A river channel extraction method based on water surface data

Through multi-source remote sensing satellite data and mathematical morphological methods combined with DEM data, Delaunay triangulation and natural breakpoint algorithm, the problem of insufficient data utilization in river channel monitoring is solved, and efficient, accurate extraction and monitoring efficiency of river channels is improved. .

CN118334530BActive Publication Date: 2024-12-06CHINA INST OF WATER RESOURCES & HYDROPOWER RES
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410449029.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-15
Publication Date
2024-12-06
Estimated Expiration
2044-04-15

AI Technical Summary

Technical Problem

The prior art lacks effective methods in river monitoring to utilize digital elevation model data and river surface data, resulting in insufficient accuracy in river extraction, especially in extreme conditions, it is difficult to ensure the real-time and efficiency of river monitoring.

Method used

Multi-source remote sensing satellite data is used to combine mathematical morphological methods for water surface extraction and reconstruction, and Delaunay triangulation algorithm and natural breakpoint algorithm are used to extract river channels, segmented analysis and buffer analysis are used to determine river channels.

Benefits of technology

It realizes efficient, accurate positioning and extraction of river channels, is suitable for large-scale river channel analysis, improves monitoring efficiency, and supports river evolution and human-land relationship analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118334530B_ABST
    Figure CN118334530B_ABST
Patent Text Reader

Abstract

The present invention discloses a river channel extraction method based on water surface data, comprising the following steps: step 1, water body extraction: using multispectral data and radar data to extract water body using water body index method to obtain water surface data; step 2, water body reconstruction: reconstructing water body based on mathematical morphology for the water surface data obtained in step 1; step 3, extracting river channels in combination with DEM natural break point algorithm, using the water body reconstructed in step 2 in combination with DEM data to extract river channels. The method of the present invention extracts river channel boundaries for subsequent related hydrological analysis, has good extraction effect, and is suitable for a wide range of river channel extraction analysis.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of water conservancy projects, and in particular to the technical field of river monitoring, and specifically is a river channel extraction method based on water surface data. Background Art

[0002] Rivers are one of the landscapes most affected by human activities. The scope of the river is the basis of river monitoring. At present, there is a lack of monitoring data on the scope of the river. Accurately locating the boundary of the river is an important part of river measurement. It has important application value for the judgment and monitoring of flood disasters, and is of great significance to water conservancy project construction, water resources development and utilization, and river water monitoring.

[0003] Traditional river monitoring and survey methods mainly rely on manual measurement. However, some areas are subject to unfavorable factors such as changeable climate, rugged roads and harsh geographical environment. Therefore, manual measurement methods have problems such as low efficiency and poor real-time performance. Especially under extreme conditions such as floods and rainstorms, it is difficult to ensure the monitoring of rivers. With the development of satellite remote sensing technology, earth science research such as hydrology has made breakthrough progress. Remote sensing technology has the unique advantage of being able to obtain all-round, multi-temporal, large-area river range, river width, water level and other information in a timely and stable manner. It has expanded the source of raw data in hydrology and provided new possibilities and research directions for river monitoring.

[0004] The extraction of river channel range is of great significance in flood simulation and river hydrological monitoring. At present, the research on river channel extraction based on remote sensing is very extensive. The existing research mainly focuses on edge detection method, contour model segmentation method, neural network model, deep learning, etc., all of which have achieved good results. Through existing research, it is found that the use of digital elevation model data (DEM) and river surface data in river channel extraction is insufficient. The research in this area is still immature and there is a technical gap. It is of great significance to study the use of DEM data and remote sensing water surface data for river channel extraction. Summary of the invention

[0005] The present invention provides a river channel extraction method based on water surface data, the purpose of which is to use multi-source remote sensing satellites to extract river channel boundaries for subsequent related hydrological analysis, river evolution human-land relationship analysis, river channel evolution monitoring, etc. The extraction effect is high and it is suitable for a wide range of river channel extraction and analysis. The purpose of the present invention is achieved through the following technical solutions:

[0006] A river channel extraction method based on water surface data comprises the following steps:

[0007] Step 1: Water surface extraction: water surface extraction is performed using multispectral data and / or radar data using a water body index method to obtain water surface data;

[0008] Step 2: Water surface reconstruction: Reconstruct the water surface data obtained in step 1 based on mathematical morphology;

[0009] Step 3: Use the water surface reconstructed in step 2 and DEM data to extract the river channel, which specifically includes the following steps:

[0010] 3-1 River triangulation: The water surface data is triangulated using the Delaunay triangulation algorithm to extract skeleton lines and obtain skeleton lines and river units. The skeleton line is the center line of the water surface; the river unit is the triangle obtained by triangulation;

[0011] 3-2 River segmentation: Segment the river according to the skeleton line and river unit setting related parameters.

[0012] 3-3 River channel extraction: Buffer analysis is performed on each river section to obtain the buffer range and thus determine the analysis range. Then, the river channel is extracted based on the natural break point algorithm based on DEM within the analysis range of each river section, and finally mosaicked into a complete river channel.

[0013] For further optimization, in step one: select one of the GF-1 multispectral data, Sentinel-1 radar data, and Sentinel-2 multispectral data of the study area during the flood season to extract the river water surface and obtain the water surface data as the water surface data for subsequent river channel extraction; or select the GF-1 multispectral data, Sentinel-1 radar data, and Sentinel-2 multispectral data of the study area during the flood season to extract the river water surface respectively, and obtain the maximum water surface of the river during the flood season from different data sources, and merge the results to obtain the final maximum river water surface data as the water surface data for subsequent river channel extraction.

