Mountainous river step-pool bed pattern intelligent recognition and classification method and device

Abstract

The present application provides a mountainous river step-pool riverbed morphology intelligent recognition and classification method and device, the method comprising: obtaining an orthographic image terrain data containing mountainous river step-pool characteristics; extracting a channel and a shoreline based on a high-precision digital elevation model and a digital orthographic image, and further identifying the range of the target river based on the extracted channel and shoreline; extracting river feature indicators from the high-precision digital elevation model; extracting sediment particle characteristics from the digital orthographic image; constructing a multi-dimensional raster data set based on the river feature indicators, the sediment particle characteristics, and the high-precision digital elevation model and the digital orthographic image; inputting the multi-dimensional raster data set into a deep learning model to obtain a classification mark. The present application effectively solves the core defects of low automation, poor recognition accuracy, and inability to support large-scale applications in the prior art, and realizes a technical breakthrough from manual semi-automatic interpretation to full-process intelligent recognition and classification of step-pool structures.

Description

Technical Field

[0001] This invention relates to the fields of geomorphological mapping of mountain rivers, hydrological sediment monitoring, and riverbed stability assessment, and particularly to a method and device for intelligent identification and classification of the morphology of stepped-deep pool riverbeds in mountainous areas. Background Technology

[0002] Under the long-term effects of water erosion, sediment transport, and riverbed adjustment, mountain rivers naturally develop a stepped-deep pool structure. This structure is a key geomorphic unit for regulating riverbed resistance, efficiently dissipating energy, stabilizing the riverbed, and maintaining the stability of the river channel and aquatic habitat. It plays an important role in controlling the stability of the riverbed, the evolution of the river channel, and the ecological restoration of mountain rivers.

[0003] Existing technologies for surveying and interpreting stepped-deep-pool structures mainly involve manual on-site investigations or the use of drones to collect river images and generate high-precision digital elevation models (DEMs) and orthophoto topographic data. Subsequent geomorphological analysis is then completed by a combination of manual and semi-automated software, with technicians manually selecting topographic features and labeling stepped-deep-pool structures, followed by basic analysis using simple topographic algorithms.

[0004] However, existing technologies still have significant shortcomings in subsequent geomorphological interpretation, structure identification, and engineering applications, mainly including: First, it cannot support the rapid identification and automated interpretation of the stepped-deep pool structure in long river sections and large areas, resulting in low efficiency and poor applicability. Second, existing algorithms have limited ability to delineate the boundaries and determine the morphology of the stepped-deep pool area, resulting in low accuracy in structure recognition, a tendency to make false or false judgments, and large fluctuations in classification results, which cannot provide reliable basic data for subsequent analysis. Summary of the Invention

[0005] This specification provides a method and device for intelligent identification and classification of the morphology of stepped-deep pool riverbeds in mountainous areas, in order to solve the problems of low identification efficiency, insufficient identification accuracy, and poor stability of classification results in existing technologies for stepped-deep pool structures.

[0006] To address the aforementioned technical problems, this specification provides a first aspect of an intelligent identification and classification method for the morphology of mountain river steps and deep pools, comprising: acquiring orthophoto topographic data containing features of mountain river steps and deep pools, wherein the orthophoto topographic data includes a digital orthophoto image and a high-precision digital elevation model (DEM) linked by unified geographic coordinates; extracting channels and shorelines based on the DEM and the DEM, and further identifying the range of the target river channel based on the extracted channels and shorelines; extracting river channel feature indicators from the DEM; extracting sediment particle features from the DEM; constructing a multidimensional raster dataset based on the river channel feature indicators, the sediment particle features, the DEM, and the DEM; and inputting the multidimensional raster dataset into a deep learning model to obtain the classification labels of the steps and deep pools output by the deep learning model.

[0007] In some embodiments, a multidimensional raster dataset is constructed based on the river channel feature indicators, the sediment particle characteristics, the high-precision digital elevation model, and the digital orthophoto, including: The river channel feature indicators, sediment particle features, and river channel feature indicators, sediment particle features, geographic location coordinates, and pixel values ​​associated with the same grid in the high-precision digital elevation model and digital orthophoto image are used to construct the multidimensional raster data corresponding to the grid according to a preset format; the multidimensional raster data associated with each grid corresponding to the target river channel constitute a multidimensional raster dataset.

[0008] In some embodiments, extracting river feature indicators from the high-precision digital elevation model includes: determining the river width value of each grid cell associated with each cross section of the river channel by: retrieving abrupt topographic change shoreline along the direction of the target cross section, determining the boundary points of the shorelines on both sides, and calculating the distance between the boundary points of the shorelines on both sides as the river width value of the target cross section, and using the river width value as the river width value of each grid cell associated with the target cross section; determining the curvature value of each grid cell associated with each cross section of the river channel by: fitting the elevation sequence of the target cross section; and determining the curvature value of the target cross section based on the fitting coefficient; wherein, the cross section of the river channel is a section perpendicular to the longitudinal profile of the river channel.

[0009] In some embodiments, extracting sediment particle features from the digital orthophoto image includes: using an automatic sediment particle identification model to identify exposed sediment particles on the riverbed in the digital orthophoto image; and extracting sediment particle features based on the sediment particle identification results, wherein the sediment particle features include particle spatial coordinates, equivalent particle size, and distribution density.

[0010] In some embodiments, the method further includes: assessing the stability of the target river channel based on the classification results of each step-pool of the target river channel.

[0011] In some embodiments, the stability calculation method includes: assigning stability quantification values ​​to each type of step-pool in the target channel, and mapping the stability quantification values ​​of each type of step-pool to a grid; and calculating the overall stability assessment value of the target channel based on the stability quantification values ​​corresponding to each grid of each step-pool in the target channel.

[0012] In some embodiments, the overall stability assessment value of the target river is calculated based on the stability quantification values ​​corresponding to each grid cell of each step-pool type structure in the target river channel. This includes: weighting and summing the step-pool type structures of the target river channel using the stability quantification value of each step-pool type structure as the weight and the number of grid cells corresponding to that step-pool type structure as the weighting factor; and normalizing the summation result using the maximum stability quantification value of the step-pool and the total number of grid cells in the target river channel to obtain the overall stability assessment value of the target river channel.

[0013] In some embodiments, channels are extracted based on the high-precision digital elevation model and the digital orthophoto as follows: The D8 eight-direction unidirectional flow algorithm is executed based on the high-precision digital elevation model and the digital orthophoto, traversing the entire grid, calculating the flow direction from each grid cell to its eight neighboring grid cells, and generating a standardized flow direction grid; correspondingly, shorelines are extracted based on the high-precision digital elevation model and the digital orthophoto as follows: Based on the standardized flow direction grid, the upstream confluence area of ​​each grid is calculated using a cumulative confluence algorithm to obtain a cumulative confluence grid; a binary river grid is determined based on a river confluence cumulative threshold; and the centerline of the main channel is determined based on the binary river grid.

[0014] In some embodiments, the extent of the target river channel is further identified based on the extracted ditch and shoreline in the following manner: perpendicular to the centerline of the ditch, a grid traversal is performed to the left and right banks of the river channel at a fixed step size, and the local topographic slope and second-order difference of elevation are calculated for each grid; grids that meet the following conditions are marked as river channel shoreline control points: the grid slope exceeds the slope change threshold, the second-order difference of elevation changes from negative to positive or from negative to positive, and the grid topographic parameters exceed the topographic parameters of the gentle riverbed section by N times the standard deviation, where N is a positive integer greater than 1; The control points on both banks of the river are connected in sequence according to the direction of the river channel, and the line elements are smoothed to form a closed river channel boundary line. Based on the river channel boundary line, the high-precision digital elevation model is cropped to obtain the target river channel high-precision digital elevation model. Based on the river channel boundary line, the digital orthophoto is cropped to obtain the target river channel digital orthophoto.

[0015] The second aspect of this specification provides an intelligent identification and classification device for the morphology of mountain river steps and deep pools, comprising: an acquisition unit for acquiring orthophoto topographic data containing the features of mountain river steps and deep pools, the orthophoto topographic data including a digital orthophoto image and a high-precision digital elevation model associated with unified geographic coordinates; a first extraction unit for extracting channels and shorelines based on the high-precision digital elevation model and the digital orthophoto image, and further identifying the range of the target river channel based on the extracted channels and shorelines; a second extraction unit for extracting river channel feature indicators from the high-precision digital elevation model; a third extraction unit for extracting sediment particle features from the digital orthophoto image; a construction unit for constructing a multidimensional raster dataset based on the river channel feature indicators, the sediment particle features, the high-precision digital elevation model, and the digital orthophoto image; and a processing unit for inputting the multidimensional raster dataset into a deep learning model to obtain the classification labels of steps and deep pools output by the deep learning model.

[0016] In some embodiments, the construction unit includes: a first constitutive subunit, used to construct multidimensional raster data corresponding to the raster by combining the river channel feature indicators, the sediment particle features, the high-precision digital elevation model, and the river channel feature indicators, sediment particle features, geographic location coordinates, and pixel values ​​associated with the same raster in the digital orthophoto according to a preset format; and a second constitutive subunit, used to construct a multidimensional raster dataset by constructing multidimensional raster data from the multidimensional raster data associated with each raster corresponding to the target river channel.

