A method for extracting embankment geometric feature parameters based on multi-source data

Abstract

The application discloses a kind of embankment geometric feature parameter extraction method based on multi-source data, including obtaining remote sensing satellite data, the embankment geometric feature parameter extraction sequence includes: S1 optical image river water body extraction, S2 optical image embankment extraction, S3 embankment result optimization, S4 embankment skeleton line extraction and backwater side embankment foot extraction, S5 embankment section shape extraction and S6 embankment parameter extraction, the application realizes the identification extraction of embankment face based on satellite remote sensing data and the calculation of embankment section, realizes the acquisition of embankment spatial feature parameter using remote sensing means.

Description

TECHNICAL FIELD

[0001] The present application relates to a kind of embankment geometric feature parameter extraction method based on multi-source data. BACKGROUND

[0002] At present, DTM / DSM data is obtained by unmanned aerial vehicle flying embankment, and centimeter-level resolution height image is used to extract embankment spatial feature parameters.

[0003] The method of obtaining centimeter-level height image by unmanned aerial vehicle for embankment parameter extraction is limited by the distance of unmanned aerial vehicle flight, and the method is only for the extraction of embankment center line and the extraction of embankment toe position above water surface and embankment slope. The extracted embankment parameters do not involve the embankment toe position on the water side and the embankment length submerged in river. The disadvantages of the prior art are: the range of embankment feature extraction parameters is small due to the limitation of unmanned aerial vehicle flight distance, the extracted feature parameters only involve the area above water surface, not the area below water surface, and the feature parameters are incomplete. SUMMARY

[0004] The purpose of the present application is to provide an embankment geometric feature parameter extraction method based on multi-source data.

[0005] In order to achieve the above purpose, the scheme of the present application is:

[0006] An embankment geometric feature parameter extraction method based on multi-source data, comprising obtaining remote sensing satellite data, the embankment geometric feature parameter extraction sequence comprising: S1 optical image river water body extraction, S2 optical image embankment extraction, S3 embankment result optimization, S4 embankment skeleton line extraction and backwater side embankment toe extraction, S5 embankment section shape extraction and S6 embankment parameter extraction, wherein:

[0007] S1 optical image river water body extraction: extracting smooth water body contour without holes and small areas from the obtained remote sensing satellite data;

[0008] S2 optical image embankment extraction: extracting multi-band data image from the obtained remote sensing satellite data, inputting the extracted multi-band data image into an embankment extraction model to obtain embankment result;

[0009] S3 embankment result optimization: according to the spatial relationship of embankment adjacent to river water body, buffer zone is made outside water body contour, region judgment is carried out, the area outside buffer zone is non-embankment, and the area inside buffer zone is embankment. The embankment surface is obtained from the embankment result by using the buffer zone as a mask, the skeleton line of the buffer zone is extracted and its angle is calculated, the MDMP multidirectional morphological processing is carried out on the embankment surface extracted by mask, and the final optimized embankment surface result is obtained;

[0010] The S4 embankment skeleton line extraction and backwater side embankment foot extraction: a smooth embankment skeleton line is extracted by the CNS algorithm, an edge of the embankment surface is detected by using a Sobel operator to extract an embankment surface contour line, a smooth curve is obtained by smoothing the embankment surface contour line, extreme points of adjacent edges are identified, end points are captured to connect polygons, and both bank backwater side embankment feet are obtained;

[0011] The S5 embankment section shape extraction: a water embankment section shape is obtained by extracting a cross section of a river according to a DEM digital elevation model, a water level elevation of a section of a neighboring hydrological station and a water level elevation of the embankment section are obtained, a water level difference between the two is calculated, a river bottom elevation value obtained from the complete section shape of the hydrological station is combined with the water level difference to calculate a river bottom elevation of the embankment section, and the complete embankment section shape is obtained by interpolating the water embankment section shape and the river bottom elevation to calculate a water under embankment section shape;

[0012] The S6 embankment parameter extraction: according to the characteristics of being low on both sides, high in the middle and constant in elevation, the complete embankment surface width, the embankment top width and the height and width of the embankment body on both bank water sides and backwater sides are extracted according to the section shape, the slope of the embankment body on both bank water sides and backwater sides is obtained according to a slope calculation formula, and the embankment length is calculated according to the embankment skeleton line.

[0013] The embankment extraction model is to make embankment sample training sets and non-embankment sample training sets from the extracted multi-band data images, to take the two kinds of sample training sets as inputs to train a random forest model, and to construct the embankment extraction model.