[0014] Furthermore, the water surface reconstruction based on mathematical morphology is a water surface reconstruction method using a direct connection method, which specifically includes the following steps:

[0015] 2-1 Artificially draw multi-segment elements at the breakwater based on water surface data and remote sensing images;

[0016] 2-2 Perform spatial connection analysis on multi-line elements and water surface data to remove redundant patches;

[0017] 2-3 Take the average water surface width at both ends of the interrupted flow to perform buffer analysis on the multi-line segments, merge the buffer surface and the water surface to fill the interrupted flow gap.

[0018] Furthermore, the DEM data in step three is 5mDEM data or 12.5mDEM data.

[0019] Furthermore, the relevant parameters set in step 3-2 include: artificial intervention points, nodes of the river skeleton line, maximum length of the river section, maximum number of units in the river section and minimum number of units in the tributaries.

[0020] Furthermore, the specific operation of river segmentation is: based on river units and skeleton lines, it is considered that multiple continuous river units form different river sections based on some points of interest (i.e., artificial intervention points: hydrological station locations, artificial water conservancy facilities, flood storage areas, etc.). The river unit is weak and cannot represent the characteristics of long river sections. After river segmentation, the requirements for computing performance will be reduced and the computing efficiency will be improved. The river segmentation standards are as follows:

[0021] (1) Artificial intervention points such as hydrological stations, artificial water conservancy facilities, flood storage areas, etc. (the rivers in these areas are greatly affected by human activities, so they need to be segmented and analyzed separately).

[0022] (2) Nodes of the river skeleton line, which is the basis for basic segmentation;

[0023] (3) Maximum length of a river section: refers to the maximum length of a river section, based on the river skeleton (the degree of meandering of the river).

[0024] Adjust the maximum length of a river section (if it is a straight river, the maximum length can be larger, and if it is a winding river, the maximum length can be smaller accordingly);

[0025] (4) Maximum number of river unit sections: This is used to compress and optimize performance based on the curvature of the river, avoid segmenting each node, and make adjustments based on the curvature. The essence of curvature is the number of river units. The more winding the river section, the smaller the maximum number of river unit sections; the straighter the river section, the larger the maximum number of river unit sections.

[0026] (5) Minimum number of tributary units: Set the minimum number of tributary units, and remove those with less than this limit. Optimize the river segmentation, extract the central axis, remove the non-important skeleton and retain the core part.

[0027] Furthermore, in step 3-3, hydrologically, the river channel is divided into the main channel and the beach according to the river cross section, and the two are defined as the river channel range. The river channel has special terrain undulations in topography - the cross section of the river channel is mostly parabolic or trapezoidal. Based on the topographic trend of the river channel, the natural break point classification algorithm can be used to group and classify the DEM and cluster and segment the river channel range. The river channel boundary can be extracted by utilizing the characteristics of the river and the undulating terrain characteristics on the cross section of the river channel. The natural break point classification algorithm is used to extract the river channel based on DEM. The natural break point algorithm is a data classification method based on the natural grouping inherent in the data, in which the histogram of the DEM terrain has a "multi-peak" feature (such as Fig. 22), each "peak" represents a typical landform; landforms at different elevations are naturally grouped according to the specified number of classifications, and the river channel group is the group with the smallest elevation in the natural grouping, and the river channel range is obtained by extracting the minimum grouping range. The advantages and beneficial effects of the present invention are:

[0028] Compared with the prior art, the river channel extraction method based on water surface data of the present invention makes maximum use of multi-source remote sensing data, and uses river water surface extraction data to play a good role in locating the river channel. It is suitable for large-scale river channel extraction and analysis. At the same time, the setting of the river channel extraction and analysis range according to the water surface data improves the analysis efficiency. In the future, it can realize annual updating of the river channel range in the basin to monitor the evolution of the river channel. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0030] Figure 1 Technology roadmap;

[0031] Figure 2 Schematic diagram of the principle of extracting skeleton lines of arbitrary polygons;

[0032] Figure 3 Water surface extraction results of Huaihe River Study Area 1: GF-1 water surface extraction results and reconstruction results, Sentinel-1 water surface extraction results and reconstruction results, Sentinel-2 water surface extraction results and reconstruction results, and the maximum water surface results of the river;

[0033] Figure 4 The results of river channel extraction based on GF-1 water surface data in the Huaihe River study area: (a1) is the river channel extraction result of 5m resolution DEM based on GF1 water surface extraction result, (a2) is the river channel extraction result of 12.5m resolution DEM based on GF1 water surface extraction result;

[0034] Figure 5 The river channel extraction results of Sentinel-1 water surface data in the Huaihe River study area: (a1) is the river channel extraction result of 5m resolution DEM based on Sentinel1 water surface extraction results, (a2) is the river channel extraction result of 12.5m resolution DEM based on Sentinel1 water surface extraction results;

[0035] Figure 6 The river channel extraction results of Sentinel-2 water surface data in the Huaihe River study area: (a1) is the river channel extraction result of 5m resolution DEM based on Sentinel2 water surface extraction results, (a2) is the river channel extraction result of 12.5m resolution DEM based on Sentinel2 water surface extraction results;

[0036] Figure 7The results of river channel extraction with the maximum water surface of the 1st river in the Huaihe River study area: (a1) is the river channel extraction result based on the 5m resolution DEM of the maximum water surface of the river, and (a2) is the river channel extraction result based on the 12.5m resolution DEM of the maximum water surface of the river;

[0037] Figure 8 The water body extraction results of Huaihe River Study Area 2: GF-1 water surface extraction results and reconstruction results, Sentinel-1 water surface extraction results and reconstruction results, Sentinel-2 water surface extraction results and reconstruction results, and the maximum water surface results of the river;