[0017] In some embodiments, the second extraction unit includes: a first determining subunit, configured to determine the river width value of each grid associated with each cross section of the river channel by: retrieving the abrupt change shoreline along the target cross section direction, determining the boundary points of the two shorelines, calculating the distance between the boundary points of the two shorelines as the river width value of the target cross section, and using the river width value as the river width value of each grid associated with the target cross section; The second determining subunit is used to determine the curvature value of each grid cell associated with each cross section of the river channel by: fitting a high-order sequence of the target cross section; and determining the curvature value of the target cross section based on the fitting coefficients; wherein the cross section of the river channel is a section perpendicular to the longitudinal profile of the river channel.

[0018] In some embodiments, the third extraction unit includes: an identification subunit for identifying exposed sediment particles on the riverbed in the digital orthophoto using an automatic sediment particle identification model; and an extraction subunit for extracting sediment particle features based on the sediment particle identification results, wherein the sediment particle features include particle spatial coordinates, equivalent particle size, and distribution density.

[0019] In some embodiments, the apparatus further includes an evaluation unit for evaluating the stability of the target river channel based on the classification results of each step-pool of the target river channel.

[0020] In some embodiments, the evaluation unit includes: an allocation subunit, configured to assign stability quantification values ​​to each type of step-pool in the target channel and process the stability quantification values ​​of each type of step-pool to a grid; and a first calculation subunit, configured to calculate the overall stability evaluation value of the target channel based on the stability quantification values ​​corresponding to each grid of each step-pool in the target channel.

[0021] In some embodiments, the first calculation subunit includes: a summation subunit, used to perform weighted summation of the stepped-deep pool type structures of the target channel using the stability quantification value of each stepped-deep pool type structure as a weight and the number of grids corresponding to the stepped-deep pool type structure as a weighting factor; and a first processing subunit, used to perform normalization processing by combining the total number of stepped-deep pool type structures with the total length of the channel to obtain the overall stability assessment value of the target channel.

[0022] In some embodiments, the first extraction unit includes: a second calculation subunit, configured to perform a D8 eight-direction unidirectional flow algorithm based on the high-precision digital elevation model and the digital orthophoto, traverse the entire grid, calculate the water flow direction from each grid cell to its eight neighboring grid cells, and generate a standardized water flow direction grid; a third calculation subunit, configured to calculate the upstream confluence area of ​​each grid cell using a cumulative confluence algorithm based on the standardized water flow direction grid, and obtain a cumulative confluence grid; a third determination subunit, configured to determine a binary river grid based on a river confluence cumulative threshold; and a fourth determination subunit, configured to determine the centerline of the main channel of the river based on the binary river grid.

[0023] In some embodiments, the first extraction unit includes: a fourth calculation subunit, used to perform a grid traversal perpendicular to the channel centerline, moving towards both banks of the river at a fixed step size, and calculating the local terrain slope and second-order elevation difference for each grid cell; a marking subunit, used to mark grid cells that meet the following conditions as riverbank control points: the grid slope exceeds the slope abrupt change threshold, the second-order elevation difference changes from negative to positive or vice versa, and the grid terrain parameters exceed the terrain parameters of the gentle riverbed section by N times the standard deviation, where N is a positive integer greater than 1; a second processing subunit, used to connect the control points on both banks of the river in sequence according to the river direction, and smooth the line features to form a closed river channel boundary line; a first clipping subunit, used to clip the high-precision digital elevation model based on the river channel boundary line to obtain a target river channel high-precision digital elevation model; and a second clipping subunit, used to clip the digital orthophoto based on the river channel boundary line to obtain a target river channel digital orthophoto.

[0024] A third aspect of this specification provides an electronic device, comprising: a memory and a processor, wherein the processor and the memory are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to realize the intelligent identification and classification method for the morphology of mountain river steps-deep pools as described in the first aspect.

[0025] The fourth aspect of this specification provides a computer storage medium storing computer program instructions, which, when executed, implement the steps of the intelligent identification and classification method for the morphology of mountain river steps-deep pools as described in the first aspect.

[0026] The fifth aspect of this specification provides a computer program product comprising a computer program that, when executed by a processor, implements the steps of the intelligent identification and classification method for the morphology of mountain river steps-deep pools as described in the first aspect.

[0027] The intelligent identification and classification of mountain river step-deep pool riverbed morphology provided in this manual effectively solves the core defects of existing technologies, such as low automation, poor identification accuracy, and inability to support large-scale applications. It achieves a technological breakthrough from manual semi-automatic interpretation to full-process intelligent identification and classification of step-deep pool structures. The specific technical effects are reflected in the following four aspects.

[0028] 1. Enables fully automated and rapid identification of long river sections and large-scale stepped-deep pool structures, significantly improving surveying efficiency and applicability.

[0029] This invention constructs a fully automated processing flow from data input to classification result output, completely eliminating the existing technology's reliance on manual screening of terrain features and marking of structural boundaries. It can complete the identification of the stepped-deep pool structure of the entire target river channel without the need for full technical personnel intervention, significantly reducing labor and time costs. It solves the problem that existing technologies cannot support rapid interpretation of long river sections and large areas, and can adapt to the needs of basin-level and large-scale geomorphological surveys in mountainous rivers.

[0030] 2. Significantly improves the accuracy of step-deep pool structure recognition and boundary positioning precision, ensuring the stability and reliability of classification results.

[0031] This invention innovatively integrates four types of data—high-precision digital elevation models, digital orthophotos, river channel feature indicators, and sediment particle characteristics—to construct a multi-dimensional raster dataset. This dataset comprehensively characterizes the core features of the stepped-deep pool structure from four dimensions: topographic morphology, surface texture, river channel morphology, and sediment distribution, overcoming the shortcomings of insufficient information from a single data source. By employing a deep learning semantic segmentation model to learn the nonlinear relationships between multi-dimensional features, it effectively improves the accuracy of boundary delineation and morphological determination of the stepped-deep pool structure, significantly reducing the probability of misclassification and missed classification. This solves the problems of low recognition accuracy and large fluctuations in classification results in existing technologies, providing reliable basic data support for subsequent riverbed stability assessment, river regulation, and ecological restoration.

[0032] 3. Output standardized raster-level classification results to lay the foundation for subsequent engineering applications and quantitative analysis.

[0033] The stepped-deep pool classification identifiers output by this invention are raster-level data that spatially match the original terrain and image data. This can directly achieve a one-to-one correspondence between structural types and spatial locations. It can not only clearly present the spatial distribution pattern of stepped-deep pools, but also seamlessly connect with subsequent stages such as quantitative calculation of river stability, risk zoning, and engineering scheme design. This solves the problem that the output results of existing technologies cannot be directly used for quantitative analysis and engineering implementation.

[0034] 4. Highly compatible, requiring no changes to existing drone data acquisition processes.

[0035] This invention directly processes standard digital orthophotos and high-precision digital elevation model data collected by existing UAVs, without the need to upgrade hardware or change the original data acquisition specifications. It can be quickly integrated into the existing mountain river survey operation system and has good engineering practicality and promotion value. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0037] Figure 1 This document presents a flowchart illustrating an intelligent identification and classification method for the stepped-deep-pool riverbed morphology of mountainous rivers. Figure 2 This is a schematic diagram of a process for extracting shorelines based on the high-precision digital elevation model and the digital orthophoto provided in this specification. Figure 3This is a flowchart illustrating a process for further identifying the scope of a target river channel based on the extracted channels and shorelines, as provided in this specification. Figure 4 This is a schematic diagram of a process for extracting river feature indicators from the high-precision digital elevation model provided in this specification. Figure 5 This is a flowchart illustrating a process for constructing a multidimensional raster dataset based on the river channel feature indicators, the sediment particle characteristics, the high-precision digital elevation model, and the digital orthophoto. Figure 6 This is a flowchart illustrating a method for calculating the stability of the target river channel provided in this specification. Figure 7 This specification provides a schematic diagram of the principle of an intelligent identification and classification device for the morphology of stepped-deep-pool riverbeds in mountainous areas. Figure 8 This is a structural block diagram of an electronic device provided in this specification. Detailed Implementation

[0038] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0039] This invention proposes an automated algorithm that integrates topographic and sediment particle information to acquire high-precision digital elevation models and high-resolution orthophoto data of mountain rivers using methods such as drones without modifying the original acquisition process. This algorithm can quickly identify and classify stepped-deep pool structures, and the classification results have good stability, which can support further automated calculation of river channel stability.

[0040] Specifically, this specification provides an intelligent identification and classification method for the morphology of mountain river steps and deep pools, such as... Figure 1 As shown, it includes the following steps S10 to S60.

[0041] S10: Obtain orthophoto topographic data containing the features of mountainous river steps and deep pools. The orthophoto topographic data includes digital orthophotos and high-precision digital elevation models linked by unified geographic coordinates.

[0042] Orthophoto topographic data is a set of digital datasets consisting of digital orthophotos (DOM) and high-precision digital elevation models (DEM). The two are linked by unified geographic coordinates and respectively represent the planar shape of surface features and the elevation of bare land. The DOM is a distortion-free vertical projection image that has been corrected for topography and projection and contains accurate geographic coordinates; the DEM is an elevation model that has eliminated interference from ground features and records surface elevation information.

[0043] The data dimensions of a digital orthophoto DOM are (X, Y, 3), where X and Y are the coordinates of the Earth's surface plane, used to locate the plane position; the number 3 represents the three dimensions, which are the three color channel data: R (red), G (green), and B (blue). The three primary colors are superimposed to synthesize a color image, thereby intuitively reflecting the true color and texture information of the Earth's surface features.

[0044] The data dimension of a high-precision digital elevation model (DEM) is (X, Y, 1), where X and Y are the surface plane coordinates used to locate the plane position; the number 1 represents a dimension, which is the single-band elevation information dimension. (X, Y, 1) together constitute the data dimension form of the DEM representing the surface elevation.

