Method and system for extracting river network remote sensing information at basin scale with high temporal and spatial resolution

Abstract

The present application belongs to the technical field of space-to-earth observation (satellite remote sensing) and watershed hydrology, geographical environment disciplines, and discloses a watershed scale high spatio-temporal resolution river network remote sensing information extraction method and system. After cloud and cloud shadow processing of optical remote sensing images, a simple non-iterative clustering superpixel segmentation algorithm is used to segment homogeneous landscape objects, and a hierarchical decision tree is constructed using a water body index and a vegetation index. In the area where the optical remote sensing image is seriously covered by clouds or the image is missing, the SAR image is filtered, the dual polarization water body index is calculated using the VV and VH polarization images, the maximum inter-class variance method is executed, and the image is binarized. After obtaining the water body result extracted from the remote sensing image, the river cut-off is repaired in combination with the water system diagram generated by the DEM model. The present application can automatically extract accurate, continuous and high spatio-temporal resolution river network remote sensing information, and can be expanded to global scale river network feature extraction.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of space-based Earth observation (satellite remote sensing) and watershed hydrology and geographical environment, and in particular relates to a method and system for extracting remote sensing information of river networks at a watershed scale with high spatiotemporal resolution. Background Technology

[0002] River networks are a crucial component of the surface water cycle, capable of storing and transporting large quantities of surface materials. The characteristics of river networks (river width, surface area, slope, etc.) form the basis of river hydrology research, reflecting the interactions between river hydrological processes, biogeochemical cycles, river ecosystems, and human activities. For a long time, under the combined influence of climate change and human activities, the water cycle and land surface processes at the watershed scale have been undergoing profound changes. The formation, distribution, and evolution of river networks exhibit significant spatiotemporal variability and high uncertainty, which will impact the sustainable development of watersheds.

[0003] However, watershed-scale river network information suffers from slow updates and low resolution. Existing research mainly focuses on large rivers with widths greater than 30m, lacking studies on smaller rivers ranging from 1 to 30m, and consequently lacking high-precision, complete studies on the spatiotemporal distribution of river system characteristics. Satellite remote sensing imagery has become an important means of rapidly acquiring surface water information, providing the possibility for accurately extracting river network information and monitoring and understanding river dynamics. Therefore, utilizing remote sensing imagery to obtain accurate and continuous river network information, achieving monthly, meter-level resolution dynamic monitoring of watershed river networks, and analyzing the spatiotemporal distribution and changes of watershed-scale river system characteristics will provide high-precision, low-cost observational data products for watershed research. This will help understand the patterns of watershed water resource changes, improve prediction and regulation capabilities, and provide a scientific basis for ensuring watershed water security and ecological security.

[0004] Currently available global and regional river network datasets are primarily extracted from Digital Elevation Models (DEMs) or remote sensing images. Developing hydrological maps using DEMs mainly involves calculating the flow direction for each pixel, as exemplified by the HydroSHEDS and MERIT Hydro datasets. HydroSHEDS is a global hydrological dataset obtained from SRTM elevation data, with a resolution of 3 arcseconds. Yamazaki et al. generated the MERIT Hydro dataset, effectively addressing the limited coverage of high-latitude watersheds. However, DEMs typically contain non-negligible vertical errors, which can distort the topographic slope used to estimate flow direction, and DEMs are slow to update, failing to reflect the dynamic changes in the river network.

[0005] Another mainstream technique utilizes remote sensing data to obtain long-term frequency, extent, and dynamic changes of river networks at a global scale. For example, Pekel et al. used Google Earth Engine (GEE) and Landsat remote sensing imagery to create a 30-meter resolution Global Surface Water (GSW) dataset and mapped the location and frequency changes of surface water at a global scale over the past 38 years. Allen and Pavelsky constructed a Global River Widths from Landsat Database (GRWL) dataset based on Landsat remote sensing imagery and estimated the total surface area of ​​rivers and streams with a width ≥ 30 meters. Compared to Landsat imagery, commercial optical remote sensing imagery typically has higher spatial resolution and richer spectral information, but it is also affected by cloud cover when identifying water bodies. Therefore, using a single optical remote sensing dataset can lead to missing data in some areas due to cloud cover, resulting in flow interruptions in the river network extraction results.

[0006] Based on the above analysis, the problems and shortcomings of the existing technology are as follows:

[0007] (1) The problems and defects of using DEM to extract river network technology are: DEM data update speed is slow and cannot reflect the dynamic changes of river network;

[0008] (2) The problems and defects of using remote sensing data to extract river network technology are: optical remote sensing data are easily affected by cloud cover, which can cause rivers to stop flowing. Furthermore, existing technologies lack research on small rivers of 1 to 30 meters, resulting in a lack of high-precision and complete spatiotemporal distribution research on the characteristics of river systems. Summary of the Invention

[0009] To overcome the problems existing in related technologies, the present invention discloses a method and system for extracting high spatiotemporal resolution remote sensing information of river networks at the watershed scale. The technical solution is as follows:

[0010] This invention is implemented as follows: a method for extracting remote sensing information of river networks with high spatiotemporal resolution at the watershed scale, comprising the following steps:

[0011] S1. After removing clouds and cloud shadows from the optical remote sensing image, the SNIC superpixel segmentation algorithm is used to segment homogeneous landscape objects, and water index features are added during segmentation.