[0038] Fig. 9 The river channel extraction results of 2GF-1 water surface data in the Huaihe River study area: (a1) is the river channel extraction result of 5m resolution DEM based on GF1 water surface extraction results, (a2) is the river channel extraction result of 12.5m resolution DEM based on GF1 water surface extraction results;

[0039] Fig.10 The river channel extraction results of 2Sentinel-1 water surface data in the Huaihe River study area: (a1) is the river channel extraction result of 5m resolution DEM based on Sentinel1 water surface extraction results, (a2) is the river channel extraction result of 12.5m resolution DEM based on Sentinel1 water surface extraction results;

[0040] Fig.11 The river channel extraction results of Sentinel-2 water surface data in the Huaihe River study area: (a1) is the river channel extraction result of 5m resolution DEM based on Sentinel2 water surface extraction results, (a2) is the river channel extraction result of 12.5m resolution DEM based on Sentinel2 water surface extraction results;

[0041] Fig.12 The results of river channel extraction with the maximum water surface in Huaihe River Study Area 2: (a1) is the river channel extraction result based on the 5m resolution DEM of the maximum water surface of the river, and (a2) is the river channel extraction result based on the 12.5m resolution DEM of the maximum water surface of the river;

[0042] Fig.13 The water surface extraction and reconstruction results of the Dawen River Plain area: GF-1 water surface extraction and reconstruction results, Sentinel-1 water surface extraction and reconstruction results, Sentinel-2 water surface extraction and reconstruction results, and the maximum water surface results of the river;

[0043] Fig.14 The river channel extraction results of the Dawen River GF-1 water surface data: (a1) is the river channel extraction result of the 5m resolution DEM based on the GF1 water surface extraction result, (a2) is the river channel extraction result of the 12.5m resolution DEM based on the GF1 water surface extraction result;

[0044] Fig.15 The river channel extraction results of the Dawen River Sentinel-1 water surface data: (a1) is the river channel extraction result of the 5m resolution DEM based on the Sentinel1 water surface extraction result, (a2) is the river channel extraction result of the 12.5m resolution DEM based on the Sentinel1 water surface extraction result;

[0045] Fig.16 The river channel extraction results of the Dawen River Sentinel-2 water surface data: (a1) is the river channel extraction result of the 5m resolution DEM based on the Sentinel2 water surface extraction result, (a2) is the river channel extraction result of the 12.5m resolution DEM based on the Sentinel2 water surface extraction result;

[0046] Fig.17 The extraction results of the river channel with the maximum water surface of the Dawen River: (a1) is the river channel extraction result based on the 5m resolution DEM of the maximum water surface of the river, and (a2) is the river channel extraction result based on the 12.5m resolution DEM of the maximum water surface extraction result of the river;

[0047] Fig.18 Distribution of verification points in the study area of ​​the upper reaches of the Huaihe River 1;

[0048] Fig.19 Distribution of 2 verification points in the study area of ​​the upper reaches of the Huaihe River;

[0049] Fig. 20 Distribution of verification points in Dawen River;

[0050] Fig.21 Illustration of calculation of average width of water surface;

[0051] Fig. 22 Example of a DEM histogram (DEM classified into four categories using natural breaks). DETAILED DESCRIPTION

[0052] A river channel extraction method based on water surface data comprises the following steps:

[0053] Step 1: Water surface extraction: water surface extraction is performed using multispectral data and / or radar data using a water body index method to obtain water surface data;

[0054] The data used include digital elevation model data (DEM), GF-1 (WFV) data, Sentinel-1 data (Sentinel-1), and Sentinel-2 data (Sentinel-2). Among them, the DEM data with various resolutions (12.5m, 5m) comes from the Geospatial Data Cloud (Geospatial Data Cloud (gscloud.cn)); the GF-1WFV optical multispectral image comes from the Resource Satellite Center (http: / / www.cresda.com / CN / ); the Sentinel-1 data and Sentinel-2 water surface data are downloaded from the GEE platform ( https: / / code.earthengine.google.com / ).

[0055] GF-1WFV 16m multispectral data preprocessing: GF-1WFV 16m data in the flood season from June to September 2022 in the study area were downloaded from the China Resources Satellite Application Center, and preprocessing such as orthorectification, radiation calibration, atmospheric correction, geometric correction, mosaicking, and cropping were performed to eliminate interference information.

[0056] Sentinel-1 data download and processing: Based on the GEE (Google Earth Engine) platform, download the Level-1 data (including VV and VH bands) in GRDH format in the interferometric wideswath (IW, used to observe land) imaging mode of Sentinel-1 data in the flood season from June to September 2022 in the study area. GEE officials have preprocessed the Sentinel-1 data, mainly including: updating orbit metadata using orbit files, eliminating GRD boundary noise, eliminating thermal noise, radiometric calibration, and terrain correction (orthorectification). Therefore, the data can be directly used on this platform to calculate the corresponding water body index (SDWI).

[0057] Download and process Sentinel-2 data: Based on the GEE platform, four bands (B2, B3, B4, and B8) of Sentinel-2 data for water surface extraction in the study area during the flood season from June to September 2022 were downloaded. GEE officials also preprocessed the Sentinel-2 data, so the data can be directly used on the platform to calculate the corresponding water body index (NDWI).

[0058] The water surface index method includes the normalized water index method and the bipolar water index method; the normalized water index calculation formula is as follows:

[0059]

[0060] In the formula: NDWI is the normalized water index, GREEN is the reflectance of the green band, and NIR is the reflectance of the near infrared band. The input of the normalized water surface index method is the green band and near infrared band of the preprocessed multispectral image, which is raster data, and the output is water surface raster data.