[0045] S20: Extract the channel and shoreline based on the high-precision digital elevation model and the digital orthophoto, and further identify the range of the target river channel based on the extracted channel and shoreline.

[0046] A gully is a linear low-lying terrain formed by the long-term erosion of surface water, serving as the main channel for water collection and transport.

[0047] The shoreline refers to the dynamic boundary between the river water and the surrounding land.

[0048] The scope of the target river channel refers to the continuous water body and tidal flat area enclosed by the shorelines on both banks, including the main channel, providing basic boundary conditions for subsequent hydrological and hydraulic analysis, river regulation engineering design, flood risk assessment, etc.

[0049] In some embodiments, channels are extracted based on the high-precision digital elevation model and the digital orthophoto image via the following step S211.

[0050] S211: Based on the high-precision digital elevation model and the digital orthophoto, execute the D8 eight-direction unidirectional flow algorithm, traverse the entire grid, calculate the water flow direction from each grid cell to the eight neighboring grid cells, and generate a standardized water flow direction grid.

[0051] The core principle of the D8 eight-direction unidirectional flow algorithm is: it is assumed that the surface runoff of each grid cell flows only in the single direction with the steepest slope among its eight neighboring grid cells; the eight neighboring grid cells refer to the eight adjacent grid cells in the target grid cell in the top, bottom, left, right and four diagonal directions.

[0052] The standardized flow direction raster is the core hydrological raster data generated after executing the D8 eight-direction single-flow direction algorithm. Each raster cell stores a unique standardized flow direction code value, which corresponds one-to-one with the raster cells of the high-precision digital elevation model and perfectly matches their geographic coordinates. This raster adopts the industry-standard 8-bit binary encoding system with a value range of {1, 2, 4, 8, 16, 32, 64, 128}, corresponding to eight discrete flow directions: east, southeast, south, southwest, west, northwest, north, and northeast, respectively. This transforms the abstract surface runoff direction into numerical data that can be directly calculated by a computer.

[0053] The specific execution process of S211 can be as follows: First, based on the high-precision digital elevation model, calculate the elevation difference and corresponding horizontal distance between each grid cell and its eight neighboring grid cells (the horizontal distance in the diagonal direction is √2 times the grid side length), and then calculate the slope value in each direction; select the direction with the largest slope value as the unique water flow direction of the grid cell; use the industry-standard 8-bit binary encoding method to standardize the flow direction, and after preprocessing special terrains such as depressions and flat areas, generate a standardized water flow direction grid covering the entire study area.

[0054] Accordingly, through Figure 2 S221 to S223 shown are shorelines extracted based on the high-precision digital elevation model and the digital orthophoto.

[0055] S221: Based on the standardized water flow direction grid, the upstream confluence area of ​​each grid is calculated using the cumulative confluence algorithm to obtain the confluence accumulation grid.

[0056] The cumulative runoff algorithm is a hydrological calculation algorithm based on the D8 unidirectional flow rule. Its core principle is: assuming that surface runoff flows downstream step by step in strict accordance with the direction defined by the standardized water flow direction grid, the catchment area of ​​each grid cell is equal to its own area plus the sum of the areas of all grid cells upstream that flow into the grid.

[0057] The specific execution process of S221 can be as follows: traverse all grid cells in the entire area, starting from the upstream source grid (without upstream inflow grids), and according to the standardized water flow direction, accumulate the catchment area (planar area of ​​a single grid cell) of each grid cell to the grid cell it points downstream; repeat this accumulation process until the outlet grid cell of the watershed, and finally obtain the total upstream catchment area corresponding to each grid cell.

[0058] The runoff accumulation raster is raster data that corresponds one-to-one with the standardized water flow direction raster and is perfectly matched with geographic coordinates. Each raster cell stores the total area of ​​all upstream catchment areas at that location. The physical meaning of this raster is: the larger the raster value, the larger the upstream catchment area, the greater the amount of surface runoff collected, and the easier it is to form stable channels and rivers.

[0059] S222: Determine the binary river grid based on the river confluence accumulation threshold.

[0060] The channel runoff accumulation threshold is the critical runoff area value that distinguishes riverine areas from non-riverine areas. Its physical meaning is that when the upstream runoff accumulation of a given grid cell reaches this threshold, the amount of surface runoff collected is sufficient to form a stable, identifiable river channel. This threshold can be determined by combining the hydrogeological conditions of the study area, field river survey data, visual interpretation results of digital orthophotos, and empirical hydrological formulas.

[0061] The specific execution process of S222 can be as follows: the confluence accumulation raster generated by S221 is binarized, and the raster cells with confluence accumulation greater than or equal to the river confluence accumulation threshold are assigned a value of 1 and marked as river areas; the raster cells with confluence accumulation less than the threshold are assigned a value of 0 and marked as non-river areas.

[0062] Binary river rasters are raster data that correspond one-to-one with runoff accumulation rasters and perfectly match geographic coordinates. They contain only two discrete values, {0,1}, and can clearly and intuitively delineate the boundaries between river channels and non-river channels within the study area.

[0063] S223: Determine the centerline of the main channel of the river based on the binary river grid.

[0064] Specifically, a binary river raster can be used as a basis to extract the skeleton of the river channel raster region through morphological thinning algorithms. While preserving the overall direction and connectivity of the river channel, the planar river channel region is gradually shrunk into a linear structure with a single pixel width. This linear structure is the centerline of the main channel. During the extraction process, short tributaries, isolated fragments, and pseudo-channel noise are removed, while the main flow path running through the study area is retained. This ensures that the centerline can accurately represent the longitudinal extension morphology of the river channel and the main direction of water flow, providing a reference axis for subsequent riverbank extraction, river range definition, and geometric parameter calculation.

[0065] In some embodiments, such as Figure 3 As shown, the extent of the target river channel is further identified based on the extracted channels and shorelines through steps S231 to S235.

[0066] S231: Perpendicular to the centerline of the ditch, perform a grid traversal to the left and right banks of the river at a fixed step size, and calculate the local terrain slope and second-order difference of elevation grid by grid.

[0067] Local terrain slope refers to the degree of inclination of the ground surface calculated by taking the current traversed grid as the center and combining the elevation values ​​of its neighboring grids. It reflects the steepness and gentleness of the terrain at that location and is used to determine the location of abrupt changes in terrain along the shoreline.

[0068] The second-order difference of elevation is a numerical value obtained by performing a second-order differential operation on elevation data. It represents the rate of change of elevation and can sensitively identify abrupt changes in terrain curvature, thereby accurately locating the boundary of riverbank.

[0069] S231 can use the centerline of the main channel of the river as a reference, and traverse the grid to the left and right banks of the river in a direction perpendicular to the centerline with a preset fixed step size. For each traversed grid, the local terrain slope and second-order difference of elevation are calculated using a high-precision digital elevation model. Through the combination and change of the two types of terrain feature parameters, a quantitative judgment basis is provided for the subsequent automatic identification and extraction of the river shoreline.

[0070] S232: Mark grids that meet the following conditions as riverbank control points: grid slope exceeds the slope change threshold, second-order elevation difference changes from negative to positive or vice versa, grid topographic parameters exceed the topographic parameters of gentle riverbed sections by N times the standard deviation, where N is a positive integer greater than 1.

[0071] The slope change threshold is a pre-set critical slope value used to distinguish between gentle areas inside the riverbed and steep areas on both sides of the riverbank.

[0072] A change in the second difference of elevation from negative to positive or vice versa indicates a significant shift in topographic curvature, corresponding to abrupt changes in topography where the riverbed transitions to the bank slope.

[0073] A value exceeding the topographic parameters of a gently sloping section of the riverbed by N times indicates that the topographic features of the grid deviate significantly from the stable range within the riverbed and belong to the boundary between the riverbed and the land.

[0074] Grids that simultaneously or separately meet the above conditions are identified as riverbank control points. These control points can accurately reflect the boundary between the river water and the land, providing reliable control nodes for subsequent fitting to form a continuous and complete riverbank.

[0075] S233: Connect the control points on both banks of the river in sequence according to the direction of the river, and smooth the line elements to form a closed river boundary line.

[0076] Based on the riverbank control points, the control points are first classified and sorted out, distinguishing between the control points on the left bank and the right bank of the river. Then, according to the extension direction of the center line of the main channel, the control points on the left bank are connected in sequence from upstream to downstream (or from downstream to upstream) to form the initial shoreline of the left bank. Similarly, the control points on the right bank are connected in sequence according to the corresponding direction to form the initial shoreline of the right bank.

[0077] Since the shoreline control points are points corresponding to discrete grids, the initial shoreline formed after concatenation may have problems such as broken lines, too many inflection points, and discontinuous lines. Therefore, it is necessary to perform smoothing processing on the initial shoreline line elements (using industry-standard smoothing algorithms, such as cubic spline smoothing, moving average smoothing, etc.) to eliminate line jitter and irregularity caused by discrete points, so that the shoreline lines conform to the actual terrain undulation characteristics.

[0078] Finally, the upstream and downstream endpoints of the smoothed left bank shoreline and right bank shoreline are connected to form a closed river channel boundary line. This boundary line accurately defines the boundary outline of the target river channel, including the river water, riverbed, and tidal flats, providing a clear and standardized boundary basis for the subsequent determination of the target river channel range and related engineering analysis.

[0079] S234: Based on the river channel boundary line, the high-precision digital elevation model is trimmed to obtain the target river channel high-precision digital elevation model.

