is marked as a riverine wetland; if patches exist... ,area complexity If the wetland falls between the two and does not meet the lake area requirement, it is marked as a marsh-type wetland (transition zone). For specific classification parameter examples, please refer to Table 4. Based on the multi-parameter fusion discrimination model, the two data are classified and judged. The classification codes (such as river=1, lake=2, marsh=3) are written into the vector attribute table. The layer color is rendered according to the code to obtain the wetland type classification result map.

[0125] Table 4: Example Table of Wetland Patch Type Discrimination Parameters and Results

[0126] ;

[0127] Table 4 shows the calculation parameters and final classification results of four typical patches. By using quantitative area and morphological index thresholds for screening, the automatic classification of different wetland types was achieved.

[0128] Please see Figure 6 The specific steps for obtaining the wetland information extraction result map are as follows:

[0129] S511: Based on the type label and spatial location of water patches in the wetland type classification result map, call vegetation cover data and soil moisture content data, perform difference comparison on the cover value and moisture content value in the buffer and fit the value distribution sequence, and perform segment aggregation to generate segment aggregation sequence;

[0130] Read the classification result vector file stored in the geographic information database, traverse the attribute table to obtain the unique identifier and classification attribute of each patch, and select the patch with the number... Taking swamp-type wetland patches as processing examples, the coordinates of their polygonal geometric centers were analyzed. and boundary node sequence, set buffer construction radius parameter The spatial analysis engine is invoked to perform an equidistant expansion operation around the patch boundary, generating a ring-shaped buffer polygon. Vegetation cover data and soil moisture content data are then retrieved, and a full-area vegetation cover (FVC) raster image and a soil moisture content (SM) inversion image are loaded, both at a resolution of 10m. The generated ring-shaped buffer is used as a mask to extract pixel values ​​from the two images. It is assumed that the number of effective pixels extracted within the buffer is... One, obtain the vegetation cover numerical sequence:

[0131] ;

[0132] Obtain soil moisture content numerical sequence The differences between the coverage and moisture content values ​​within the buffer zone are compared and a value distribution sequence is fitted. The feature difference index of each corresponding pixel is then calculated. Calculate the first group of pixels Statistically analyze the differences and sort them in ascending order. Calculate the percentiles of the data, with the quantile nodes as follows: The process involves performing segment aggregation to generate a segment aggregation sequence, and then dividing the difference range into segments based on the quantile nodes. Given a series of continuous feature aggregation intervals, assuming the minimum calculated difference value is... The maximum value is The constructed segment sequence is The sequence structure is then stored in an in-memory hash table for subsequent normalization calls.

[0133] S512: Based on the segment aggregation sequence, call the vegetation coverage data and soil moisture data, monitor the values ​​in the buffer zone around the water body patch, and perform segment normalization calculation based on the segment values ​​in the segment aggregation sequence to obtain the segment normalization matrix.

[0134] Call the memory stored Feature aggregation interval definition corresponding to the patch The system retrieves vegetation cover and soil moisture data, repositions the pixel coordinates within the buffer, reads the original feature values ​​of the pixels, and recalculates the difference index. The values ​​were monitored within the buffer zone surrounding the water patch. For the pixel with coordinates (500, 501), the difference index was calculated as follows: Based on the values ​​of the segments in the segment aggregation sequence, traverse... Multiple intervals in the data, determine Falling into the second interval Determine the lower bound of the interval. Upper Realm The segment normalization calculation is performed using a local linear mapping algorithm, and the calculation formula is as follows: Substitute into numerical calculation This value represents the relative position probability of a pixel within its respective feature segment. For a difference index of , The pixel falls into the fifth interval. ,calculate The above calculation is completed by traversing 25 pixels in the buffer. The normalized values ​​obtained are written into a two-dimensional array according to the spatial distribution of pixels to obtain the segment normalization matrix.

[0135] S513: Based on the segment normalization matrix, call the water patch type label in the wetland type classification result map, compare the probability value in the matrix with the corresponding label value of vegetation coverage data and soil moisture content data, and adjust the mismatched labels to obtain the wetland information extraction result map.