[0012] S2. Based on the obtained image segmentation objects, analyze the time series features of Automatic Water Extraction Index (AWEI), Improved Normalized Difference Water Index (MNDWI), Normalized Difference Vegetation Index (NDVI), and Enhanced Vegetation Index (EVI), construct a hierarchical decision tree classification model, and remove paddy fields from the results.

[0013] S3, mark the pixels in the missing areas of the optical image as nodata, select the SAR image and use the sensor to transmit vertically polarized waves, and receive vertically polarized wave VV or horizontally polarized wave VH images, calculate the dual-polarized water index SDWI; use the Otsu method to threshold segment the SAR image to obtain the binarized water extraction result and replace the nodata pixels in the optical image.

[0014] S4. Use DEM data to fill depressions, use the D8 algorithm to calculate flow direction and estimate the cumulative flow; construct a river network model to generate river buffer zones; fuse the river network model generated based on DEM data with the river network results extracted from remote sensing images, and directionally connect the broken river networks to generate a continuous river network.

[0015] In step S1, the optical remote sensing image is processed to remove clouds and cloud shadows, including: performing bit operations on the cloud layer using the QA60 band of Sentinel-2 data to mask the area covered by the cloud layer.

[0016] In step S1, homogeneous landscape objects are segmented using the SNIC superpixel segmentation algorithm, including:

[0017] (1) Initialize seed points: Divide the image with N pixels into K superpixels of the same size, and build a priority queue to store the K created elements;

[0018] (2) Cluster center update and distance measurement: Select the next pixel to be added to the superpixel cluster through a priority queue and return the candidate pixel with the smallest distance from the centroid.

[0019] Furthermore, the priority queue contains 4 or 8 candidate pixels connected to the currently growing superpixel cluster, and each new pixel added to the superpixel cluster is used to perform the cluster center update.

[0020] Furthermore, in the initialization of seed points, the distance from the centroid of the Kth superpixel to the jth candidate pixel is calculated as follows:

[0021]

[0022] In the formula, d j，k Let X be the distance from the j-th pixel to the k-th centroid. j Let X be the spatial coordinates of the j-th pixel. kFor the space of the Kth centroid, C j Let C be the 3D color feature vector of the j-th pixel in the CIELAB color space. k Let be the 3D color feature vector of the CIELAB color space with the Kth centroid, s be the normalization factor of spatial distance and color distance, and m be the normalization factor of color distance.

[0023] In step S2, the time series characteristics of the Automatic Water Extraction Index (AWEI), the Modified Normalized Difference Water Index (MNDWI), the Normalized Difference Vegetation Index (NDVI), and the Enhanced Vegetation Index (EVI) are analyzed, and the expressions are as follows:

[0024] AWEI nsh =4×(ρ Green -ρ SWIR1 )-(0.25×ρ NIR +2.75×ρ SWIR2 )

[0025] AWEI sh =ρ Blue +2.5×ρ Green -1.5×(ρ NIR +ρ SWIR1 -0.25×ρ SWIR2

[0026]

[0027]

[0028]

[0029] In the formula, the AWEI index is divided into AWEI based on the differences in noise sources that cause errors in different land cover types. nsh and AWEI sh AWEI nsh It can effectively eliminate non-water pixels on dark building surfaces in urban background areas, and is suitable for situations where shadows are not the main source of noise; AWEI sh Applicable to areas where shadows are the primary source of noise; MNDWI is the Modified Normalized Difference Water Index, EVI is the Enhanced Vegetation Index, NDVI is the Normalized Difference Vegetation Index, and ρ Green For the green band, ρ Blue For the blue band, ρ SWIR1 For shortwave infrared band 1, ρ NIR For the near-infrared band, ρ SWIR2 It is a shortwave infrared band 2, ρ Red It is the red band.

[0030] In step S3, SAR images are selected and vertically polarized waves are transmitted using a sensor. Vertically polarized wave (VV) or horizontally polarized wave (VH) images are received, and the dual-polarization water index SDWI is calculated, including:

[0031] SAR images with imaging times close to those of optical images are selected and refined Lee filtering is applied to suppress speckle noise. SDWI is calculated using vertically polarized wave (VV) and horizontally polarized wave (VH) polarized images to amplify the differences between water bodies and non-water bodies, resulting in a bimodal distribution of the water body histogram.

[0032] Furthermore, SDWI is calculated using vertically polarized wave (VV) and horizontally polarized wave (VH) polarization images, with the following expression:

[0033] SDWI = ln(10 × VV × VH) - 8

[0034] In the formula, SDWI is the dual polarization water index, VV is the vertical polarization wave emitted and received by the sensor, and VH is the vertical polarization wave emitted and received by the sensor.

[0035] In step S4, DEM data is used to fill depressions and calculate the D8 flow direction, including: in a 3×3 neighborhood, the gradient between the center cell and the eight surrounding neighboring cells is calculated using the D8 algorithm, and the direction with the largest gradient is determined as the flow direction; the cumulative flow is calculated by assigning cells with a cumulative value greater than or equal to 1 to the water flow and cells with a cumulative value of 0 to the ridgeline.