[0080] The high-precision digital elevation model of the entire area can be spatially clipped and raster extracted using the closed river channel boundary as the boundary, retaining only the elevation raster data corresponding to the area inside the river channel boundary, and removing the elevation data of the external non-river channel area.

[0081] The high-precision digital elevation model of the target river channel obtained after cropping has the same coordinate reference, grid size and elevation accuracy as the global DEM. It only contains the topographic elevation information inside the river channel, which can effectively reduce data interference from non-river areas and provide simplified and accurate dedicated topographic data for subsequent river channel topographic analysis, hydraulic calculation and cross-section extraction.

[0082] S235: Based on the river channel boundary line, crop the digital orthophoto to obtain the target river channel digital orthophoto.

[0083] Using the closed river channel boundary as the spatial boundary, the entire digital orthophoto image is cropped, retaining only the image pixel data within the river channel boundary, while pixels in non-river areas are masked or removed. The resulting target river channel digital orthophoto image has the same coordinate system, resolution, and spectral information as the original DOM, containing only the texture and color features of land features within the river channel, eliminating interference from irrelevant areas, and facilitating subsequent riverbank verification, terrain feature comparison, and fine-tuning of the river channel boundary.

[0084] S30: Extract river channel feature indicators from the high-precision digital elevation model.

[0085] "River characteristic indicators" are quantitative parameters used to quantitatively characterize the geometric shape, topographic relief, and hydraulic properties of a river. They can objectively reflect the spatial structure, evolution law, and hydrodynamic conditions of a river, providing a quantitative basis for river identification, regulation planning, and hydrological simulation.

[0086] In some embodiments, river channel index characteristics may include river width W.

[0087] In some embodiments, river channel index characteristics may include cross-sectional curvature C.

[0088] Both river width W and cross-sectional curvature C can be calculated based on cross-sections, which are sections perpendicular to the longitudinal profile of the river channel. For example, river width can be defined as the vertical distance between the left and right banks of the river in the cross-sectional direction, i.e., the straight-line distance between the control points on the banks at the same cross-section. Cross-sectional curvature can be defined as the degree of curvature of the riverbed topography curve on a cross-section perpendicular to the centerline of the main channel, using the centerline of the main channel as a reference. It is obtained by performing second-order differential calculations or curve fitting on the high-order sequence of the cross-section and is used to characterize the steepness and gentleness of the riverbed cross-section and the degree of abrupt changes in topography.

[0089] In some embodiments, such as Figure 4 As shown, S30 may include extracting river feature indicators from the high-precision digital elevation model through the following S31 and S32.

[0090] S31: The river width value of each grid cell associated with each cross section of the river channel is determined by the following method: searching the shoreline of terrain change along the direction of the target cross section, determining the boundary points of the shoreline on both sides, calculating the distance between the boundary points of the shoreline on both sides as the river width value of the target cross section, and using the river width value as the river width value of each grid cell associated with the target cross section.

[0091] S32: Determine the curvature values ​​of each grid cell associated with each cross section of the river channel by fitting a high-order sequence of the target cross section; and determine the curvature values ​​of the target cross section based on the fitting coefficients.

[0092] In some embodiments, for computational simplicity, a quadratic polynomial can be used to fit the height sequence of the target cross-section to determine the curvature value of the target cross-section. Specifically, the quadratic polynomial is Z(y) = ay 2 +by+c, Z represents the elevation of the target cross section, y is the elevation, and a, b, and c are fitting coefficients. Then the curvature value of the target cross section can be C=|2a|.

[0093] It should be noted that the quadratic polynomial is only one specific fitting method provided in this manual. Other fitting methods can also be used if the computational difficulty is not considered. These methods will not be listed here.

[0094] S30 can densely select cross-sections along the river channel, calculate the river width and cross-sectional curvature on each cross-section, and form a sequence of river channel characteristic indicators distributed longitudinally along the river channel, thereby achieving a refined description of the overall morphology of the river channel and local topographic changes.

[0095] All selected cross-sections can cover all grids. If a grid corresponds to different cross-sections, resulting in a grid corresponding to different channel feature indicators, then the average (or extreme value) of each channel feature indicator corresponding to that grid can be used as the final channel feature indicator corresponding to that grid.

[0096] It should be noted that river channel characteristic indicators may include at least one of river width and cross-sectional curvature, but are not limited to these two indicators; other characteristic indicators may also be included. Of course, in some embodiments, at least one of river width and cross-sectional curvature may be omitted, and other river channel characteristic indicators may be used instead. For example, other river channel characteristic indicators may include river longitudinal slope, cross-sectional area, extreme values ​​of runoff accumulation, and shoreline slope.

[0097] S40: Extract sediment particle features from the digital orthophoto.

[0098] Sediment particle characteristics can include particle spatial coordinates. Particle spatial coordinates are the planar position coordinates (X, Y) of a single sediment particle within the target channel in the digital orthophoto coordinate system. These coordinates are consistent with the topographic data coordinate system mentioned earlier and can be directly associated with the surface location corresponding to the high-precision digital elevation model.

[0099] Particle spatial coordinates can be obtained as follows: by preprocessing digital orthophotos such as image enhancement, noise removal, and threshold segmentation, the contour boundaries of individual sediment particles in the image are identified, and the geometric center of each particle contour is extracted. The plane coordinates corresponding to the geometric center are the spatial coordinates of the sediment particle.

[0100] The characteristics of sediment particles can also include the equivalent particle size D. This equivalent particle size D is used to uniformly quantify the size of sediment particles. Irregularly shaped sediment particles can be equivalent to a spherical particle, and the diameter of this spherical particle is the equivalent particle size D of the sediment particle.

[0101] Sediment particle characteristics can also include distribution density M. Distribution density M refers to the number of sediment particles per unit area, used to characterize the density of sediment particle distribution in the target river channel. The larger the value, the denser the sediment particles in the area, and vice versa.

[0102] In some embodiments, S40 may include extracting sediment particle features from the digital orthophoto image by: using an automatic sediment particle identification model to identify exposed sediment particles on the riverbed in the digital orthophoto image; and extracting sediment particle features based on the sediment particle identification results, wherein the sediment particle features include particle spatial coordinates, equivalent particle size, and distribution density.

[0103] Specifically, the digital orthophoto image of the target river channel is first preprocessed. Image enhancement algorithms are used to improve the contrast between sediment particles and background features. Threshold segmentation combined with morphological operations is used to remove impurities and noise, accurately identifying the contours of individual sediment particles. Then, the geometric center of each particle contour is extracted to obtain the particle's spatial coordinates (X, Y). The projected area of ​​each particle contour is calculated, and the equivalent particle size D is derived. A sampling unit with a fixed area is selected, and the number of sediment particles in the unit is counted to calculate the distribution density M. Finally, the above three feature parameters are integrated to form a target river channel sediment particle feature dataset, providing basic data support for subsequent related analyses.

[0104] It should be noted that the characteristics of sediment particles may include at least one of particle spatial coordinates, equivalent particle size D, and distribution density M, but are not limited to these three types of characteristics. Other sediment-related characteristic indicators may also be included. Of course, in some embodiments, at least one of particle spatial coordinates, equivalent particle size D, and distribution density M may be omitted, and other sediment particle characteristic indicators may be used instead. For example, other sediment particle characteristic indicators may include sediment particle size sorting coefficient, coarse particle enrichment ratio, particle alignment consistency, longitudinal distribution period of sediment, and regularity of particle deposition texture, etc.

[0105] S50: Based on the river channel feature indicators, the sediment particle characteristics, the high-precision digital elevation model, and the digital orthophoto, construct a multidimensional raster dataset.

[0106] Each raster in the multidimensional raster dataset corresponds to a raster in the high-precision digital elevation model (DEM) and the digital orthophoto. That is, each raster cell in the multidimensional raster dataset is perfectly matched one-to-one with the raster cells of the original high-precision digital elevation model (DEM) and digital orthophoto (DOM) in terms of projection coordinate system, spatial resolution, row and column numbering, and geographic plane coordinates, without spatial offset, row and column misalignment, or resolution difference. This ensures that topographic elevation, image spectrum, river morphology, and sediment particle data at the same spatial location can be accurately superimposed and assigned values ​​without bias.

[0107] The above-described S50 only describes the construction of a multidimensional raster dataset based on river channel characteristic indicators, sediment particle characteristics, high-precision digital elevation models, and digital orthophotos, and does not mean that the multidimensional raster dataset is constructed solely based on these contents. In some embodiments, other data (e.g., standardized water flow direction data, runoff accumulation data, local topographic slope data, second-order elevation difference data, sediment particle size sorting coefficient, coarse particle enrichment ratio, and other topographic and sediment-derived characteristic data) can be further combined to construct a multidimensional raster dataset.

[0108] In some embodiments, such as Figure 5 As shown, S50 may include the following S51 and S52.

[0109] S51: The river channel feature indicators, the sediment particle features, the river channel feature indicators, sediment particle features, geographical location coordinates, and pixel values ​​associated with the same grid in the high-precision digital elevation model and the digital orthophoto image are used to construct the multidimensional grid data corresponding to the grid according to a preset format.

[0110] A pixel value refers to the spectral brightness value recorded in each raster unit (pixel) of a digital orthophoto (DOM).

[0111] S51 can use a single raster under a unified spatial reference as the basic processing unit to integrate multi-source heterogeneous data raster-by-raster. Within the same raster cell corresponding to the same geographic coordinates and row and column numbers, the elevation values, digital orthophoto RGB pixel values, geographic location coordinates, river channel characteristic indicators, and sediment particle characteristics of that raster are collected and combined according to a preset band order or data organization form to form a multidimensional feature vector corresponding to a single raster, that is, the multidimensional raster data corresponding to that raster.