[0136] Read the normalized probability values ​​stored in the matrix, calculate the global average probability value of the matrix (0.62), and retrieve the water patch type labels from the wetland type classification result image to obtain... The current label for the patch is "marshland," and the feature matching baseline interval for marshland is set as follows: The probability values ​​in the matrix are compared with the corresponding label values ​​of vegetation cover data and soil moisture content data, and mismatched labels are adjusted. The calculated values ​​are then used to determine the correct values. Compare with the baseline interval to determine If the result is a match, the original label is retained; if the target is another patch... The current label is "lake-type wetland", and the calculated normalized mean of the segment is The baseline range requirement for lacustrine wetlands is... (Indicating a significant difference between the water body and its surrounding environment), judgment If a mismatch occurs between tag attributes and environmental characteristics, an adjustment mechanism is triggered, based on... The alternative types corresponding to the numerical range (such as seasonal waterlogging areas or high-humidity meadows) will be The label is changed to "non-wetland" or "transition zone". The specific correction parameters are shown in Table 5. This verification logic is executed by traversing all patches in the map, updating the Type field in the vector attribute table, and finally rendering the output to obtain the wetland information extraction result map.

[0137] Table 5: Comparison Table of Correction Parameters for Wetland Information Extraction

[0138] ;

[0139] Table 5 lists the verification data for four typical patches. and Due to environmental characteristic indicators If a data point does not fall within the preset reasonable range, the system automatically identifies and corrects the classification result, thereby eliminating the classification error.

[0140] A wetland information extraction system based on multi-source remote sensing data fusion is used to execute the aforementioned wetland information extraction method based on multi-source remote sensing data fusion. The system includes:

[0141] The data fusion module acquires optical images, radar images, and water level height time series data of wetland areas, aligns the optical images and radar images spatially, and uses a band fusion algorithm to superimpose and fuse multispectral band data and backscattering coefficients according to pixel positions to generate a multi-source fused image dataset, which is then transmitted to the optical radar discrimination module.

[0142] The optical radar discrimination module calls the multispectral reflectance and backscattering coefficient of pixels in the multi-source fusion image dataset, extracts the near-infrared band reflectance and radar backscattering coefficient, identifies water body pixels and non-water body pixels based on the dual characteristics of water body combined with the water body feature judgment benchmark value, generates a wetland water body distribution mask, and transmits it to the water body connectivity analysis module.

[0143] The water connectivity analysis module calls the pixel positions marked as water bodies in the wetland water body distribution mask, counts the spatial connectivity of water body pixels, calculates the Euclidean distance between adjacent water body pixels, merges water body pixels into water body patches, extracts the boundary pixel coordinates, forms a wetland water body boundary vector set, and transmits it to the wetland structure classification module.

[0144] The wetland structure classification module calls the boundary coordinates of water body patches in the wetland water body boundary vector set, calculates the area, perimeter and shape complexity index of the water body patches, and classifies them into river type, lake type and marsh type wetlands through a multi-parameter fusion discrimination model, outputs the wetland type classification result map, and passes it to the wetland label calibration module;

[0145] The wetland label calibration module calls the type labels and spatial locations of water patches in the wetland type classification result map, combines vegetation coverage data and soil moisture content data, calculates the distribution probability of vegetation type in the surrounding buffer zone, adjusts the wetland type labels of mismatched areas, and forms a wetland information extraction result map.