[0014] The scheme further comprises the following steps:

[0015] The optical image data set is obtained by orthorectifying the panchromatic and multispectral data in the obtained observation satellite data, the panchromatic band and the multispectral data in the optical image data set are fused to improve the multispectral resolution to 0.8 m, the normalized water index NDWI and the normalized vegetation index NDVI are calculated, the calculated NDWI, NDVI bands and the red, green, blue and near-infrared bands contained in the image are synthesized into an image data set containing multiple bands, the water threshold is automatically extracted by using the OTSU threshold method according to the NDWI band, the river water range is extracted by masking, the small holes and smooth water body contour in the water body are removed by morphology, and the smooth water body contour without holes and small areas is obtained.

[0016] NDWI is the normalized water index, NDWI=(Green-NIR) / (Green+NIR);

[0017] NDVI is the normalized vegetation index, NDVI=(NIR-R) / (NIR+R);

[0018] Green is the reflectivity of green light band, NIR is the reflectivity of near infrared band, R is the reflectivity of red light band.

[0019] The scheme is further: the multiple bands are 6 bands.

[0020] The scheme is further: the method for removing the small holes in the water body and smoothing the water body contour through morphology comprises: performing image opening operation and image closing operation on the image, the image opening operation smoothes the contour of the object, disconnects narrow necks and eliminates fine protrusions, the image closing operation fills narrow gaps and long gullies, eliminates small holes and fills the broken contour lines, wherein the opening operation is first etching and then inflation, and the closing operation is first inflation and then etching.

[0021] The scheme is further: in the dike result optimization, first, small holes in the dike result are removed through morphology and the contour of the dike result is smoothed to obtain a preliminary optimization result of the dike.

[0022] The scheme is further: the dike surface extracted by the mask is subjected to MDMP multi-directional morphology processing, which is used for preventing structure breakage or loss when processing the dike linear object.

[0023] Compared with the prior art, the satellite remote sensing data dike surface result is accurately extracted, the existing section data and DEM are used to calculate the dike section terrain, the dike spatial feature parameter extraction is realized, and effective technical support is provided for the dike parameter extraction.

[0024] The application will be further described in detail in combination with the description of the drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The flowchart of the extraction method of the application is shown in the figure;

[0026] Figure 2 The dike surface diagram extracted according to the application is shown in the figure;

[0027] Figure 3 The dike skeleton line diagram extracted according to the application is shown in the figure;

[0028] Figure 4 The dike section diagram deduced according to the application is shown in the figure;

[0029] Figure 5 The left bank dike cross section diagram extracted according to the application is shown in the figure;

[0030] Figure 6 The right bank dike cross section diagram extracted according to the application is shown in the figure. DETAILED DESCRIPTION

[0031] A kind of embankment geometric feature parameter extraction method based on multi-source data, including obtaining remote sensing satellite data for embankment, it is in remote sensing satellite data service platform screening less than 10% GF-7BWD surface reflectivity product, the full-color band resolution in surface reflectivity product is 0.8m, and the resolution of multispectral is 3.2m, as shown in Figure 1 The embankment geometric feature parameter extraction sequence includes: S1 optical image river water body extraction, S2 optical image embankment extraction, S3 embankment result optimization, S4 embankment skeleton line extraction and backwater side embankment foot extraction, S5 embankment section shape extraction and S6 embankment parameter extraction, wherein:

[0032] S1 optical image river water body extraction: extract smooth water body contour without hole and small area from the obtained remote sensing satellite data;

[0033] S2 optical image embankment extraction: extract multi-band data image from the obtained remote sensing satellite data, input the extracted multi-band data image into an embankment extraction model, and obtain the embankment result;

[0034] S3 embankment result optimization: according to the spatial relationship that embankment is adjacent to river water body, buffer zone is made outside water body contour, region judgment is carried out, non-embankment outside buffer zone and embankment inside buffer zone, embankment surface is obtained from embankment result by using buffer zone as mask, skeleton line of buffer zone is extracted and its angle is calculated, MDMP multidirectional morphological processing is carried out on embankment surface extracted by mask, and finally optimized embankment surface result is obtained;Traditional morphological structure element is isotropic, and when processing linear object such as embankment, structure fracture or loss may occur, and the improved MDMP (multidirectional morphological) processing can solve this problem;