[0036] Another objective of this invention is to provide a watershed-scale high spatiotemporal resolution river network remote sensing information extraction system. This system implements the aforementioned watershed-scale high spatiotemporal resolution river network remote sensing information extraction method. The system includes:

[0037] The image segmentation module performs cloud removal and cloud shadow processing on the optical remote sensing image, and then uses the SNIC superpixel segmentation algorithm to segment homogeneous landscape objects, while incorporating water index features during segmentation.

[0038] The classification model construction module is used to segment objects based on the obtained images, analyze the time series features of Automatic Water Extraction Index (AWEI), Improved Normalized Difference Water Index (MNDWI), Normalized Difference Vegetation Index (NDVI), and Enhanced Vegetation Index (EVI), construct a hierarchical decision tree classification model, and remove paddy fields from the results.

[0039] The SDWI calculation module marks pixels in the missing areas of the optical image as nodata, selects SAR images and uses the sensor to transmit vertically polarized waves, and receives vertically polarized wave VV or horizontally polarized wave VH images to calculate the dual-polarization water index SDWI.

[0040] The image segmentation module is used to perform threshold segmentation on SAR images using the Otsu method to obtain binarized water body extraction results and replace nodata pixels in the optical images.

[0041] The river buffer generation module is used to fill depressions with DEM data, calculate flow direction using the D8 algorithm, estimate confluence accumulation, construct a river network model, and generate river buffers.

[0042] The continuous river network generation module is used to fuse the river network model generated based on DEM data with the river network results extracted from remote sensing images, and directionally connect the broken river networks to generate a continuous river network.

[0043] Combining all the above technical solutions, the advantages and positive effects of this invention are as follows: The method for extracting river network remote sensing information by fusing optics, synthetic aperture radar (SAR), and digital elevation model (DEM) provided by this invention extracts water bodies from optical and SAR images on an Earth big data cloud computing platform, and is divided into two parts: First, cloud removal and cloud shadow processing are performed on the optical remote sensing images, and a simple non-iterative clustering (SNIC) superpixel segmentation algorithm is used to segment homogeneous landscape objects. A hierarchical decision tree is constructed using water body indices and vegetation indices to reduce the impact of other surface water bodies such as paddy fields on river network extraction. Second, in areas where the optical remote sensing images are severely covered by clouds or where the images are missing, the SAR images are filtered, and the dual-polarized water body index (Sentinel-1 Dual-Polarized) is calculated using VV and VH polarization images. The WaterIndex (SDWI) is used to amplify the difference between water bodies and other land features, and then the Otsu's method (OTSU) is performed to binarize the image. After obtaining the water body results extracted from the remote sensing image, the river system map generated by the DEM model is combined to repair river flow interruptions and reduce background noise. The river network remote sensing information extraction method of this invention can automatically extract accurate, continuous, and high spatiotemporal resolution river network remote sensing information. It is suitable for extracting river network remote sensing information at the watershed scale and can also be extended to extract river network features at the global scale.

[0044] SAR imagery is capable of operating in all weather conditions, penetrating cloud cover and vegetation cover. In areas where optical imagery data is severely obscured by cloud cover, it can provide alternative observational information and improve the accuracy of results.

[0045] By combining superpixel segmentation with hierarchical decision trees, we can improve data processing efficiency and apply it to the rapid extraction of large-scale water bodies. Hierarchical decision tree classification, on the other hand, considers the temporal variation characteristics of water bodies and can effectively distinguish surface water bodies or other land features that are easily confused with rivers.

[0046] By leveraging the constraints of DEM data on river flow direction, broken river segments extracted from remote sensing images can be connected to improve the connectivity of the river network.

[0047] The technical solution of this invention enables monthly dynamic monitoring of the Yellow River basin's river network at a resolution of 10m. This allows for timely, accurate, quantitative, and comprehensive acquisition of the spatial distribution and area variation patterns of the Yellow River basin's river network, providing high-precision, low-cost observational data products for Yellow River basin research. This contributes to understanding the patterns of water resources and sediment changes in the Yellow River basin, enhancing prediction and regulation capabilities, and providing a scientific basis for the implementation of the national strategy for ecological protection and high-quality development of the Yellow River basin, as well as ensuring water and ecological security. Furthermore, this technical solution has the potential to be extended to extract river networks at a global scale, contributing to a better understanding of the global river system's response mechanisms to climate change and human activities.

[0048] Current domestic and international technologies mainly focus on extracting data from large rivers with a width greater than 30m, lacking research on smaller rivers with a width of 1-30m. Furthermore, the temporal resolution is mostly around one year, resulting in a lack of high spatiotemporal resolution and comprehensive research on the spatiotemporal distribution of river network characteristics. The technical solution of this invention can achieve monthly monitoring of dynamic changes in river networks at a resolution of 10m.