[0112] S51 achieves a high degree of fusion of topographic, imagery, river morphology, sediment distribution, and spatial location information at the raster cell level, providing standardized and structured single-raster feature data for subsequent model input.

[0113] S52: The multidimensional raster data associated with each raster corresponding to the target river channel constitutes a multidimensional raster dataset.

[0114] After constructing the multidimensional raster data for all rasters within the target river channel, the multidimensional raster data corresponding to all rasters are integrated according to a unified spatial arrangement rule to form a multidimensional raster dataset covering the entire target river channel. This dataset spatially matches the original high-precision digital elevation model and digital orthophoto, maintaining consistent row and column counts, resolution, and geographical extent. In terms of data dimensions, it includes topographic elevation, spectral information, spatial coordinates, river channel morphology parameters, and sediment particle parameters, forming standardized multidimensional tensor data that can be directly input into deep learning models. This provides a complete and unified data foundation for the intelligent identification and classification of stepped-deep pool structures.

[0115] For example, the original high-precision digital elevation model (DEM) has dimensions of (X, Y, 1); the original digital orthophoto has dimensions of (X, Y, 3). The two can be fused to obtain a four-band basic raster matrix (X, Y, 4). Then, based on the one-to-one correspondence between the rasteres, four types of feature parameters, namely river width W, cross-sectional curvature C, equivalent particle size D, and distribution density M, are superimposed on each raster to finally generate a multi-dimensional raster dataset that integrates multiple elements. The overall data dimension is upgraded to (X, Y, 8), providing standardized input data for subsequent deep learning semantic segmentation models.

[0116] It should be noted that the current technology has not considered combining four types of data—river channel characteristic indicators, sediment particle characteristics, high-precision digital elevation models, and digital orthophotos—to construct a multidimensional raster dataset. This is not a simple technological oversight, but rather a result of multiple constraints, including limitations in cognition, data acquisition capabilities, algorithm adaptability, and industry technological inertia. These constraints are specifically reflected in the following four aspects.

[0117] I. Core cognitive limitation: There is a misunderstanding of the essential characteristics of the stepped-deep pool.

[0118] Existing technologies have long regarded steps and deep pools as pure topographic and geomorphic units, believing that identification can be completed solely through elevation fluctuations. They have failed to recognize that steps and deep pools are the product of long-term coupled evolution of topography, water flow, and sediment. Their core distinguishing features include not only topographic morphology but also sediment sorting patterns and transverse morphological characteristics of the river channel.

[0119] Specifically: 1. Existing research and engineering applications generally focus on the macroscopic topographic differences of "step-like protrusions and deep pool depressions", ignoring the most essential and stable sediment sorting characteristic of "enrichment of coarse particles in steps and concentration of fine particles in deep pools"; 2. They have not realized that morphological indicators such as river channel cross-sectional curvature and river width changes can effectively distinguish between steps-deep pools and ordinary riverbed undulations, and can greatly improve the accuracy of structural boundary positioning.

[0120] This limitation of single-dimensional cognition fundamentally prevents existing technologies from actively incorporating sediment characteristics and riverbed morphology into the identification system.

[0121] II. Hard constraints on data acquisition and processing capabilities.

[0122] Prior to this invention, the synchronous acquisition and rasterization of four types of data faced insurmountable technical bottlenecks, specifically in the following three aspects.

[0123] 1. High-precision sediment particle features cannot be automatically extracted: Before the widespread use of centimeter-level orthophotos from UAVs, existing technologies can only obtain local sediment particle size data through manual sampling, which cannot achieve sediment particle identification and quantification for the entire river section and grid by grid. Even with high-definition images, there is a lack of mature automated sediment particle extraction algorithms, and manual annotation is extremely inefficient and has no engineering application value.

[0124] 2. Lack of automated extraction technology for river feature indicators: Existing technology can only manually measure the width and curvature of a small number of cross sections, and cannot realize the automatic calculation of river features for the entire river section on a grid-by-grid basis, and cannot form a rasterized feature layer that can be matched with terrain and image data.

[0125] 3. Insufficient spatial matching accuracy of multi-source data: Early UAV mapping technology had large coordinate registration errors, and the spatial offset between terrain data and image data could reach several meters. It was impossible to achieve precise grid-by-grid overlay, and forced fusion would lead to feature confusion and decreased recognition accuracy.

[0126] III. Adaptability limitations of algorithm technology.

[0127] Before the maturity of deep learning semantic segmentation technology, traditional ladder-deep pool recognition algorithms (such as threshold segmentation, edge detection, and traditional machine learning) could only handle low-dimensional, single-source inputs and could not effectively utilize the spatial correlation features of multi-dimensional raster data. This is specifically reflected in the following two aspects.

[0128] 1. Traditional algorithms have limited ability to process high-dimensional data. Multi-feature fusion leads to an exponential increase in computational cost and is prone to overfitting. 2. Existing algorithms are all rule-based deterministic algorithms, which cannot learn the complex nonlinear relationships between terrain, imagery, river morphology, and sediment characteristics. Even if the four types of data are forcibly combined, the recognition accuracy cannot be improved, but more interference will be introduced instead.

[0129] Fourth, the lag between industry technological inertia and application needs. This is specifically reflected in the following two aspects.

[0130] 1. Technological inertia: Mountain river surveying and geomorphological interpretation have long followed the traditional process of "topographic mapping + manual visual interpretation". The industry has formed solidified technical standards and operating procedures, and lacks the motivation to explore new multi-data fusion technology routes.

[0131] 2. Lagging application needs: Early mountain river management mainly focused on flood control and hydropower development, only requiring an understanding of the overall river topography and scouring and sedimentation trends, without the need for refined step-deep pool classification and stability assessment; it was not until the recent surge in demand for ecological restoration and refined river management that higher requirements were placed on the identification accuracy and classification detail of step-deep pools.

[0132] In summary, existing technologies are constrained by multiple limitations such as cognition, data, algorithms, and industry inertia. They neither recognize the necessity of combining the four types of data nor possess the technical conditions to achieve such a combination. Therefore, they have failed to propose the multidimensional raster dataset construction method described in this invention.

[0133] S60: Input the multidimensional raster dataset into the deep learning model to obtain the classification label of the ladder-deep pool output by the deep learning model.

[0134] The deep learning model is pre-trained. Training can be performed using the following steps: 1. When constructing the standard training label set, annotation tools such as Labelme and QuPath can be used to manually and accurately annotate the orthophotos of mountain rivers from UAVs and the corresponding original high-precision digital elevation models, constructing a supervised learning label set for stepped-deep pool structures; assign 0 values ​​to the grids in non-stepped-deep pool areas, and assign values ​​1-6 to the stepped-deep pool areas according to their type, corresponding to six sub-structures: stacked, imbricate, interlocked, congested, megalithic, and bedrock, ultimately generating standard single-channel supervised labels with dimensions (X, Y, 1); 2. When determining the training input and supervised output, the above... The multidimensional raster dataset is used as the input for model training, and the manually labeled standard classification labels are used as the model supervision output. The dataset is divided into training, validation, and test sets according to a certain ratio to ensure the balanced distribution of training samples. 3. During iterative optimization training of the model, the training set is input into the semantic segmentation model, with classification accuracy as the optimization objective and cross-entropy as the loss function. The model weights are updated iteratively through backpropagation. The validation set is used to monitor the model training effect to prevent overfitting. After the model's classification accuracy and recall reach a stable threshold and converge, the training is completed and the model is saved as a pre-trained deep learning model for subsequent fully automatic reasoning and classification of ladder-deep pool structures.

[0135] During the training and inference process of the aforementioned deep learning model, the features of sediment particles (particle spatial coordinates, equivalent particle size D, distribution density M) mentioned above are fully integrated. The core features such as the size, spatial location, and distribution density of sediment particles are embedded into multi-dimensional raster data as input data, enabling the model to accurately capture the unique sediment sorting, deposition, and arrangement patterns of the stepped-deep pool structure, further improving the recognition accuracy and classification stability of stepped-deep pool structures in complex mountain river landforms.

[0136] Deep learning models such as U-Net, YOLO-Seg, DeepLabV3+, and PSPNet are suitable for pixel-by-pixel classification of raster data. U-Net or a lightweight YOLO semantic segmentation model can be prioritized, as they are well-suited for learning the spatial features of multi-dimensional raster data from river channels, balancing classification accuracy and inference efficiency to meet the need for rapid identification of stepped-deep pool structures in large-scale, long river sections in mountainous areas.

[0137] It is important to note that S60's contribution to existing technologies does not lie in applying deep learning models to the ladder-deep pool recognition scenario, but in the innovative construction of input data and output classification labels for deep learning models. This is a core innovation that no existing technology has achieved. Specifically: 1. Regarding input data, existing technologies only use a single high-precision digital elevation model or a single orthophoto as model input, resulting in a lack of feature dimensions and failure to couple key information about river morphology and sediment particles, leading to a high misjudgment rate and unstable classification results in the identification of stepped-deep pools. This invention constructs a multi-dimensional raster dataset that integrates elevation, RGB spectrum, river width, cross-sectional curvature, equivalent particle size D, and distribution density M, uniformly inputting all features of topography, geomorphology, and sediment into the model. This allows the model to learn the complete morphology of stepped-deep pool structures and the distribution patterns of sediment, improving the accuracy of identification and classification from the data source. 2. Regarding output data, existing technologies can only output binary results for river channels / non-river channels, failing to achieve detailed classification of stepped-deep pools. This invention's model outputs precise classification labels grid-by-grid, directly corresponding to 6 typical stepped-deep pool structures. The output results are standardized and can be directly used for river stability assessment, solving the problems of existing technologies being unable to automatically subdivide and classify data and the difficulty in implementing the results in engineering.