[0061] The dual polarization water surface index method has the following formula:

[0062] K SDWI =ln(10×VV×VH)

[0063] Where: K SDWI Represents the result value of band operation, VV and VH represent Sentinel-1 dual-polarization data; the dual-polarization water surface index method is used as the input of the HV and HH bands of the preprocessed radar data, which is raster data, and the output is water surface raster data, which is subsequently converted into vectors using conversion tools.

[0064] Threshold segmentation uses OTSU algorithm for segmentation. Dilation and erosion are used to process these redundant patches and interrupted flow. Single ground object raster vectorization is input as raster data of water surface results and output as vector data; the parameters include: minimum number of pixels of ground object patches, patches with more than the specified number of pixels are regarded as water surface areas, gap radius, gaps with less than the specified radius will be filled.

[0065] Select one of the GF-1 multispectral data, Sentinel-1 radar data, and Sentinel-2 multispectral data of the flood season in the study area to extract the river water surface and obtain the water surface data as the water surface data for subsequent river channel extraction; or select the GF-1 multispectral data, Sentinel-1 radar data, and Sentinel-2 multispectral data of the flood season in the study area to extract the river water surface respectively, and obtain the maximum water surface of the river in the flood season from different data sources, and merge the results to obtain the final maximum river water surface data as the water surface data for subsequent river channel extraction.

[0066] Step 2: Water surface reconstruction: Reconstruct the water surface data obtained in step 1 based on mathematical morphology;

[0067] The water surface reconstruction based on mathematical morphology is a water surface reconstruction method using a direct connection method, which specifically includes the following steps:

[0068] 2-1 Artificially draw multi-segment elements at the breakwater based on water surface data and remote sensing images;

[0069] 2-2 Perform spatial connection analysis on multi-line elements and water surface data to remove redundant patches;

[0070] 2-3 Take the average water surface width at both ends of the dry flow to perform buffer analysis on the multi-line segments, where the buffer width of the multi-line segments between the dry flow sections is calculated by the average river width of the two ends of the river sections, where the average river width calculation formula at both ends of the river sections is W = 2 × S / L (W: river section width; S: river section area; L: river section perimeter; Fig.21 As shown in the figure, the dotted part is the gap river section, the sum of the areas of the river sections at both ends (S1+S2) is S, and the sum of the perimeters of the river sections at both ends (L1+L2) is L). After obtaining the width, a buffer analysis is performed on the line segment (buffer analysis refers to automatically establishing a buffer zone within a certain width range around the point, line, and surface entities) to obtain the buffer surface, and the buffer surface and the water surface are merged to fill the broken flow gap.

[0071] Step 3: Use the water surface reconstructed in step 2 and DEM data to extract the river channel, which specifically includes the following steps:

[0072] 3-1 River triangulation: Extract skeleton lines from water surface data to form skeleton lines and river units (internal adjacent triangles obtained by constrained Delaunay triangulation of the water surface). The skeleton line is the center line of the water surface.

[0073] The specific operation of river triangulation is as follows: river surface data belongs to narrow patches in map making, which are polygons. The skeleton line of the polygon is the trajectory of points with equal distances to two or more sides of the polygon. Its essence is the boundary set of the minimum distance between each point in the polygon and each side. The nodes of any polygon skeleton line are mainly located at the midpoints of the common sides of adjacent triangles, which are exactly the midpoints of the common sides of the internal adjacent triangles (the triangle is defined as the river unit) obtained by the constrained Delaunay triangulation of any polygon. The principle is as follows Figure 1 .

[0074] The entire process is implemented by programming using Python object-oriented programming language combined with ArcPy site package:

[0075] (1) Use the Create TIN and Convert TIN to Triangle tools in the 3D Analysis Toolbox and combine them with the Feature Vertex to Point tool to generate a constrained Delaunay triangulated mesh of any polygon;

[0076] (2) Use ArcGIS's algorithmic tools such as line splitting at inflection points, feature conversion points, and spatial connection to generate skeleton line nodes;

[0077] The Remove Duplicates and Spatial Join tools post-process the generated skeleton line nodes.

[0078] 3-2 River segmentation: Segment the river according to the skeleton line and river unit setting related parameters.

[0079] The specific operation of river segmentation is: based on river units and skeleton lines, multiple continuous river units are considered to form different river sections based on some points of interest (i.e., artificial intervention points: hydrological station locations, artificial water conservancy facilities, flood storage areas, etc.). The river unit is weak and cannot represent the characteristics of long river sections. After river segmentation, the requirements for computing performance will be reduced and the computing efficiency will be improved. The river segmentation standards are as follows:

[0080] (1) Artificial intervention points such as hydrological stations, artificial water conservancy facilities, flood storage areas, etc. (the rivers in these areas are greatly affected by human activities, so they need to be segmented and analyzed separately).

[0081] (2) Nodes of the river skeleton line, which is the basis for basic segmentation;

[0082] (3) Maximum length of a river section: refers to the maximum length of a river section. The maximum length of a river section is adjusted based on the river skeleton (the degree of meandering of the river). If the river is straight, the maximum length can be larger; if the river is more winding, the maximum length can be smaller.

[0083] (4) Maximum number of river unit sections: This is used to compress and optimize performance based on the curvature of the river, avoid segmenting each node, and make adjustments based on the curvature. The essence of curvature is the number of river units. The more winding the river section, the smaller the maximum number of river unit sections; the straighter the river section, the larger the maximum number of river unit sections.

[0084] (5) Minimum number of tributary units: Set the minimum number of tributary units, and remove those with less than this limit. Optimize the river segmentation, extract the central axis, remove the non-important skeleton and retain the core part.

[0085] 3-3 River channel extraction: Determine the analysis scope according to each river section, then use the natural break point algorithm to extract the river channel based on DEM, and finally mosaic it into a complete river channel.