[0049] Rivers extracted from remote sensing images are prone to flow interruptions, while rivers extracted from DEMs have a slow update rate and cannot reflect dynamic changes in the river. The technical solution of this invention combines remote sensing images with DEMs, effectively overcoming the technical bias inherent in extracting rivers from a single data source. Attached Figure Description

[0050] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure;

[0051] Figure 1 This is a flowchart of the method for extracting remote sensing information of river networks at a watershed scale with high spatiotemporal resolution provided in an embodiment of the present invention;

[0052] Figure 2 This is a schematic diagram of the Yellow River Basin, an example study area provided in this embodiment of the invention;

[0053] Figure 3 This is a flowchart illustrating the technical operation of an example study provided in this embodiment of the invention;

[0054] Figure 4 This is the water body threshold segmentation model provided in this embodiment of the invention, wherein water body and non-water body sample points are respectively located in (a) AWEI nsh (b)AWEI nsh With (MNDWI-NDVI), (c)AWEI sh With (MNDWI-EVI), (d)AWEI sh The scatter plots of (MNDWI-NDVI) show the water and non-water sample points at (e) (MNDWI-EVI), (f) (MNDWI-NDVI), and (g) AWEI, respectively. nsh (h)AWEI sh Frequency distribution diagram;

[0055] Figure 5 This is a hierarchical decision tree classification model provided in this embodiment of the invention, where SD represents standard deviation, TS represents time series, and T1 is determined by calculating the SD value of the NDVI time series of the river sample.

[0056] Figure 6 This is a schematic diagram illustrating the temporal variation analysis of NDVI in the Yellow River (YR) and Paddy Fields (PF) provided in the embodiments of the present invention.

[0057] Figure 7 This is a schematic diagram of example results of each step of superpixel segmentation and hierarchical decision tree provided in the embodiments of the present invention; wherein, (a) is Sentinel-2 image, (b) is AWEI image, (c) is SNIC result, (d) is water extraction result, (e) is paddy field removal result, and (f) is final river result;

[0058] Figure 8 This is a schematic diagram illustrating the principle of fusion of Sentinel-2 and DEM modeling of river networks provided in this embodiment of the invention.

[0059] Figure 9 This is a comparison chart of the results obtained by the method of this invention and previous methods in different land cover types (a) urban areas and (b) mountainous areas, provided by the embodiments of this invention; wherein (c) and (g) are the results obtained by the method of this invention, (d) and (h) are the results obtained by the method of Li et al., (e) and (i) are the results obtained by the method of Zou et al., and (f) and (j) are the results obtained by the Active-Passive Surface Water Classification (APWC) method;

[0060] Figure 10 This is a comparison chart of the results obtained by using the method of this invention with existing river datasets provided in the embodiments of this invention; the base map shows the comparison between the Yellow River Basin river network extraction results and the existing Global River Widths from Landsat Database (GRWL); (a), (c) and (e) are displays with the existing Global Surface Water (GSW) dataset; (b), (d) and (f) are the Yellow River Basin river network extraction results using this method; (g) is the confusion matrix for accuracy evaluation of this method; (h) is the comparison of the Yellow River Basin river network extraction results using this method with the GRWL dataset and the Global River Network (GRN) dataset in terms of river length; and (i) is the relationship between the surface area of ​​the Yellow River Basin river network and precipitation, temperature and evaporation. Detailed Implementation

[0061] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0062] Example 1, such as Figure 1 As shown, the method for extracting high spatiotemporal resolution remote sensing information of river networks at the watershed scale provided in this embodiment of the invention includes the following steps:

[0063] S1. After removing clouds and cloud shadows from the optical remote sensing image, the SNIC superpixel segmentation algorithm is used to segment homogeneous landscape objects, and water index features are added during segmentation.

[0064] S2. Based on the obtained image segmentation objects, analyze the time series features of Automatic Water Extraction Index (AWEI), Improved Normalized Difference Water Index (MNDWI), Normalized Difference Vegetation Index (NDVI), and Enhanced Vegetation Index (EVI), construct a hierarchical decision tree classification model, and remove paddy fields from the results.

[0065] S3, mark the pixels in the missing areas of the optical image as nodata, select the SAR image and use the sensor to transmit vertically polarized waves, and receive vertically polarized wave VV or horizontally polarized wave VH images, calculate the dual-polarized water index SDWI; use the Otsu method to threshold segment the SAR image to obtain the binarized water extraction result and replace the nodata pixels in the optical image.

[0066] S4. Use DEM data to fill depressions, use the D8 algorithm to calculate flow direction and estimate the cumulative flow; construct a river network model to generate river buffer zones; fuse the river network model generated based on DEM data with the river network results extracted from remote sensing images, and directionally connect the broken river networks to generate a continuous river network.

[0067] Specifically, it includes:

[0068] (1) The optical remote sensing image (taking Sentinel-2 as an example) was processed to remove clouds and cloud shadows. The QA60 band of Sentinel-2 data was used, which can perform bit operations on clouds and mask the area covered by clouds. Then, by analyzing the difference in reflection intensity of water bodies and their surrounding heterogeneous backgrounds in different band images, the SNIC superpixel segmentation algorithm was used to segment homogeneous landscape objects. Water body index features were added during segmentation to improve image segmentation accuracy.

[0069] SNIC is an improved version of Simple Linear Iterative Clustering (SLIC). It is a method that combines color similarity and spatial proximity. The main idea is to convert RGB images to the CIELAB color space, and to form a 5-dimensional feature vector by combining the three color components and two coordinate values ​​of each pixel in the image. Then, the pixels are locally clustered by measuring the distance between the vectors.