[0138] The aforementioned intelligent identification and classification of the morphology of mountain river steps and deep pools effectively solves the core defects of existing technologies, such as low automation, poor identification accuracy, and inability to support large-scale applications. It achieves a technological breakthrough from semi-automatic manual interpretation of step-deep pool structures to intelligent identification and classification of the entire process. The specific technical effects are reflected in the following four aspects.

[0139] 1. Enables fully automated and rapid identification of long river sections and large-scale stepped-deep pool structures, significantly improving surveying efficiency and applicability.

[0140] This invention constructs a fully automated processing flow from data input to classification result output, completely eliminating the existing technology's reliance on manual screening of terrain features and marking of structural boundaries. It can complete the identification of the stepped-deep pool structure of the entire target river channel without the need for full-process technical personnel intervention, significantly reducing labor and time costs. It solves the problem that existing technologies cannot support rapid interpretation of long river sections and large areas, and can adapt to the needs of basin-level and large-scale geomorphological surveys in mountainous rivers.

[0141] 2. Significantly improves the accuracy of step-deep pool structure recognition and boundary positioning precision, ensuring the stability and reliability of classification results.

[0142] This invention innovatively integrates four types of data—high-precision digital elevation models, digital orthophotos, river channel feature indicators, and sediment particle characteristics—to construct a multi-dimensional raster dataset. This dataset comprehensively characterizes the core features of the stepped-deep pool structure from four dimensions: topographic morphology, surface texture, river channel morphology, and sediment distribution, overcoming the shortcomings of insufficient information from a single data source. By employing a deep learning semantic segmentation model to learn the nonlinear relationships between multi-dimensional features, it effectively improves the accuracy of boundary delineation and morphological determination of the stepped-deep pool structure, significantly reducing the probability of misclassification and missed classification. This solves the problems of low recognition accuracy and large fluctuations in classification results in existing technologies, providing reliable basic data support for subsequent riverbed stability assessment, river regulation, and ecological restoration.

[0143] 3. Output standardized raster-level classification results to lay the foundation for subsequent engineering applications and quantitative analysis.

[0144] The stepped-deep pool classification identifiers output by this invention are raster-level data that spatially match the original terrain and image data. This can directly achieve a one-to-one correspondence between structural types and spatial locations. It can not only clearly present the spatial distribution pattern of stepped-deep pools, but also seamlessly connect with subsequent stages such as quantitative calculation of river stability, risk zoning, and engineering scheme design. This solves the problem that the output results of existing technologies cannot be directly used for quantitative analysis and engineering implementation.

[0145] 4. Highly compatible, requiring no changes to existing drone data acquisition processes.

[0146] This invention directly processes standard digital orthophotos and high-precision digital elevation model data collected by existing UAVs, without the need to upgrade hardware or change the original data acquisition specifications. It can be quickly integrated into the existing mountain river survey operation system and has good engineering practicality and promotion value.

[0147] In numerous engineering practices and scientific research scenarios, stability analysis of riverbed morphology is required to support river regulation, ecological restoration, and flood control safety assessments. Existing technologies have developed relatively mature stability assessment systems for ordinary rivers in plain areas. These systems typically rely on average hydraulic parameters of the river section, sediment characteristics, and overall scour and deposition trends, employing empirical formulas, hydrodynamic numerical simulations, or cross-sectional monitoring and statistics to evaluate the scour and deposition balance at the macroscopic scale of the river channel. However, their assessment logic is based on uniform water flow, uniform sediment distribution, and a gentle and continuous riverbed morphology, making them unsuitable for mountain rivers with stepped-deep pool structures.

[0148] As a unique geomorphic unit in mountain rivers, the stepped-pool formation exhibits distinctive characteristics that differentiate it from ordinary rivers in terms of topography, flow regime, and sediment distribution: First, the riverbed alternates between steps and pools along the flow direction, with abrupt changes in topography and extremely high spatial heterogeneity; second, the flow exhibits a complex pattern of frequent alternations between rapid and slow currents, with local hydraulic phenomena such as hydraulic jumps, waterfalls, and eddies, which cannot be simplified using uniform flow conditions; third, the riverbed sediment is highly sorted, with coarse-grained boulders and gravels dominating the stepped sections, while fine-grained sediment is predominantly deposited in the pool sections, resulting in significant differences in particle size and distribution. These characteristics render stability assessment methods for ordinary rivers, which rely on averaging, macroscopic analysis, and homogenization, inapplicable.

[0149] Compared to the stability mechanism of ordinary rivers, which is based on the overall balance between scour and sedimentation, the stability of stepped-pool riverbeds depends on the coupled equilibrium of local structures. The scour resistance of a single step and the energy dissipation effect of the corresponding pool are interdependent and synergistically stable. Damage to local structures can easily trigger a chain reaction of instability in the river section. Its stability assessment must take into account the refined differences in structural type, spatial distribution, sediment sorting, and topographic morphology. However, existing assessment methods for ordinary rivers cannot characterize the local structural differences of stepped-pools, cannot quantify the stability contribution of different types of structures, and cannot reflect the impact of spatial combinations on overall stability. Therefore, they cannot be directly used for the stability assessment of stepped-pool channels.

[0150] Based on material composition, depositional morphology, and formation mechanism, the stepped-deep pool can be divided into six typical structures: bedrock type, megalithic type, blockage type, interlocking type, imbricate type, and depositional type. The formation conditions, erosion resistance, and structural stability of different types vary significantly. The stability level from high to low is as follows: bedrock type > megalithic type > blockage type > interlocking type > imbricate type > depositional type.

[0151] Multiple types of stepped pools often coexist within the same river channel. The number, combination of types, and spatial arrangement of these pools collectively determine the overall stability of the river channel. Generally speaking, the higher the proportion of high-stability types, the more complete the structure, and the more continuous the arrangement, the stronger the overall stability of the river channel; conversely, the higher the proportion of low-stability types, the more severe the damage, and the more discontinuities, the weaker the overall stability of the river channel.

[0152] Current technologies for stability assessment of stepped-pool river structures mostly rely on on-site qualitative observation, empirical judgment, or single-indicator estimation, resulting in the following common shortcomings: 1. They can only provide qualitative descriptions of stability / instability, failing to achieve quantitative, comparable, and reproducible numerical assessments; 2. They do not differentiate stability differences according to stepped-pool type, treating all structures as the same level, leading to large errors and low reliability; 3. They cannot match structural stability with a global river grid, failing to reflect the impact of spatial distribution on overall stability; 4. They cannot perform weighted calculations for mixed river sections with multiple types, failing to objectively reflect the contribution of type combinations and quantitative structures. These shortcomings directly result in existing methods being unable to accurately quantify and assess the true stability state of river channels, making it difficult to support engineering design, ecological restoration, and risk warning.

[0153] It is important to note that the aforementioned deficiency, "4. Inability to perform weighted calculations on multi-type mixed river sections, failing to objectively reflect the contribution of type combinations and quantitative structures," is primarily due to the inability of existing technologies to achieve grid-level fine-grained classification of stepped-deep pools. Existing technologies largely rely on manual interpretation, local cross-sectional measurements, or macro-topographic statistics, which can only determine the existence of stepped-deep pools at the river section or profile scale. They struggle to apply the classification results to each spatial grid unit, failing to establish a one-to-one correspondence between structural types and spatial locations. Consequently, they lack the spatialized, gridded data foundation required for weighted calculations.

[0154] Specifically, the core reasons why existing technologies cannot achieve raster-level classification include: First, existing methods only use single terrain or image data for analysis, without integrating river morphology features and sediment particle distribution features, lacking refined discrimination criteria to distinguish different stepped-deep pool structure types, making it difficult to determine the structure category on a raster-by-raster basis in continuous space; Second, existing technologies have not constructed a multi-dimensional fusion dataset with unified coordinates and resolution, making it impossible to achieve raster-by-raster matching and overlay analysis of terrain, imagery, river morphology, and sediment features, and not supporting raster-by-raster feature extraction and type discrimination; Third, existing technologies do not adopt a pixel-level deep learning semantic segmentation framework, making it impossible to carry out automated, raster-by-raster standardized classification of the entire river area, and can only manually label discrete points or local sections, failing to generate raster-level classification results covering the entire area.

[0155] This specification, based on the aforementioned intelligent identification and classification method for the stepped-deep-pool riverbed morphology of mountain rivers, further evaluates the stability of the target river channel based on the classification results of each stepped-deep-pool, thus providing a stability assessment method. For example... Figure 6 As shown, the stability calculation method includes the following steps S71 and S72.

[0156] S71: Assign stability quantization values ​​to each type of step-pool in the target river channel, and process the stability quantization values ​​of each type of step-pool to correspond with the grid.

[0157] Based on the differences in material composition, structural strength, erosion resistance, and formation mechanism among the six types of stepped-deep pool structures, a unique and fixed stability quantification value can be pre-assigned to each structural type. The value directly reflects the stability of that type of stepped-deep pool itself.

[0158] In some embodiments, the stability quantification values ​​can be set as follows: bedrock type stability value is 6, corresponding to an extremely high stability level; boulder type stability value is 5, corresponding to a very high stability level; congested type stability value is 4, corresponding to a high stability level; interlocked type stability value is 3, corresponding to a medium stability level; shingled type stability value is 2, corresponding to a low stability level; and stacked type stability value is 1, corresponding to a very low stability level.