[0086] The natural breaks algorithm is a data classification method based on the natural groupings inherent in the data. It seeks to minimize the average deviation of each class from the class mean while maximizing the deviation of each class from the mean of other groups, that is, to reduce the variance within the class and maximize the variance between classes, thereby determining the best arrangement of values ​​in different categories. The variance goodness of fit (GVF) is obtained by dividing the difference between the squared deviation of the array mean (SDAM) and the squared deviation of the class mean (SDCM) by SDAM:

[0087]

[0088] SDAM=∑(x i -X) 2

[0089] SDCM=∑∑(x i -Z) 2

[0090] x = array mean; Z = class mean.

[0091] The closer the GVF is to 1, the better the effect is. A GVF of 1 is only possible when the intra-class variation is zero.

[0092] The histogram of DEM terrain has a typical "multi-peak" feature, where each "peak" represents a typical feature. The river classification of this model adopts the "natural break point" algorithm. Features of different elevations are naturally grouped according to the specified number of classifications, and the river group is the group with the smallest elevation in the natural grouping. The number of characteristic features is specified according to the number of characteristic features in the DEM. The characteristic features are reflected in different areas in the DEM histogram, generally 3 to 10.

[0093] In order to accurately and quickly extract the river range, this experiment conducted a study on river range extraction based on DEM data combined with GF-1 data, combined the natural break point algorithm with DEM data, and proposed a new method for river extraction. This method is based on the typical "multi-peak feature" of the DEM histogram and uses the natural break point algorithm to extract DEM rivers.

[0094] Based on the river segmentation, the DEM data is combined to extract the river channel. The study area is determined according to the river section, and then the river channel is extracted for each study area, and finally the complete river channel is mosaicked. Buffer analysis is performed on the river section to determine the study scope of each river section: create a temporary folder to traverse the river section buffer, and each buffer is used to clip the raster. The minimum elevation is extracted for each unit to extract the river channel range. The buffer distance is calculated using the river section, and the minimum bounding geometry is used to determine the minimum width of the river unit. The river unit is buffered using buffer analysis, and a field (buffer distance) is added to the vector. The field is calculated:

[0095] Buffer distance = width × coefficient

[0096] (1) Calculate the minimum polygon and minimum width of the river section;

[0097] (2) Calculate buffer distance: add fields, calculate fields: buffer distance = width * coefficient;

[0098] (3) Calculate the river reach and create a buffer zone;

[0099] (4) Clip the river section DEM according to the range;

[0100] (5) The river channel is extracted in each segment using the natural break point algorithm;

[0101] (6) Calculate the river channel in the study area: mosaic into a new grid and merge the river channel extraction results.

[0102] Embodiment 1:

[0103] Two areas in the upper reaches of the Huaihe River were selected for research. The river water surface data generated by multi-source remote sensing images during the flood season were used to reconstruct the water body. The resulting data were combined with 5mDEM and 12.5mDEM for river channel extraction, and 2m full-color high-definition images were superimposed for visual analysis.

[0104] Results of water body reconstruction in study area 1 ( Figure 3 ) It can be seen that the water surface extraction result of GF-1 data is the best, while the water surface extraction result of Sentinel-2 data is poor and the water surface is small.

[0105] The study area 1 in the upper reaches of the Huaihe River is located in a mixed area of ​​mountainous areas and hills, with a large slope and obvious elevation changes, which is more suitable for classification based on the natural break point algorithm. Figures 4 to 7 It can be seen that the water surface data extracted based on the three types of remote sensing data and the DEM data of two precisions can extract the river range very well, and the river extraction results are basically consistent with the real river. From the perspective of different remote sensing data sources, the results based on the GF-1 remote sensing image can extract the river and conform to the boundary of the actual river. However, the river results of the combination of Sentinel-2 remote sensing images and 12.5mDEM are not extracted at the end of the river section, indicating that the extraction results are affected by the accuracy of the terrain data. The river extraction results based on Sentinel-1 radar data and Sentinel-2 remote sensing image data have the problem that some river sections are narrower than the actual ones. This is affected by the water surface results. The water surface of the water body results of these two data is narrow, and the water surface extraction effect is lower than that of the GF-1 data. The larger the extracted water surface, the larger the range of the final analysis, and the more accurate the result. Therefore, the extraction effect of the water surface data has a certain influence on the extraction results of the river.

[0106] The results of water body reconstruction in the study area 2 in the upper reaches of the Huaihe River (such as Figure 8 ) It can be seen that GF-1 has the best water surface extraction and reconstruction effect, while the results of the other two data are poor. Study area 2 is located in a mountainous area with many puddles and small ditches on both sides. Figure 8 to Figure 11It can be seen that the three remote sensing data combined with the maximum water surface data have a better river channel extraction effect overall. Sentinel-1 and Sentinel-2 data also have unsatisfactory water surface results, and their corresponding river channel results are narrower than the actual river channel. The result boundary of the GF-1 remote sensing data is not smooth enough, which may be affected by the multi-spectral image of the same object with different spectra, resulting in irregular river channel boundaries, but the overall river channel range is extracted completely. The Sentinel-2 water surface results and the maximum water surface results are combined with 12.5mDEM to extract the river channel. The results have irregular boundaries. On the one hand, they are affected by the resolution of DEM. On the other hand, the existing DEMs are all made from multi-source remote sensing data. There is inevitably interference information in the DEM, which leads to inaccurate river channel extraction results.

[0107] On the whole, the river extraction method combined with water surface data is more applicable to rivers in mountainous areas.