[0070] The specific steps for image segmentation using SNIC superpixels are as follows:

[0071] ① Initialize seed points: Divide the image with N pixels into K superpixels of the same size, and build a priority queue to store the K created elements;

[0072] ② Cluster Center Update and Distance Measurement: A priority queue is used to select the next pixel to add to the superpixel cluster. The priority queue contains 4 or 8 candidate pixels connected to the currently growing superpixel cluster, and returns the candidate pixel with the smallest distance to the centroid. Each new pixel added to the superpixel cluster is used to update the cluster center. Therefore, unlike SLIC, which requires multiple K-means iterations to update the cluster centers, SNIC can complete the cluster center update in one iteration. The distance from the centroid of the Kth superpixel to the jth candidate pixel is calculated as follows:

[0073]

[0074] In the formula, d j，k Let X be the distance from the j-th pixel to the k-th centroid. j Let X be the spatial coordinates of the j-th pixel.k For the space of the Kth centroid, C j Let C be the 3D color feature vector of the j-th pixel in the CIELAB color space. k Let be the 3D color feature vector of the CIELAB color space with the Kth centroid; s is the normalization factor for spatial distance and color distance, and m is the normalization factor for color distance.

[0075] (2) Various surface water bodies and other land features may affect the correct extraction of river networks, such as aquaculture ponds and paddy fields. They generally have similar spectral and water color characteristics to rivers, which can easily cause confusion and reduce the extraction accuracy. Based on the image segmentation objects obtained in step (1), the time series features of their Automated Water Extraction Index (AWEI), Modified Normalized Difference Water Index (MNDWI), Normalized Difference Vegetation Index (NDVI), and Enhanced Vegetation Index (EVI) are analyzed. A hierarchical decision tree classification model is constructed, and paddy fields and aquaculture ponds are removed from the results.

[0076] The time-series characteristics of the Automatic Water Extraction Index (AWEI), the Modified Normalized Difference Water Index (MNDWI), the Normalized Difference Vegetation Index (NDVI), and the Enhanced Vegetation Index (EVI) were analyzed, and the expression is as follows:

[0077] AWEI nsh =4×(ρ Green -ρ SWIR1 )-(0.25×ρ NIR +2.75×ρ SWIR2 )

[0078] AWEI sh =ρ Blue +2.5×ρ Green -1.5×(ρ NIR +ρ SWIR1 -0.25×ρ SWIR2

[0079]

[0080]

[0081]

[0082] In the formula, the AWEI index is divided into AWEI based on the differences in noise sources that cause errors in different land cover types. nsh and AWEI sh AWEI nsh It can effectively eliminate non-water pixels on dark building surfaces in urban background areas, making it suitable for situations where shadows are not the primary source of noise; AWEI sh Primarily applicable to areas where shadows are the main source of noise. MNDWI is the Modified Normalized Difference Water Index, EVI is the Enhanced Vegetation Index, NDVI is the Normalized Difference Vegetation Index, and ρ... Green For the green band, ρ Blue For the blue band, ρ SWIR1 For shortwave infrared band 1, ρ NIR For the near-infrared band, ρ SWIR2 It is a shortwave infrared band 2, ρ Red It is the red band.

[0083] Taking Sentinel-2 multispectral imagery as an example, ρ Blue , ρ Green , ρ Red , ρ NIR , ρ SWIR1 , ρ SWIR2 They are bands 2, 3, 4, 8, 11, and 12, respectively.

[0084] This invention collects a large number of water body sample points and non-water body sample points, and generates scatter density maps of water bodies and non-water bodies (see...). Figure 4 99.81% of the non-aquatic sample points were located in areas where MNDWI-EVI < 0, and 92.39% of the water sample points were located in areas where MNDWI-EVI > 0 (see...). Figure 4 e). 99.83% of non-aquatic sample points had MNDWI-NDVI < 0, and 91.43% of aquatic sample points had MNDWI-NDVI > 0 (see...). Figure 4 f). 95.31% of the water sample points had an AWEIsh > -0.6 (see f). Figure 4 g), 91.21% of the water sample points had AWEIsh>0 (see g). Figure 4 h). To address this, this invention proposes a novel threshold segmentation algorithm for water body extraction from optical images. Pixels meeting the conditions (AWEIsh > -0.6 or AWEIsh > 0) and (MNDWI > EVI or MNDWI > NDVI) are classified as water bodies, while other pixels are classified as non-water bodies. The spectral and color characteristics of paddy fields and aquaculture ponds are similar to those of rivers, easily causing confusion and reducing the accuracy of river network extraction. The NDVI time-series characteristics of water body targets are analyzed to remove paddy fields from the results.

[0085] (3) Based on steps (1) and (2), pixels in the missing areas of the optical image (taking Sentinel-2 as an example) are marked as nodata. SAR images (taking Sentinel-1 as an example) with similar imaging times to the optical images are selected. After performing refined Lee filtering to suppress speckle noise, SDWI is calculated using VV and VH polarimetric images to amplify the differences between water and non-water bodies, resulting in a bimodal distribution of the water body histogram. Subsequently, OTSU is used to perform threshold segmentation on the SAR images to obtain binarized water body extraction results, which are then used to replace the nodata pixels in the optical images. This step uses SAR images as a supplement to optical images in areas with cloud cover or missing data, compensating for the susceptibility of optical images to cloud cover, ensuring complete image coverage of the entire watershed every month, and improving temporal resolution.