[0035] S4 embankment skeleton line extraction and backwater side embankment foot extraction: smooth embankment skeleton line is extracted by CNS algorithm, edge detection is carried out on embankment surface by Sobel operator to extract embankment contour line, smooth curve is obtained by smoothing embankment contour line, extreme points of adjacent edges are identified, end point capture connects polygon, and backwater side embankment foot on both sides is obtained;Wherein: the extracted embankment surface result applies Zhang-Suen skeleton algorithm, which can cause redundant branch line of skeleton and multiple skeleton lines, too many redundant points on line segment and disconnected center line, so that it is difficult to obtain a complete smooth center line, and the improved CNS algorithm (connected optimization skeleton) can solve the problem, and the smooth embankment skeleton line is extracted;

[0036] The S5 embankment section shape extraction: according to the DEM digital elevation model, the cross section of the river is extracted to obtain the water embankment section shape, and the water level elevation of the adjacent hydrological station section and the water level elevation of the embankment section are obtained, and the water level difference of the two is calculated, the river bottom elevation value obtained from the complete section shape of the hydrological station is combined with the water level difference to calculate the river bottom elevation of the embankment section, and the river bottom elevation of the embankment section is calculated by interpolation from the embankment water section shape and the river bottom elevation, so as to obtain the complete embankment section shape; the complete embankment section shape is obtained by calculating the embankment underwater section shape by Kriging interpolation.

[0037] Kriging interpolation formula:

[0038] Z B = Z A +(H B -H A )

[0039] Wherein Z B is the embankment section river bottom elevation value, Z A is the adjacent hydrological station river bottom elevation value, H B is the embankment section water level elevation value, H A is the adjacent hydrological station water level elevation value.

[0040] The S6 embankment parameter extraction: according to the characteristics of low in the middle and high in the two sides of the embankment with constant elevation, the complete embankment width, embankment width and the height and width of the embankment body on the two sides of the water side and the backwater side are extracted according to the section shape, and the slope of the embankment body on the two sides of the water side and the backwater side is obtained according to the slope calculation formula; the length of the embankment is calculated by applying the embankment skeleton line.

[0041] The embankment extraction model is to make embankment sample training set and non-embankment sample training set from the extracted multi-band data image, to train the random forest model by inputting the two kinds of sample training sets as input, and to build the embankment extraction model, the specific method comprising:

[0042] Select 70% of the sample number as the training random forest model, and 30% of the sample as the test model, the input sample contains 6 band characteristics and 0, 1 characteristics (1 for embankment, 0 for non-embankment), input the data containing 6 band characteristics to the built random forest model, and get the binary image.

[0043] Figures 2 to 6 The schematic diagram of extracting dam parameters is shown.

[0044] The method for extracting smooth water body contour without holes and small area is:

[0045] In the acquired observation satellite data (in the surface reflectance product), optical image data sets are obtained by orthorectifying the panchromatic and multispectral data, the panchromatic band and the multispectral data in the optical image data sets are applied for band fusion, the multispectral resolution is increased to 0.8 m, the normalized water index NDWI and the normalized vegetation index NDVI are calculated, the calculated NDWI, NDVI bands and the red light, green light, blue light and near-infrared bands contained in the image are synthesized into an image data set containing multiple bands (at least 6 bands), the water body threshold is automatically extracted according to the NDWI band by using the OTSU threshold method (maximum inter-class variance method), the river water body range is extracted by using a mask, small holes in the water body and the smooth water body contour are removed by using morphological methods, and a smooth water body contour without holes and small areas is obtained.

[0046] NDWI is a normalized water index, NDWI = (Green-NIR) / (Green+NIR);

[0047] NDVI is a normalized vegetation index, NDVI = (NIR-R) / (NIR+R);

[0048] Green is the reflectivity of the green band, NIR is the reflectivity of the near-infrared band, and R is the reflectivity of the red band.

[0049] The method for removing small holes in the water body and smoothing the water body contour by using morphological methods includes: performing image opening operation and image closing operation on the image, the image opening operation smoothes the contour of the object, disconnects narrow necks and eliminates small protrusions, the image closing operation fills narrow gaps and long gullies, eliminates small holes and fills the breaks in the contour line, wherein the opening operation is first eroded and then expanded; and the closing operation is first expanded and then eroded.

[0050] In the dike result optimization, small holes in the dike result and the dike result contour are first removed by using morphological methods to obtain a preliminary optimization result of the dike.