[0108] Evaluation results of river extraction accuracy in this example (mountainous area):

[0109] Verification points were evenly selected in the upper reaches of the Huaihe River in Study Area 1 and Study Area 2, and the verification points were divided into river channels and non-river channels. Visual interpretation was performed using Google Maps and 2m high-precision full-color satellite images to determine whether the verification points belonged to the river channel area. There were a total of 10,320 verification points in Study Area 1 (e.g. Fig.18 ), of which 518 are rivers and 9802 are non-rivers; there are 16170 verification points in the study area 2 (such as Fig.19 ), of which 450 are river channels and 15720 are non-river channels. The confusion matrix of the result data was calculated to obtain the overall accuracy, user accuracy, producer accuracy and Kappa coefficient, and the accuracy evaluation tables of river channel extraction results based on different accuracy DEM combined with multi-source remote sensing image data were obtained for study area 1 (Tables 1-4) and study area 2 (Tables 11-14).

[0110] Table 1. Precision evaluation of 1GF1 river channel extraction results in the upper Huaihe River study area

[0111]

[0112] The results of river channel extraction based on water surface data extracted from GF-1 remote sensing image data (Table 7) show that the overall accuracy of the extraction results has reached above 0.98, the Kappa coefficient of 5mDEM is 0.8765, and the Kappa coefficient of 12.5mDEM is 0.8767; the producer accuracy of the river channel area has reached above 0.90, the user accuracy is above 0.80, the missed error is less than 0.1, and the misjudgment error is less than 0.2. Among them, the misjudgment error of the river channel based on 12.5mDEM is greater than that of 5mDEM.

[0113] Table 2 Precision evaluation of river channel extraction results of 1Sentinel-1 in the upper Huaihe River study area

[0114]

[0115]

[0116] The results of river channel extraction based on water surface data extracted from Sentinel-1 remote sensing image data (Table 2) show that the overall accuracy of the extraction results has reached above 0.98, the Kappa coefficient of 5mDEM is 0.8104, and the Kappa coefficient of 12.5mDEM is 0.8369; the producer accuracy of the river channel area is 0.7046 and 0.7471 respectively, the user accuracy is above 0.95, and the misjudgment error is below 0.05. This is because the water surface data is small, the corresponding buffer analysis range is small, the obtained river channel range is relatively narrow, and the misjudgment error of the river channel area is correspondingly small. The missed error and misjudgment error based on Sentinel1 remote sensing image data are greater than the accuracy evaluation of GF-1, which is related to the effect of river water surface extraction. The river water surface extracted based on GF-1 remote sensing image is better than the extraction result of Sentinel1 remote sensing image.

[0117] Table 3 Results of 1Sentinel-2 river channel extraction in the upper Huaihe River study area

[0118]

[0119] The results of river channel extraction based on water surface data extracted from Sentinel-2 remote sensing image data in Study Area 1 (Table 3) show that the overall accuracy of the extraction results has reached above 0.98, the Kappa coefficient of 5mDEM is 0.8836, and the Kappa coefficient of 12.5mDEM is 0.8089. The misjudgment error of the river channel area of ​​12.5mDEM is 0.2029, while the misjudgment error of 5mDEM is only 0.0974. The corresponding user accuracy of the river channel area of ​​the former is 0.7971, which is much smaller than 0.9026 of 5mDEM. When the water surface data is not ideal, the extraction results are more affected by the resolution.

[0120] Table 4 Extraction results of the largest river surface in the study area 1 in the upper reaches of the Huaihe River

[0121]

[0122]

[0123] The results of river channel extraction based on the maximum water surface data in study area 1 (Table 4) show that the Kappa coefficients of the river channel extraction results of the two precision DEMs are greater than those of the other three cases. The Kappa coefficient of the river channel results extracted by 5mDEM is 0.8869, and the Kappa coefficient of the river channel results extracted by 12.5mDEM is 0.8865. This is because the larger the range of water surface data, the larger the algorithm analysis range, and the more accurate the river channel extraction results.

[0124] Table 5 2GF-1 river channel extraction results in the upper Huaihe River study area

[0125]

[0126] The results of river channel extraction based on water surface data extracted from GF-1 remote sensing image data in study area 2 (Table 5) show that the overall accuracies of 5mDEM and 12.5mDEM are 0.9854 and 0.9891 respectively; the Kappa coefficients are 0.7382 and 0.7935 respectively.

[0127] Table 6 2Sentinel-1 river channel extraction results in the upper Huaihe River study area

[0128]

[0129]

[0130] The results of river channel extraction based on water surface data extracted from Sentinel-1 remote sensing image data in Study Area 2 (Table 6) show that the overall accuracy of the two precision DEMs is above 0.98, but the Kappa coefficient is lower than the accuracy result of GF-1, among which the Kappa coefficient of 5mDEM is 0.5736 and the Kappa coefficient of 12.5mDEM is 0.6473. This is because the water surface data results of Sentinel-1 remote sensing images in Study Area 2 are poor, and the extracted river channel range is also smaller than the actual river channel range.

[0131] Table 7 Accuracy evaluation of 2Sentinel-2 river channel extraction results in the upper Huaihe River study area

[0132]

[0133] The results of river channel extraction based on water surface data extracted from Sentinel-2 remote sensing image data in Study Area 2 (Table 7) show that the overall accuracy of both DEMs is above 0.98, the Kappa coefficient of 5mDEM is 0.7738, and the Kappa coefficient of 12.5mDEM is 0.71007, which is better than the accuracy of Sentinel-1. The analysis is that the water body extraction results of Sentinel-2 in Study Area 2 are better than those of Sentinel-1. The error of the river channel area misjudgment rate of 12.5mDEM is 0.3443, and the error of 5mDEM is 0.1311. Since the river channel result range of 12.5mDEM is larger, it exceeds the actual river too much, thus causing errors.