[0086] The SDWI is calculated using VV and VH polarimetric images, expressed as follows:

[0087] SDWI = ln(10 × VV × VH) - 8

[0088] In the formula, SDWI is the dual polarization water index, VV is the vertical polarization wave emitted and received by the sensor, and VH is the vertical polarization wave emitted and received by the sensor.

[0089] (4) Fill depressions using DEM data and perform flow direction analysis using the D8 algorithm. The D8 algorithm calculates the gradient between the center pixel and its eight neighboring pixels in a 3×3 neighborhood, and the direction of the largest gradient is determined as the flow direction. For example... Figure 3 As shown, the values ​​for each direction starting from the center are 1, 2, 4, 8, 16, 32, 64, and 128, respectively. For example, if the direction of maximum gradient is to the left of the current processing unit, then its flow direction will be encoded as 16. Calculating the confluence accumulation involves assigning cells with an accumulation value greater than or equal to 1 to water flow and cells with an accumulation value of 0 to ridgelines. This is used to construct a river network model and generate river buffers. This step effectively reduces errors caused by mountain shadows and maximizes the integrity of water body information.

[0090] (5) According to step (4), the river network model generated based on DEM data is fused with the river network results extracted from remote sensing images, and the broken river network is connected in a directional manner to finally generate an accurate continuous river network.

[0091] like Figure 8 As shown, the river origin is located using the river network extracted from remote sensing images. The starting and ending points of river segment 1 in the DEM model connect the beginning and end of the two broken river lines extracted from remote sensing images, which increases the river connectivity and should be retained. However, river segment 2 in the DEM, as a tributary, does not increase the river connectivity and should be deleted.

[0092] The watershed-scale high spatiotemporal resolution river network remote sensing information extraction system provided in this embodiment of the invention includes:

[0093] The image segmentation module is used to segment homogeneous landscape objects using the SNIC superpixel segmentation algorithm after removing clouds and cloud shadows from optical remote sensing images, and to incorporate water index features during segmentation.

[0094] The classification model building module is used to segment objects from images, analyze the time series features of Automatic Water Extraction Index (AWEI), Improved Normalized Difference Water Index (MNDWI), Normalized Difference Vegetation Index (NDVI), and Enhanced Vegetation Index (EVI), construct a hierarchical decision tree classification model, and remove paddy fields from the results.

[0095] The SDWI calculation module is used to mark pixels in the missing areas of optical images as nodata, select SAR images, and calculate SDWI using VV and VH polarization images.

[0096] The image segmentation module is used to perform threshold segmentation on SAR images using OTSU to obtain binarized water body extraction results and replace nodata pixels in optical images.

[0097] The river buffer generation module is used to fill depressions with DEM data, calculate the D8 flow direction, estimate the cumulative runoff volume, construct a river network model, and generate river buffers.

[0098] The continuous river network generation module is used to fuse the river network model generated based on DEM data with the river network results extracted from remote sensing images, and directionally connect the broken river networks to generate a continuous river network.

[0099] Example 2

[0100] Taking the Yellow River Basin as an example, this paper analyzes in detail the process and advantages of a method for extracting high spatiotemporal resolution continuous river network remote sensing information at the basin scale. It compares the method with previous proposed river network extraction methods and river network datasets to evaluate the accuracy of the proposed method. Using Sentinel-1 SAR and Sentinel-2 optical remote sensing data as examples, this paper proposes an algorithm that combines superpixel segmentation with a hierarchical decision tree classification model, leveraging the constraint of terrain on river flow direction to achieve rapid extraction of accurate, continuous, and high spatiotemporal resolution river network remote sensing information at the basin scale.

[0101] like Figure 2 As shown, the Yellow River is China's second longest river. It originates in the Bayan Har Mountains of the Qinghai-Tibet Plateau, meanders eastward across the Loess Plateau, and flows into the Bohai Sea. The Yellow River is 5,464 km long, with a total drainage area of ​​795,000 km². 2The Yellow River, with only 2.2% of the nation's runoff, supplies water to 15% of the country's arable land and 12% of its population. It also bears the responsibility of sediment transport, a task not typically undertaken by clear-water rivers. For a long time, under the combined influence of climate change and human activities, the Yellow River basin's water cycle and land surface processes have been undergoing profound changes. The formation, distribution, and evolution of its river network exhibit significant spatiotemporal variability and high uncertainty, which will affect the basin's sustainable development. However, information on the Yellow River basin's river network suffers from slow updates and low resolution. Existing analyses mainly focus on large rivers with widths greater than 30 meters, lacking research on smaller rivers ranging from 1 to 30 meters, thus lacking high-precision, complete spatiotemporal distribution analysis of the river system characteristics.