[0159] After assigning stability quantification values, the step-pool classification raster output by S60 is mapped one-to-one with the stability quantification values. Each raster is assigned a stability value corresponding to its associated step-pool type, ultimately forming a stability-assigned raster covering the entire target river channel. "One-to-one mapping" means that the stability-assigned raster is completely consistent with each raster in the high-precision digital elevation model, digital orthophoto, and multidimensional raster dataset in terms of spatial coordinates, ground resolution, and raster row and column numbers, achieving precise matching between the step-pool structure type, stability value, and spatial location.

[0160] S71 transforms existing experience-based qualitative judgments into a standardized, calculable, and quantifiable numerical system by assigning stability quantification values ​​according to structural type. This effectively addresses the shortcomings of existing technologies, such as the inability to quantify the stability of stepped-deep pools, distinguish the stability differences between different structures, and achieve spatial expression of stability. It provides a unified and standardized quantitative basis for subsequent calculations of the overall stability assessment value of the target river channel.

[0161] It is important to note that existing technologies lack standardized, reusable, and engineered quantitative stability value assignments for different types of stepped-deep pools. Current assessments of stepped-deep pool stability remain at the level of manual qualitative judgment, only able to provide vague descriptions of individual structures as "stable / relatively stable / unstable" based on field surveys. Even though a few academic studies have provided scattered empirical values ​​for specific structures in small-scale experiments, a unified assignment standard has not been established, a quantitative mapping relationship between type and stability has not been created, and it has not become a universally applicable technical means.

[0162] The current technology has failed to achieve this operation due to three fundamental constraints: cognition, data, and technology. First, at the cognitive level, there is no unified quantitative basis. Existing research only qualitatively recognizes the differences in stability among different steps-deep pools, but has not systematically established a quantitative evaluation system for stability based on material composition, structural strength, erosion resistance, and formation mechanism. Different scholars have vastly different judgments on the stability of the same type of structure, making it impossible to form a reusable assignment standard. Second, there is a lack of data foundation for large-scale automated classification. Current technology cannot achieve automatic identification of step-deep pool types across the entire river section at the grid level. It can only manually label a small number of discrete structures and does not have the prerequisite for batch assignment of values ​​to all structures within a large river channel. Third, the evaluation system has not formed a quantitative logical closed loop. Current technology lacks a supporting weighted calculation method for overall stability. Even if a value is assigned to a single structure, it cannot be converted into a quantitative result of the overall stability of the river channel, lacking the technological driving force and application value to carry out this operation.

[0163] S72: Calculate the overall stability assessment value of the target river based on the stability quantification values ​​corresponding to each grid cell of each step-pool in the target river channel.

[0164] In some embodiments, S72 can be based on the stability assignment grid of the global grid obtained in S71, and comprehensively consider key factors such as the type, spatial distribution, structural length and quantity ratio of the stepped-deep pool, and use a weighted average calculation method to integrate the local structural stability values ​​into a unified overall stability assessment value for the entire river channel, thereby realizing a quantitative evaluation from single grid stability to river system stability.

[0165] In the calculation, the stability quantification value of each stepped-deep pool type structure is used as the weight, and the number of grid cells corresponding to that stepped-deep pool type structure is used as the weighting factor to perform a weighted summation of the stepped-deep pool type structures in the target channel. The summation result is then normalized using the maximum stability quantification value of the stepped-deep pool and the total number of grid cells in the target channel, ultimately obtaining a comparable, repeatable, and standardized overall stability assessment value. This calculation method can be expressed as: , Where Z is the overall stability assessment value, N is the number of stepped-deep pool structure types, and x i Let n be the stability quantification value of the i-th step-deep pool type structure. i Let P be the number of grid cells in the i-th step-deep pool type structure, P be the maximum stability quantization value, and Q be the total number of grid cells in the target channel.

[0166] This overall stability assessment value can objectively reflect the contribution of different types, quantities, and lengths of stepped-deep pool structures to the overall stability of the river channel. It can be directly used for determining the stability level of the river channel, risk zoning, and evaluating the priority of engineering repairs.

[0167] The aforementioned “step-deep pool structure type” refers to the classification based on the material composition, formation mechanism, morphological characteristics and erosion resistance of step-deep pools. It is used to distinguish different types of step-deep pool structures (entities), including but not limited to bedrock type, megalithic type, blockage type, interlocking type, imbricate type, and depositional type.

[0168] The aforementioned “step-deep pool type structure” refers to a specific river geomorphic structure (entity) belonging to a certain step-deep pool structure type. It is a step-deep pool entity that actually exists in the river channel and has a spatial location and morphological scale.

[0169] This specification provides an intelligent recognition and classification device for the morphology of mountain river steps and deep pools, which can be used to implement the above-mentioned method. For example... Figure 7 As shown, the device includes an acquisition unit 10, a first extraction unit 20, a second extraction unit 30, a third extraction unit 40, a construction unit 50, and a processing unit 60.

[0170] The acquisition unit 10 is used to acquire orthophoto topographic data containing the features of mountainous river steps and deep pools. The orthophoto topographic data includes digital orthophotos and high-precision digital elevation models associated with unified geographic coordinates.

[0171] The first extraction unit 20 is used to extract channels and shorelines based on the high-precision digital elevation model and the digital orthophoto, and to further identify the range of the target river channel based on the extracted channels and shorelines.

[0172] The second extraction unit 30 is used to extract river feature indicators from the high-precision digital elevation model.

[0173] The third extraction unit 40 is used to extract sediment particle features from the digital orthophoto.

[0174] The construction unit 50 is used to construct a multidimensional raster dataset based on the river channel feature indicators, the sediment particle characteristics, the high-precision digital elevation model, and the digital orthophoto.

[0175] The processing unit 60 is used to input the multidimensional raster dataset into the deep learning model to obtain the ladder-deep pool classification label output by the deep learning model.

[0176] In some embodiments, the building unit includes a first building subunit and a second building subunit.

[0177] The first constituent subunit is used to construct the multidimensional raster data corresponding to the raster by combining the river feature indicators, the sediment particle features, the river feature indicators, sediment particle features, geographical coordinates, and pixel values ​​associated with the same raster in the high-precision digital elevation model and the digital orthophoto image according to a preset format.

[0178] The second constituent sub-unit is used to construct a multidimensional raster dataset by associating the multidimensional raster data of each raster corresponding to the target river channel.

[0179] In some embodiments, the second extraction unit includes a first determining subunit and a second determining subunit.

[0180] The first determining subunit is used to determine the river width value of each grid associated with each cross section of the river channel by: searching the abrupt change shoreline along the direction of the target cross section, determining the boundary points of the two shorelines, calculating the distance between the boundary points of the two shorelines as the river width value of the target cross section, and using the river width value as the river width value of each grid associated with the target cross section.

[0181] The second determining subunit is used to determine the curvature values ​​of each grid cell associated with each cross section of the river channel by fitting a high-order sequence of the target cross section and determining the curvature values ​​of the target cross section based on the fitting coefficients.

[0182] The cross-section of the river channel is a section perpendicular to the longitudinal section of the river channel.

[0183] In some embodiments, the third extraction unit includes an identification subunit and an extraction subunit.

[0184] The identification subunit is used to identify exposed sediment particles on the riverbed in the digital orthophoto using an automatic sediment particle identification model.

[0185] The extraction subunit is used to extract sediment particle features based on the sediment particle identification results. The sediment particle features include particle spatial coordinates, equivalent particle size, and distribution density.

[0186] In some embodiments, the apparatus further includes an evaluation unit for evaluating the stability of the target river channel based on the classification results of each step-pool in the target river channel.

[0187] In some embodiments, the evaluation unit includes an allocation subunit and a first calculation subunit.

[0188] The allocation sub-unit is used to assign stability quantization values ​​to each type of step-pool in the target channel and to map the stability quantization values ​​of each type of step-pool to the grid.

[0189] The first calculation subunit is used to calculate the overall stability assessment value of the target river based on the stability quantification values ​​corresponding to each grid of each step-pool in the target river channel.

[0190] In some embodiments, the first calculation subunit includes a summation subunit and a first processing subunit.

[0191] The summation sub-unit is used to perform weighted summation on the stepped-deep pool type structures of the target channel, with the stability quantification value of each stepped-deep pool type structure as the weight and the number of grids corresponding to that stepped-deep pool type structure as the weighting factor.

[0192] The first processing subunit is used to normalize the summation result by taking the maximum stability quantification value of the stepped-deep pool and the total number of grids in the target channel, which is the overall stability assessment value of the target channel.

[0193] In some embodiments, the first extraction unit includes a second calculation subunit, a third calculation subunit, a third determination subunit, and a fourth determination subunit.

[0194] The second calculation subunit is used to execute the D8 eight-direction unidirectional flow algorithm based on the high-precision digital elevation model and the digital orthophoto, traverse the entire grid, calculate the water flow direction from each grid cell to the eight neighboring grid cells, and generate a standardized water flow direction grid.

[0195] The third calculation subunit is used to calculate the upstream confluence area of ​​each grid based on the standardized water flow direction grid and through the cumulative confluence algorithm to obtain the confluence accumulation grid.

[0196] The third determination sub-unit is used to determine the binary river grid based on the river confluence accumulation threshold.

[0197] The fourth determining sub-unit is used to determine the centerline of the main channel of the river based on the binary river grid.

[0198] In some embodiments, the first extraction unit includes a fourth calculation subunit, a marking subunit, a second processing subunit, a first cropping subunit, and a second cropping subunit.