[0146] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for extracting wetland information from multi-source remote sensing data fusion, characterized in that, The method comprises the following steps: S1: acquiring optical image, radar image and water level time series data of wetland area, aligning the spatial positions of the optical image and the radar image, superimposing and fusing the multi-spectral band data and the backscattering coefficient according to the pixel positions by using a band fusion algorithm to generate a multi-source fusion image data set; The specific steps of S1 are: S111: based on the pixel spatial coordinates of the optical image and the radar image of the wetland area, calling the row and column indexes of the two types of images and comparing the corresponding coordinate values with the image coordinate reference, performing grid position correction on the image coordinates based on the offset difference, and generating an optical-radar alignment matrix; S112: according to the pixel correspondence relationship of the optical-radar alignment matrix, calling the multi-spectral band values of the optical image and the backscattering coefficient of the radar image, performing aggregation on the two types of values for the same pixel, comparing according to the spectral reference value, and performing serialization processing to obtain a pixel fusion feature frame; S113: for the sequence structure of the pixel fusion feature frame, matching the row and column indexes of the sequence with the regional grid index, mapping the feature frame to the corresponding geographical grid unit according to the grid index, and generating a multi-source fusion image data set; S2: calling the multi-spectral reflectivity and backscattering coefficient of the pixels in the multi-source fusion image data set, extracting the near-infrared band reflectivity and the radar backscattering coefficient, identifying water body pixels and non-water body pixels according to the water body double feature combined with the water body feature judgment reference value, and generating a wetland water body distribution mask; The specific steps of S2 are: S211: based on the multi-spectral reflectivity of the pixels in the multi-source fusion image data set, acquiring near-infrared band reflectivity data, taking pixel coordinates and spectral values as the basis, performing spectral data extraction, using the image pixel position relationship, extracting the near-infrared reflectivity value of the corresponding pixel, and generating a near-infrared reflectivity matrix; S212: based on the near-infrared reflectivity matrix and the radar backscattering coefficient, performing joint threshold judgment operation on the two types of data, combining the near-infrared reflectivity value and the radar scattering value, and using the formula: ； The double-feature decision column value is calculated, the water body and non-water body pixels are determined according to the water body feature determination criterion value, and a pixel water body determination matrix is generated, wherein, represents the double-feature decision column value of the i-th row j-th column pixel, represents the k-th value in the near-infrared reflectivity sequence of the i-th pixel, represents the k-th value in the near-infrared reflectivity sequence of the j-th pixel, represents the radar backscattering coefficient of the i-th row j-th column pixel, represents the geometric spatial neighborhood scattering increment of the i-th row j-th column pixel, represents the local background scattering energy value of the i-th row j-th column pixel, and the summation symbol k represents the index count of the pixel spectrum sequence, and n represents the length of the pixel near-infrared reflectivity sequence. S213: based on the pixel water body judgment matrix, performing binary mapping operation on each pixel, marking the water body pixels as water body code and marking the non-water body pixels as non-water body code, and generating a wetland water body distribution mask; S3: calling the pixel positions marked as water body in the wetland water body distribution mask, counting the spatial connectivity of the water body pixels, calculating the Euclidean distance between adjacent water body pixels, merging the water body pixels into water body patches, extracting the boundary pixel coordinates, and forming a wetland water body boundary vector set; The specific steps of S3 are: S311: based on the pixel positions marked as water body in the wetland water body distribution mask, acquiring a two-dimensional coordinate matrix of the marked pixels, performing Euclidean distance calculation, judging whether the water body pixels are adjacent, and obtaining a pixel adjacency judgment coefficient according to the Euclidean distance comparison of the adjacency relationship; S312: calling the pixel adjacency judgment coefficient, performing a connected relationship clustering operation on the pixel number, calculating the water body pixel aggregation intensity value according to the spatial position relationship between the pixel and the adjacent pixel, and obtaining a water body clustering paragraph matrix; S313: According to the water body clustering paragraph matrix, water body connected region pixels are extracted, neighborhood topological detection is performed, boundary pixels are determined, and contour tracking is performed in spatial order, closed curves are obtained, and unified vector expression is performed to obtain a wetland water body boundary vector set; S4: The boundary coordinates of the water body patch in the wetland water body boundary vector set are called, the area, perimeter and shape complexity index of the water body patch are calculated, the wetland is divided into river type, lake type and marsh type wetland through a multi-parameter fusion discrimination model, and a wetland type classification result map is output; S5: The type label and spatial position of the water body patch in the wetland type classification result map are called, the vegetation coverage data and soil water content data are combined, the vegetation type distribution probability in the surrounding buffer area is calculated, the wetland type label of the mismatching area is adjusted, and a wetland information extraction result map is formed; The wetland information extraction result map includes a wetland type adjustment layer, a wetland ecological attribute layer and a wetland spatial consistency layer. 2.The method according to claim 1, wherein, The multi-source fusion image data set includes a spectral reflectance feature set, a radar scattering coefficient feature set and a ground object attribute feature set, the wetland water body distribution mask includes a water body indication grid set, a water body classification grid set and a water body spatial region set, the wetland water body boundary vector set includes a boundary node set, a boundary line set and a boundary topology set, and the wetland type classification result map includes a wetland type attribute layer, a wetland type distribution layer and a wetland zoning layer. The water body feature judgment reference value is a judgment threshold value determined by statistics samples according to the joint characteristics of low reflectivity of water body in near-infrared band and low backscattering in radar image. 3.The method according to claim 1, wherein, The specific steps of S4 are: S411: Based on the boundary coordinates of the water body patch in the wetland water body boundary vector, the boundary coordinate sequence of each water body patch is obtained, the difference vector of adjacent coordinate points is calculated, the area is calculated by using the cross product method of the coordinate point difference vector, the closed correction is combined with the first and last coordinate point difference, and the water body patch area matrix is established; S412: According to the water body patch area matrix, the Euclidean distance between any coordinate point in the water body patch boundary coordinate sequence and the next coordinate point in sequence is extracted, the ratio of the coordinate sequence length and the distance sequence is calculated, and the shape complexity parameter set is generated; S413: The shape complexity parameter set and the water body patch area matrix are called, and the two data are classified and judged based on a multi-parameter fusion discrimination model, and are respectively marked as river type wetland, lake type wetland and marsh type wetland to obtain a wetland type classification result map.