[0134] Table 8 Accuracy evaluation of the extraction results of the largest river surface channel in the upper Huaihe River study area 2

[0135]

[0136]

[0137] The results of river channel extraction based on the maximum river surface data in study area 2 (Table 8) show that the overall accuracy of the two DEMs is above 0.98, the Kappa coefficient of 5mDEM is 0.7935, and the Kappa coefficient of 12.5mDEM is 0.7742, which shows that when the water surface data is good enough, the river channel extraction results are also affected by the DEM resolution.

[0138] Example 2

[0139] The Dawen River Basin has a variety of terrains, including mountains, hills, and plains. However, many areas of the river have low water volume all year round. Therefore, during the river surface extraction process, many tributaries cannot be extracted due to the influence of river water volume and resolution. The final water surface result is only the water surface of the main river of the Dawen River. The plain area selected in the Dawen River is in an area with abundant water volume, so the water surface extraction effect is naturally very good (such as Fig.13 ). From the perspective of DEMs with different resolutions, the river channel extraction results of 5mDEM are better than those of 12.5m, which also verifies that the higher the accuracy of DEM data, the higher the accuracy of river channel extraction. The extraction results of 12.5mDEM ( Fig.14 (a2), Fig.15 (a2), Fig.16 (a2), Fig.17(a2)) have rough and uneven boundaries and fail to extract the real river channel. However, the 5mDEM data results extract areas outside the river channel as the river channel. This is because the plain area has a flat and low-lying terrain. When classifying the natural discontinuities of the DEM, it is easy to extract only the main channel of the river channel, but not the riverbed and beach. On the other hand, it is also easy to classify the flat and low-lying areas around the river as the river channel.

[0140] In summary, in mountainous areas, due to obvious changes in elevation, the river extraction results are likely to be larger than the actual river range, while in plain areas, due to the insignificant changes in elevation, it is easy to extract a small river or extract surrounding fields during classification. Since the plain area studied has abundant water, the method of the present invention is mainly related to the resolution of DEM when the water surface data is good.

[0141] Evaluation results of river extraction accuracy in this example:

[0142] Verification points were established by evenly selecting points in the Dawen River Basin ( Fig. 20 ), the verification points are divided into river channels and non-river channels, and visual interpretation is performed in combination with Google Maps and 2m high-precision full-color satellite images to determine whether the verification points belong to the river channel area. There are 39,496 verification points in the basin, including 1,498 river channels and 37,998 non-river channels; the area in the yellow box in the figure below is selected as the plain river research area. There are 966 verification points in the research area, including 74 river channels and 892 non-river channels. The confusion matrix of the result data is calculated to obtain the overall accuracy, user accuracy, producer accuracy and Kappa coefficient, and the accuracy evaluation table of the river extraction results based on different precision DEM combined with multi-source remote sensing image data in the Dawen River area is obtained.

[0143] Results of river channel accuracy evaluation for the entire Dawen River basin:

[0144] Table 9 Accuracy evaluation of Dawen River GF-1 river channel extraction results

[0145]

[0146]

[0147] Table 10 Precision evaluation of river channel extraction results of Sentinel-1 in Dawen River

[0148]

[0149] Table 11 Precision evaluation of Sentinel-2 river channel extraction results of Dawen River

[0150]

[0151] Table 12 Accuracy evaluation of the extraction results of the largest river surface channel in Dawen River

[0152]

[0153]

[0154] Tables 9 to 12 show that the overall accuracy of the main river channels in the Dawen River Basin is above 0.98, and the Kappa coefficient is above 0.75, which meets the better Kappa coefficient accuracy standard (0.61-0.8). The misjudgment error in the river channel area is below 0.15, and the missed judgment error in the river channel area does not exceed 0.31.

[0155] Results of river channel accuracy evaluation in the Dawen River Plain area:

[0156] The results of river channel accuracy evaluation in the Dawen River Plain area are shown in Tables 13 to 16.

[0157] Table 13 Accuracy evaluation of GF-1 river channel extraction results in Dawen River Plain

[0158]

[0159] The results of river extraction based on GF-1 remote sensing image water surface data in the Dawen River Plain (Table 13) show that the overall accuracy of 5mDEM is 0.9824, the overall accuracy of 12.5mDEM is 0.9638, the Kappa coefficient of 5mDEM is 0.8647, and the Kappa coefficient of 12.5mDEM is 0.6730. This is because 12.5mDEM does not recognize the fields in the river as the river, while the field area data in the river and the beach area in the river are farmland but still part of the river.

[0160] Table 14 Precision evaluation of Sentinel-1 river channel extraction results in the Dawen River Plain

[0161]

[0162]

[0163] The results of the trace extraction based on the water surface data of Sentinel-1 remote sensing images in the Dawen River Plain (Table 14) show that the overall accuracy of 5mDEM is 0.9565, the overall accuracy of 12.5mDEM is 0.9586, the Kappa coefficient of 5mDEM is 0.5914, and the Kappa coefficient of 12.5mDEM is 0.6172. The accuracy difference between the two is not large, which is also related to the small number of selected points in the area.

[0164] Table 15 Precision evaluation of Sentinel-2 river channel extraction results in Dawen River Plain

[0165]

[0166] The channel extraction results based on Sentinel-2 remote sensing image water surface data in the Dawen River Plain (Table 15) show that the overall accuracy of 5mDEM is 0.9658, the overall accuracy of 12.5mDEM is 0.9627, the Kappa coefficient of 5mDEM is 0.7338, and the Kappa coefficient of 12.5mDEM is 0.6663. The channel extraction results of 5mDEM are better than those of 12.5m. This is because the edges of the 12.5mDEM river channel results are rough and the range of the extraction results is larger than the range of the real river channel.