[0199] The fourth calculation sub-unit is used to perform a grid traversal on both sides of the river channel at a fixed step size, perpendicular to the center line of the channel, and calculate the local terrain slope and second-order difference of elevation grid by grid.

[0200] The marker sub-unit is used to mark grids that meet the following conditions as riverbank control points: the grid slope exceeds the slope change threshold, the second difference of elevation changes from negative to positive or vice versa, and the grid topographic parameters exceed the topographic parameters of the gentle riverbed section by N times the standard deviation, where N is a positive integer greater than 1.

[0201] The second processing subunit is used to connect the control points on both banks of the river in sequence according to the direction of the river, and to smooth the line elements to form a closed river boundary line.

[0202] The first trimming subunit is used to trim the high-precision digital elevation model based on the river channel boundary line to obtain the target river channel high-precision digital elevation model.

[0203] The second cropping subunit is used to crop the digital orthophoto based on the river channel boundary line to obtain the target river channel digital orthophoto.

[0204] The descriptions and functions of the above devices can be found in the section on "Intelligent Recognition and Classification Methods and Systems for Mountain River Step-Deep Pool Riverbed Morphology," and will not be repeated here.

[0205] This invention also provides an electronic device, such as... Figure 8 As shown, the electronic device may include a processor 801 and a memory 802, wherein the processor 801 and the memory 802 may be connected via a bus or other means. Figure 8 Taking the example of a connection between China and Israel via a bus.

[0206] Processor 801 can be a central processing unit (CPU). Processor 801 can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0207] Memory 802, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the intelligent identification and classification method for the stepped-deep-pool riverbed morphology of mountain rivers in this embodiment of the invention (e.g., Figure 8 The diagram shows an acquisition unit 10, a first extraction unit 20, a second extraction unit 30, a third extraction unit 40, a construction unit 50, and a processing unit 60. The processor 801 executes various functional applications and data processing by running non-transitory software programs, instructions, and modules stored in the memory 802, thereby realizing the intelligent identification and classification method for the morphology of mountain river steps-deep pools in the above method embodiment.

[0208] The memory 802 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor 801, etc. Furthermore, the memory 802 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 802 may optionally include memory remotely located relative to the processor 801, and these remote memories may be connected to the processor 801 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0209] The one or more modules are stored in the memory 802, and when executed by the processor 801, they perform the following: Figure 1 The embodiment shown illustrates an intelligent recognition and classification method for the morphology of mountainous river steps and deep pools.

[0210] The specific details of the aforementioned electronic device can be understood by referring to the relevant descriptions and effects in the method embodiments, and will not be repeated here.

[0211] This specification also provides a computer storage medium storing computer program instructions, which, when executed by a processor, implement the steps of the above-described intelligent identification and classification method for the morphology of mountainous river steps and deep pools.

[0212] This specification also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the above-described intelligent identification and classification method for the morphology of mountain river steps-deep pools.

[0213] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

[0214] The various embodiments in this specification are described in a progressive manner. For the same or similar parts between the various embodiments, please refer to each other. The focus of each embodiment is to describe the differences from other embodiments.

[0215] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions.

[0216] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0217] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute certain parts of the methods of various embodiments of this application.

[0218] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0219] This application can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0220] Although this application has been described through embodiments, those skilled in the art will know that this application has many modifications and variations without departing from the spirit of this application, and it is intended that the appended claims cover such modifications and variations without departing from the spirit of this application.

Claims

1. A method for intelligent recognition and classification of the morphology of mountain river steps and deep pools, characterized in that, include: Obtain orthophoto topographic data containing the features of mountainous river steps and deep pools. The orthophoto topographic data includes digital orthophotos and high-precision digital elevation models linked by unified geographic coordinates. Based on the high-precision digital elevation model and the digital orthophoto, the channel and shoreline are extracted, and the range of the target river channel is further identified based on the extracted channel and shoreline. Extract river channel feature indicators from the high-precision digital elevation model; Extract sediment particle features from the digital orthophoto; Based on the river channel feature indicators, the sediment particle characteristics, the high-precision digital elevation model, and the digital orthophoto, a multidimensional raster dataset is constructed. The multidimensional raster dataset is input into a deep learning model to obtain the ladder-deep pool classification labels output by the deep learning model.

2. The method according to claim 1, characterized in that, Based on the aforementioned river channel characteristic indicators, sediment particle characteristics, the high-precision digital elevation model, and the digital orthophoto, a multidimensional raster dataset is constructed, including: The river channel feature indicators, sediment particle features, and river channel feature indicators, sediment particle features, geographic location coordinates, and pixel values ​​associated with the same grid in the high-precision digital elevation model and digital orthophoto image are used to construct the multidimensional grid data corresponding to the grid according to a preset format. The multidimensional raster data associated with each raster corresponding to the target river channel constitute a multidimensional raster dataset.

3. The method according to claim 1, characterized in that, River feature indicators are extracted from the high-precision digital elevation model, including: The river width value of each grid cell associated with each cross section of the river channel is determined by the following method: searching the shoreline of terrain change along the direction of the target cross section, determining the boundary points of the shoreline on both sides, calculating the distance between the boundary points of the shoreline on both sides as the river width value of the target cross section, and using the river width value as the river width value of each grid cell associated with the target cross section. The curvature values ​​of each grid cell associated with each cross section of the river channel are determined by: fitting a high-order sequence of the target cross section; and determining the curvature values ​​of the target cross section based on the fitting coefficients. The cross-section of the river channel is a section perpendicular to the longitudinal section of the river channel.

4. The method according to claim 1, characterized in that, Extracting sediment particle features from the digital orthophoto image includes: An automatic sediment particle identification model was used to identify exposed sediment particles on the riverbed in the digital orthophoto image. Based on the sediment particle identification results, sediment particle features are extracted, including particle spatial coordinates, equivalent particle size, and distribution density.

5. The method according to claim 1, characterized in that, The method further includes: The stability of the target river channel is assessed based on the classification results of each step-pool in the target river channel.

6. The method according to claim 5, characterized in that, Stability calculation methods include: Assign stability quantization values ​​to each type of step-pool in the target river channel, and process the stability quantization values ​​of each type of step-pool to the grid. Based on the stability quantification values ​​corresponding to each step-pool and each grid in the target river channel, the overall stability assessment value of the target river channel is calculated.

7. The method according to claim 1, characterized in that, Based on the stability quantification values ​​corresponding to each grid cell of each step-pool in the target river channel, the overall stability assessment value of the target river channel is calculated, including: The stability quantification value of each stepped-deep pool type structure is used as the weight, and the number of grids corresponding to the stepped-deep pool type structure is used as the weighting factor to perform a weighted summation of the stepped-deep pool type structures of the target river channel; The summation result is normalized by using the maximum stability quantification value of the stepped-deep pool and the total number of grids in the target channel to obtain the overall stability assessment value of the target channel.

8. The method according to claim 1, characterized in that, The trench is extracted based on the high-precision digital elevation model and the digital orthophoto image in the following manner: Based on the high-precision digital elevation model and the digital orthophoto, the D8 eight-direction unidirectional flow algorithm is executed to traverse the entire grid and calculate the water flow direction from each grid cell to the eight neighboring grid cells, generating a standardized water flow direction grid. Accordingly, the shoreline is extracted based on the high-precision digital elevation model and the digital orthophoto as follows: Based on the standardized water flow direction grid, the upstream confluence area of ​​each grid is calculated using the cumulative confluence algorithm to obtain the grid of cumulative confluence. The binary river grid is determined based on the river confluence accumulation threshold; The centerline of the main channel of the river is determined based on the binary river grid.

9. The method according to claim 1, characterized in that, The extent of the target river channel can be further identified based on the extracted channels and shorelines using the following methods: Perpendicular to the centerline of the ditch, perform a grid traversal to the left and right banks of the river at a fixed step size, and calculate the local terrain slope and second-order elevation difference for each grid. Rasteres meeting the following conditions are marked as riverbank control points: the raster slope exceeds the slope abrupt change threshold, the second difference of elevation changes from negative to positive or vice versa, and the raster topographic parameters exceed the topographic parameters of the gentle riverbed section by N times the standard deviation, where N is a positive integer greater than 1. The control points on both sides of the river are connected in sequence according to the direction of the river, and the line elements are smoothed to form a closed river boundary line. The high-precision digital elevation model of the target river channel is obtained by trimming the high-precision digital elevation model based on the river channel boundary line. The digital orthophoto image of the target river is obtained by cropping the digital orthophoto image based on the river channel boundary line.

10. A smart identification and classification device for the morphology of mountain river steps and deep pools, characterized in that, include: The acquisition unit is used to acquire orthophoto topographic data containing the features of mountainous river steps and deep pools. The orthophoto topographic data includes digital orthophotos and high-precision digital elevation models associated with unified geographic coordinates. The first extraction unit is used to extract channels and shorelines based on the high-precision digital elevation model and the digital orthophoto, and to further identify the range of the target river channel based on the extracted channels and shorelines. The second extraction unit is used to extract river feature indicators from the high-precision digital elevation model; The third extraction unit is used to extract sediment particle features from the digital orthophoto; A construction unit is used to construct a multidimensional raster dataset based on the river channel feature indicators, the sediment particle characteristics, the high-precision digital elevation model, and the digital orthophoto. The processing unit is used to input the multidimensional raster dataset into a deep learning model to obtain the ladder-deep pool classification labels output by the deep learning model.

11. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes these computer instructions to implement the intelligent identification and classification method for the morphology of mountain river steps and deep pools in any one of claims 1 to 9.

Images