4. The method according to claim 3, wherein, The specific steps of S5 are: S511: Based on the type label and spatial position of the water body patch in the wetland type classification result map, the vegetation coverage data and soil water content data are called, the difference between the coverage value and the water content value in the buffer area is compared and the value distribution sequence is fitted, and the section aggregation sequence is generated by performing section aggregation; S512: According to the section aggregation sequence, the vegetation coverage data and soil water content data are called, the values in the water body patch surrounding buffer area are monitored, the section normalization calculation is performed according to the value of the section in the section aggregation sequence, and the section normalization matrix is obtained. S513: Based on the segment normalization matrix, the water body patch type label in the wetland type classification result map is called, the probability value in the matrix is compared with the corresponding label value of the vegetation coverage data and the soil moisture content data, and the unmatched label is adjusted to obtain a wetland information extraction result map.

5. A wetland information extraction system of multi-source remote sensing data fusion, characterized in that, The system is used to implement the wetland information extraction method of multi-source remote sensing data fusion according to any one of claims 1-4, and the system comprises: A data fusion module acquires optical images, radar images and water level time series data of a wetland area, aligns the spatial positions of the optical images and the radar images, superimposes and fuses the multi-spectral band data and the backscattering coefficient according to the pixel positions by using a band fusion algorithm to generate a multi-source fusion image data set, and delivers the data set to an optical radar discrimination module. The optical radar discrimination module calls the multi-spectral reflectivity and the backscattering coefficient of the pixels in the multi-source fusion image data set, extracts the near-infrared band reflectivity and the radar backscattering coefficient, identifies water body pixels and non-water body pixels according to the water body double characteristics combined with the water body characteristic determination reference value, generates a wetland water body distribution mask, and delivers the mask to a water body connectivity analysis module. The water body connectivity analysis module calls the pixel positions marked as water bodies in the wetland water body distribution mask, counts the spatial connectivity of the water body pixels, calculates the Euclidean distance between adjacent water body pixels, merges the water body pixels into water body patches, extracts the boundary pixel coordinates, forms a wetland water body boundary vector set, and delivers the set to a wetland structure classification module. The wetland structure classification module calls the boundary coordinates of the water body patches in the wetland water body boundary vector set, calculates the area, perimeter and shape complexity index of the water body patches, divides the water body patches into river type, lake type and marsh type wetlands through a multi-parameter fusion discrimination model, outputs a wetland type classification result map, and delivers the map to a wetland label calibration module. The wetland label calibration module calls the type label and the spatial position of the water body patches in the wetland type classification result map, combines the vegetation coverage data and the soil moisture content data, calculates the vegetation type distribution probability in the surrounding buffer area, adjusts the wetland type label of the unmatched area, and forms a wetland information extraction result map.

Images