[0051] The dike surface extracted by using the mask is subjected to MDMP multi-directional morphological processing, which is used for preventing structure breakage or loss when processing dike linear objects.

[0052] In example S3, the specific method of the MDMP multi-directional morphological processing includes:

[0053] The directional structure element is B θ , and θ is the calculated angle.

[0054]

[0055] Wherein:

[0056] For open operation, f·B θ For closed operation.

[0057] Θ = {θ1, θ2,..., θ n}

[0058]

[0059] Wherein:

[0060] f θ (x, y) is open and closed operation, w i is weight determination, f fused (x, y) is the final processing result.

[0061] In example S4: the specific method for extracting smooth dike skeleton line by CNS algorithm comprises:

[0062] Traverse all non-zero pixels in the binary image, when judging whether to delete or keep each pixel, the position farthest from the center is the "center" point of the dike, and the skeleton should be arranged along these high-value connection areas.

[0063]

[0064] Wherein:

[0065] D(x, y) is the distance from point (x, y) to the nearest boundary

[0066] Extracting "local maximum value" or "ridge line" as skeleton point on the distance map can avoid multiple skeleton lines, and the skeleton point is closer to the "center",

[0067]

[0068] Wherein:

[0069] N is the neighborhood window;

[0070] After extracting the skeleton point, extract all skeleton end points Pi, for example, connect only one pixel point through 8-neighborhood:

[0071]

[0072] Wherein α·dist(i, j) is the Euclidean distance, is the gradient, and α and β are adjustment parameters.

[0073] In the embodiment S4, the edge detection by the Sobel operator is used to identify and locate the boundary part with significant change in pixel intensity in the image, and the specific method of extracting the contour line of the embankment surface includes: extracting the geometric contour and surface features of the embankment by the edge detection, detecting the edge by calculating the gradient of the image by the Sobel operator, including two convolution kernels, respectively calculating the gradient in the horizontal G x and vertical G y directions:

[0074]

[0075] wherein I is the input image, and * represents the convolution operation.

[0076] After the two kernels are respectively convolved with the image, the final gradient amplitude G is calculated in the following manner:

[0077]

[0078] The gradient direction (edge direction) is:

[0079]

[0080] In the embodiment S5, the specific method used in the Kriging interpolation includes:

[0081]

[0082] wherein:

[0083] is the estimated value of the point (x o ,y o ), that is, z o =z(x o ,y o ); wherein λ i is the weight coefficient, but is not the inverse of the distance, but is a set of optimal coefficients that can satisfy the difference between the estimated value z at the point (x o ,y o ) and the true value z o is the minimum, that is:

[0084]

[0085] while satisfying the condition of unbiased estimation

[0086] The above-mentioned one embodiment of the method for extracting the geometric feature parameters of the embankment based on multi-source data realizes the extraction of the spatial feature parameters of the embankment by accurately extracting the embankment surface of the satellite remote sensing data, calculates the terrain of the embankment section by using the existing section data and DEM, and provides effective technical support for the extraction of the embankment parameters.

Claims

1. A method for extracting parameters of embankment geometric features based on multi-source data, comprising obtaining remote sensing satellite data, characterized in that, The embankment geometric feature parameter extraction sequence comprises: S1, optical image river channel water body extraction; S2, optical image embankment extraction; S3, embankment result optimization; S4, embankment skeleton line extraction and backwater side embankment foot extraction; S5, embankment section shape extraction; and S6, embankment parameter extraction. The S1 optical image river channel water body extraction is to extract a smooth water body contour without holes and small areas from acquired remote sensing satellite data; The S2 optical image embankment extraction is to extract multi-band data images from acquired remote sensing satellite data, input the extracted multi-band data images into an embankment extraction model, and obtain an embankment result; The S3 embankment result optimization is to perform region judgment on a buffer zone outside the water body contour according to the spatial relationship between the embankment and the river channel water body, the buffer zone outside the region is non-embankment, and the buffer zone inside the region is embankment, a dam surface is obtained from the embankment result by using the buffer zone for mask, a skeleton line of the buffer zone is extracted and an angle thereof is calculated, MDMP multi-directional morphological processing is performed on the dam surface extracted by using the mask, and a final optimized dam surface result is obtained; The S4 embankment skeleton line extraction and backwater side embankment foot extraction is to extract a smooth embankment skeleton line through a CNS algorithm, perform Sobel operator edge detection on the dam surface to extract a dam surface contour line, perform smoothing processing on the dam surface contour line to obtain a smooth curve, identify extreme points of adjacent edges, capture endpoints to connect polygons, and obtain backwater side embankment feet on both banks; The S5 embankment section shape extraction is to extract a water-on embankment section shape according to a DEM digital elevation model in a river channel, acquire a water level elevation of a section of a hydrological station and a water level elevation of an embankment section, calculate a water level difference between the two, obtain a river bottom elevation value from a complete section shape of the hydrological station, combine the water level difference to calculate a river bottom elevation of the embankment section, interpolate the embankment water-on section shape and the river bottom elevation to calculate an embankment water-under section shape, and thus obtain a complete embankment section shape; The S6 embankment parameter extraction is to extract complete embankment surface widths, embankment top widths, and heights and widths of embankment bodies on both banks according to the characteristics of two sides being low, the middle being high, and the elevations being constant, acquire slopes of the embankment bodies on both banks according to a slope calculation formula, and calculate an embankment length by using the embankment skeleton line. The embankment extraction model is to make embankment sample training sets and non-embankment sample training sets from the extracted multi-band data images, input the two kinds of sample training sets for random forest model training, and construct the embankment extraction model.