[0167] Table 16 Accuracy evaluation of the maximum water surface river extraction results in the Dawen River Plain

[0168]

[0169]

[0170] The channel extraction results based on Sentinel-2 remote sensing image water surface data in the Dawen River Plain (Table 16) show that the overall accuracy of 5mDEM is 0.9669, the overall accuracy of 12.5mDEM is 0.9658, the Kappa coefficient of 5mDEM is 0.7289, and the Kappa coefficient of 12.5mDEM is 0.6965. The effect of 5mDEM river channel extraction is better than that of 12.5m. Analysis shows that the river channel range of some river sections in 12.5mDEM is slightly narrower than the real river channel.

[0171] According to the river extraction effects in different regions and different data sources, the water body extraction effect has a greater impact on the river range extraction. Among them, the river extraction effect combined with GF-1 remote sensing image data is better than that of Sentinel-1 and Sentinel-2. When the water surface data is very good or very poor, the accuracy of DEM has a greater impact on the effect. The higher the accuracy, the better the river extraction effect.

[0172] According to the water body and river channel extraction effects in the Dawen River Basin, the method of the present invention is more suitable for rivers with water flow or river channel width greater than 10m. And the method of the present invention is more suitable in mountainous areas, which is affected by topographic factors and the mountainous terrain changes significantly. The "natural break point" algorithm can better separate the river channel part, while the shallows and low-lying areas in the plains will affect the extraction of the river channel. Therefore, this method is more suitable for river channel extraction in mountainous and hilly areas, and can also be applied in plains, but the accuracy is not as good as that in hilly areas.

[0173] Finally, it should be noted that the above is only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred arrangement scheme, a person skilled in the art should understand that the technical solution of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A river channel extraction method based on water surface data, characterized in that: The following steps are involved: Step 1: Water surface extraction: Use multispectral data and / or radar data to extract water surface using the water index method. Use dilation and erosion to process the redundant patches and interruptions in the water surface extraction results, and perform vectorization of single ground feature grids to obtain water surface data. The parameters include: the minimum number of pixels in the ground object patch, patches with more than the specified number of pixels are considered as water surface areas, and the gap radius, gaps smaller than the specified radius will be filled; Step 2: Water surface reconstruction: Perform water surface reconstruction based on mathematical morphology on the water surface data obtained in step 1 to obtain complete water surface data; The water surface reconstruction specifically includes the following steps: 2-1 Artificially draw multi-segment elements at the breakwater based on water surface data and remote sensing images; 2-2 Perform spatial connection analysis on multi-line elements and water surface data to remove redundant patches; 2-3 Take the average water surface width at both ends of the interrupted flow to perform buffer analysis on multiple line segments. The buffer width of the multi-line segments between the dry river sections is calculated by the average river width of the river sections at both ends, where the average river width calculation formula at both ends is W = 2 × S / L. Where: W: river section width; S: river section area; L: river section perimeter; the sum of the areas S1+S2 of the river sections at both ends is S, and the sum of the perimeters L1+L2 of the river sections at both ends is L. After the width is obtained, a buffer analysis is performed on the line segment. The buffer analysis refers to automatically establishing a buffer zone within a certain width range around the point, line, and surface entities based on the point, line, and surface entities to obtain a buffer surface, and merge the buffer surface and the water surface to fill the broken flow gap; Step 3: Use the water surface reconstructed in step 2 and DEM data to extract the river channel, which specifically includes the following steps: 3-1 River triangulation: The water surface data is triangulated using the Delaunay triangulation algorithm to extract the skeleton line and obtain the skeleton line and river unit; the river unit is the triangle obtained by triangulation, and the skeleton line is the center line of the water surface; 3-2 River segmentation: Segment the river according to the skeleton line and river unit setting related parameters; the setting related parameters include: artificial intervention points, nodes of the river skeleton line, maximum length of the river section, maximum number of river section units and minimum number of tributary units; 3-3 River channel extraction: Buffer analysis is performed on each river section to obtain the buffer range and thus determine the analysis range. Then, the natural break point algorithm is used to extract the river channel based on the DEM analysis range of each river section. Finally, the river channel extraction results of each segment are mosaicked into a complete river channel. In the buffer analysis: the buffer distance is calculated using the river section, the minimum bounding geometry is used to determine the minimum width of the river unit, and the buffer zone analysis is used to buffer the river unit. Buffer distance = width × coefficient.

2. The river channel extraction method based on water surface data according to claim 1 is characterized in that: In step 1: select one of the GF-1 multispectral data, Sentinel-1 radar data, and Sentinel-2 multispectral data of the flood season in the study area to extract the river water surface and obtain the water surface data as the water surface data for subsequent river channel extraction; or select the GF-1 multispectral data, Sentinel-1 radar data, and Sentinel-2 multispectral data of the flood season in the study area to extract the river water surface respectively, and obtain the maximum water surface of the river in the flood season from different data sources, and merge the results to obtain the final maximum river water surface data as the water surface data for subsequent river channel extraction.

3. The river channel extraction method based on water surface data according to claim 1 is characterized in that: The DEM data in step 3 is 5mDEM data or 12.5mDEM data.

4. The river channel extraction method based on water surface data according to claim 1 is characterized in that: In step 3-3, the natural break point algorithm is used based on DEM to extract the river channel, wherein the histogram of the DEM terrain has a "multi-peak" feature, and each "peak" represents a type of typical landform; landforms at different elevations are naturally grouped according to the specified number of classifications, and the river channel group is the group with the smallest elevation in the natural grouping, and the river channel range is obtained by extracting the minimum grouping range.