[0102] like Figure 3 As shown, the technical process of this invention consists of two parts. First, cloud removal and cloud shadow processing are performed on Sentinel-2 imagery. The SNIC superpixel segmentation algorithm is used to segment homogeneous landscape objects, and a hierarchical decision tree is constructed using water and vegetation indices to reduce the impact of other surface water bodies such as paddy fields and aquaculture ponds on river network extraction. Second, in areas with severe cloud cover or missing images in the Sentinel-2 imagery, Sentinel-1 SAR imagery is used as a substitute. SDWI is calculated, and the OTSU method is executed to binarize the imagery. After obtaining the water body extraction results from the remote sensing imagery, a river system map generated from the DEM model is combined to repair river flow interruptions and reduce background noise. Finally, the accuracy of the extraction results is evaluated by comparing with existing datasets.

[0103] like Figure 4 As shown, this invention collected 19,998 sample points, including 10,093 water sample points and 9,905 non-water sample points, and generated a scatter density map of water and non-water bodies.

[0104] like Figure 5 , Figure 6 and Figure 7 As shown, AWEI, MNDWI, NDVI, and EVI indices were added during SNIC superpixel segmentation. After obtaining the segmented land cover classifications, a threshold segmentation model was used to maximize the retention of water bodies. The NDVI time-series characteristics of water bodies were analyzed to remove paddy fields from the results.

[0105] like Figure 8 As shown, the river starting point is located using the river network extracted from remote sensing images, and the river segments modeled by DEM are used to directionally connect the beginning and end of the river breaks extracted from remote sensing images, thereby increasing river connectivity.

[0106] like Figure 9As shown, compared with previous methods, the method of this invention can effectively suppress shadow noise caused by excessively tall buildings or small building spacing when extracting rivers in urban areas. When extracting rivers in mountainous areas, this method can identify small rivers and maximize the preservation of the integrity of the river network.

[0107] As shown in Table 1, the overall accuracy of the method of the present invention reaches 95.77%, and the Kappa coefficient is 0.92. Compared with the Li, Zou and APWC methods, its accuracy is significantly improved.

[0108] Table 1. Accuracy comparison results of this invention with other algorithms

[0109]

[0110] like Figure 10 As shown, the river network results of the Yellow River Basin are overlaid with existing GRWL and GSW datasets to qualitatively evaluate the spatial distribution and detail of the river network. The results show that the method of this invention can more effectively enhance the contrast between small rivers and their surrounding background, and can extract more small rivers. After classifying the river network results of the Yellow River Basin, the river lengths are compared with those of the GRWL and GRN datasets. Overall, the total river length obtained by the method of this invention is 28,587 km longer than that of the GRWL dataset and 19,389 km longer than that of the GRN dataset. The lengths of rivers of level 3 and above are significantly greater than those in existing datasets, indicating that the results of this invention have better coverage of medium and small rivers.

[0111] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0112] The information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0113] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this invention. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments.

[0114] According to embodiments of this application, the present invention also provides a computer device comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the steps in any of the above-described method embodiments.

[0115] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps described in the various method embodiments above.

[0116] This invention also provides an information data processing terminal, which, when executed on an electronic device, provides a user input interface to implement the steps described in the above method embodiments. The information data processing terminal is not limited to mobile phones, computers, or switches.

[0117] This invention also provides a server that, when executed on an electronic device, provides a user input interface to implement the steps described in the above method embodiments.

[0118] This invention also provides a computer program product that, when run on an electronic device, enables the electronic device to implement the steps described in the various method embodiments above.

[0119] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying the computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0120] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0121] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention and within the spirit and principles of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for extracting high spatiotemporal resolution remote sensing information of river networks at the watershed scale, characterized in that, The method includes the following steps: S1. After removing clouds and cloud shadows from the optical remote sensing image, the SNIC superpixel segmentation algorithm is used to segment homogeneous landscape objects, and water index features are added during segmentation. S2. Based on the obtained image segmentation objects, analyze the time series features of Automatic Water Extraction Index (AWEI), Improved Normalized Difference Water Index (MNDWI), Normalized Difference Vegetation Index (NDVI), and Enhanced Vegetation Index (EVI), construct a hierarchical decision tree classification model, and remove paddy fields from the results. S3, mark the pixels in the missing areas of the optical image as nodata, select the SAR image and use the sensor to transmit vertically polarized waves, and receive vertically polarized wave VV or horizontally polarized wave VH images, calculate the dual-polarized water index SDWI; use the Otsu method to threshold segment the SAR image to obtain the binarized water extraction result and replace the nodata pixels in the optical image. S4. Use DEM data to fill depressions, use the D8 algorithm to calculate flow direction and estimate the cumulative flow; construct a river network model and generate river buffer zones; fuse the river network model generated based on DEM data with the river network results extracted from remote sensing images, and directionally connect the broken river networks to generate a continuous river network. In step S1, homogeneous landscape objects are segmented using the SNIC superpixel segmentation algorithm, including: (1) Initialize seed points: Divide the image with N pixels into K superpixels of the same size, and build a priority queue to store the K created elements; (2) Cluster center update and distance measurement: Select the next pixel to be added to the superpixel cluster through a priority queue, and return the candidate pixel with the smallest distance from the centroid; In step S4, DEM data is used to fill depressions and calculate the D8 flow direction, including: in a 3×3 neighborhood, the gradient between the center cell and the eight surrounding neighboring cells is calculated using the D8 algorithm, and the direction with the largest gradient is determined as the flow direction; the cumulative flow is calculated by assigning cells with a cumulative value greater than or equal to 1 to the water flow and cells with a cumulative value of 0 to the ridgeline.