2. The extraction method according to claim 1, characterized in that, The method for extracting the smooth water body contour without holes and small areas is: In the acquired observation satellite data, optical image data set is obtained by orthorectification of the panchromatic and multispectral data, the panchromatic band and multispectral data in the optical image data set are applied for band fusion, the multispectral resolution is increased to 0.8 m, normalized water index NDWI and normalized vegetation index NDVI are calculated, the calculated NDWI, NDVI band and the red light, green light, blue light and near-infrared band contained in the image are synthesized into an image data set containing multiple bands, the water threshold is automatically extracted according to the NDWI band by using OTSU threshold value method, the river water range is extracted by mask, small holes in the water body and smooth water body contour are removed through morphology, and the smooth water body contour without holes and small areas is obtained; wherein: NDWI is normalized water index, NDWI=(Green-NIR) / (Green+NIR); NDVI is normalized vegetation index, NDVI=(NIR-R) / (NIR+R); Green is the reflectivity of the green band, NIR is the reflectivity of the near-infrared band, and R is the reflectivity of the red band.

3. The extraction method according to claim 2, characterized in that, The multiple bands are six bands.

4. The extraction method of claim 2, wherein, The method for removing small holes in the water body and smoothing the water body contour through morphology includes: image opening operation and image closing operation are performed on the image, the image opening operation smoothes the contour of the object, disconnects narrow necks and eliminates fine protrusions, the image closing operation fills narrow discontinuities and long gullies, eliminates small holes and fills the breaks in the contour line, wherein the opening operation is first eroded and then expanded; and the closing operation is first expanded and then eroded.

5. The extraction method of claim 1, wherein, In the dike result optimization, small holes in the dike result and the dike result contour are removed through morphology, and the preliminary optimization result of the dike is obtained.

6. The extraction method of claim 1, wherein, The dike surface extracted by the mask is subjected to MDMP multidirectional morphological processing, which is used for preventing structure breakage or loss when processing the dike linear object.

7. The extraction method of claim 1, wherein, In S4, the specific method for extracting the smooth dike skeleton line through the CNS algorithm includes: All non-zero pixels in the binary image are traversed, and when judging whether to delete or retain each pixel, the position farthest from the pixel is the "centermost" point of the dike, and the skeleton should be arranged along the high-value connection area.

8. The extraction method of claim 1, wherein, In S4: edge detection with the Sobel operator is used to identify and locate border portions in the image where the pixel intensity changes significantly, the specific method of extracting the contour line of the embankment surface includes: using edge detection to extract the geometric contour and surface features of the embankment, detecting the edge by calculating the gradient of the image with the Sobel operator, including two convolution kernels, respectively calculating the gradient in the horizontal and vertical direction: where I is the input image, denotes a convolution operation.

9. The extraction method of claim 1, wherein, In S5, the dike section shape is obtained by calculating the dike underwater section shape through Kriging interpolation, so as to obtain the complete dike section shape; the Kriging interpolation formula is: wherein is the levee cross-section river bottom elevation value, is the adjacent hydrological station river bottom elevation value, is the levee cross-section water level elevation value, is the adjacent hydrological station water level elevation value.

Images