2. The method for extracting high spatiotemporal resolution remote sensing information of river networks at the watershed scale according to claim 1, characterized in that, In step S1, the optical remote sensing image is processed to remove clouds and cloud shadows, including: performing bit operations on the cloud layer using the QA60 band of Sentinel-2 data to mask the area covered by the cloud layer.

3. The method for extracting high spatiotemporal resolution remote sensing information of river networks at the watershed scale according to claim 2, characterized in that, The priority queue contains 4 or 8 candidate pixels connected to the currently growing superpixel cluster. Each new pixel added to the superpixel cluster is used to perform the cluster center update.

4. The method for extracting high spatiotemporal resolution remote sensing information of river networks at the watershed scale according to claim 1, characterized in that, In the initialization of seed points, the distance from the centroid of the Kth superpixel to the jth candidate pixel is calculated as follows: ； In the formula, Let be the distance from the j-th pixel to the k-th centroid. Let j be the spatial coordinates of the j-th pixel. For the space of the Kth centroid, Let j be the 3D color feature vector of the CIELAB color space for the j-th pixel. Let be the 3D color feature vector of the CIELAB color space with the Kth centroid, s be the normalization factor for spatial distance and color distance, and m be the normalization factor for color distance.

5. The method for extracting high spatiotemporal resolution remote sensing information of river networks at the watershed scale according to claim 1, characterized in that, In step S2, the time series characteristics of the Automatic Water Extraction Index (AWEI), the Modified Normalized Difference Water Index (MNDWI), the Normalized Difference Vegetation Index (NDVI), and the Enhanced Vegetation Index (EVI) are analyzed, and the expressions are as follows: ； In the formula, the AWEI index is divided into AWEI based on the differences in noise sources that cause errors in different land cover types. nsh and AWEI sh AWEI nsh It can effectively eliminate non-water pixels on dark building surfaces in urban background areas, and is suitable for situations where shadows are not the main source of noise; AWEI sh Applicable to areas where shadow is the primary noise source; MNDWI is the Modified Normalized Difference Water Index, EVI is the Enhanced Vegetation Index, and NDVI is the Normalized Difference Vegetation Index. Green band For blue band, It is the shortwave infrared band 1. It is in the near-infrared band. It is a shortwave infrared band 2. It is the red band.

6. The method for extracting high spatiotemporal resolution remote sensing information of river networks at the watershed scale according to claim 1, characterized in that, In step S3, SAR images are selected and vertically polarized waves are transmitted using a sensor. Vertically polarized wave (VV) or horizontally polarized wave (VH) images are received, and the dual-polarization water index SDWI is calculated, including: SAR images with imaging times close to those of optical images are selected and refined Lee filtering is applied to suppress speckle noise. SDWI is calculated using vertically polarized wave (VV) and horizontally polarized wave (VH) polarized images to amplify the differences between water bodies and non-water bodies, resulting in a bimodal distribution of the water body histogram.

7. The method for extracting high spatiotemporal resolution remote sensing information of river networks at the watershed scale according to claim 6, characterized in that, SDWI is calculated using vertically polarized wave (VV) and horizontally polarized wave (VH) polarization images. The expression is: ； In the formula, SDWI is the dual polarization water index, VV is the vertical polarization wave emitted and received by the sensor, and VH is the vertical polarization wave emitted and received by the sensor.

8. A watershed-scale, high spatiotemporal resolution remote sensing information extraction system for river networks, characterized in that, This system implements the watershed-scale high spatiotemporal resolution river network remote sensing information extraction method as described in any one of claims 1 to 7, and the system includes: The image segmentation module performs cloud removal and cloud shadow processing on the optical remote sensing image, and then uses the SNIC superpixel segmentation algorithm to segment homogeneous landscape objects, while incorporating water index features during segmentation. The classification model construction module is used to segment objects based on the obtained images, analyze the time series features of Automatic Water Extraction Index (AWEI), Improved Normalized Difference Water Index (MNDWI), Normalized Difference Vegetation Index (NDVI), and Enhanced Vegetation Index (EVI), construct a hierarchical decision tree classification model, and remove paddy fields from the results. The SDWI calculation module marks pixels in the missing areas of the optical image as nodata, selects SAR images and uses the sensor to transmit vertically polarized waves, and receives vertically polarized wave VV or horizontally polarized wave VH images to calculate the dual-polarization water index SDWI. The image segmentation module is used to perform threshold segmentation on SAR images using the Otsu method to obtain binarized water body extraction results and replace nodata pixels in the optical images. The river buffer generation module is used to fill depressions with DEM data, calculate flow direction using the D8 algorithm, estimate confluence accumulation, construct a river network model, and generate river buffers. The continuous river network generation module is used to fuse the river network model generated based on DEM data with the river network results extracted from remote sensing images, and directionally connect the broken river networks to generate a continuous river network.

Images