A river surface flow field physical rule reconstruction system

Abstract

The application discloses a river surface flow field physical rule reconstruction system, and relates to the technical field of hydrological monitoring and computer vision, which comprises a visual sensor, a visual observation branch, a physical rule deduction branch, a data fusion module and a result output module; the visual sensor is arranged in a river observation area and is used for continuously collecting video image sequences of the river surface. The river surface flow field physical rule reconstruction system constructs a closed loop reconstruction mechanism with visual observation and physical rule deduction in parallel and bidirectional verification, effectively overcomes the flow field measurement failure problem caused by the degradation of visual information in the traditional method under complex environment, can fully retain the visual measurement details when the image quality is good, automatically enhances the constraint effect of the physical model when the visual information is unreliable, ensures the integrity and continuity of the output flow field in space, and improves the comprehensive performance of the reconstruction result in terms of physical rationality and data reliability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hydrological monitoring and computer vision, in particular to a river surface flow field physical rule reconstruction system. BACKGROUND

[0002] The current river surface flow field observation technology based on video images can achieve good results under ideal lighting and calm weather conditions. However, in actual engineering applications, such as when it rains, the water surface reflects strong light, or it is misty, the technology system exposes its inherent vulnerability. Raindrops hitting the water surface will form dense interference ripples, strong light will cause local water surface areas to be overexposed and lose texture details, and fog will cause the image contrast to drop significantly. These complex environmental factors directly cause the number of visual feature points available for flow field calculation in the video sequence to decrease sharply and the tracking trajectory to break down in large areas, ultimately resulting in a large number of blank areas in the reconstructed flow field vector diagram and serious inaccuracies in the calculated flow speed. Existing improvement schemes mostly focus on optimizing image preprocessing algorithms or using more complex feature matching strategies, but these efforts are still essentially within the scope of visual perception and have not broken through the limitations of relying solely on apparent image information. The processing flow often deviates from the physical movement rules of river water bodies, and when the image information itself becomes unreliable due to environmental interference, the subsequent calculations lose a solid foundation, making it difficult to ensure the robustness of the entire system at critical moments. The technical problem to be solved by the present application is how to overcome the disturbance of complex environments on the visual observation of river surface flow fields and achieve high-robustness physical rule reconstruction. SUMMARY

[0003] The present application aims to provide a river surface flow field physical rule reconstruction system to solve the problems raised in the background.

[0004] To solve the above technical problems, the present application provides the following technical scheme: a river surface flow field physical rule reconstruction system, comprising a visual sensor, a visual observation branch, a physical rule deduction branch, a data fusion module, and a result output module;

[0005] The visual sensor is arranged in the river observation area to continuously collect video image sequences of the river surface;

[0006] The visual observation branch is connected to the visual sensor and comprises a feature extraction unit and a flow field calculation unit. The feature extraction unit identifies water surface feature points from the video image sequences and generates feature point motion trajectories. The flow field calculation unit generates an initial surface flow field based on the feature point motion trajectories and simultaneously generates a visual uncertainty matrix representing the reliability of each position in the initial surface flow field;

[0007] The physical rule deduction branch includes a hydrological data receiving unit and a physical model calculation unit, the hydrological data receiving unit acquires river terrain data, boundary condition data and real-time hydrological measurement data, and the physical model calculation unit independently calculates and generates a physical reference flow field based on a coupling model of a shallow water equation and a vorticity constraint according to the received data;

[0008] The data fusion module is connected with the visual observation branch and the physical rule deduction branch respectively, and includes a weight distribution unit and a fusion calculation unit, the weight distribution unit dynamically adjusts the fusion weight of the initial surface flow field and the physical reference flow field at the corresponding position according to the reliability value of each position in the visual uncertainty matrix, and the fusion calculation unit performs fusion calculation on the two flow fields by using an adaptive weighted algorithm to generate a first version of the reconstructed flow field;

[0009] The data fusion module further feeds back the first version of the reconstructed flow field to the physical rule deduction branch, and the physical model calculation unit performs secondary calculation with the first version of the reconstructed flow field as the initial condition to generate a secondary verification flow field;

[0010] The result output module is connected with the data fusion module and includes a time-space continuity verification unit, the time-space continuity verification unit performs continuity analysis on the secondary verification flow field in the time and space dimensions, eliminates abnormal vectors that do not conform to the fluid motion law, and outputs a final surface flow field.

[0011] Further, the generation of the visual uncertainty matrix includes: a feature extraction unit performs quality evaluation on the video image sequence, different reliability values are set for the raindrop interference area, the water surface reflection area and the low contrast area respectively, and a flow field calculation unit further corrects the reliability value of each position according to the success rate of feature point tracking and the trajectory continuity to form a complete visual uncertainty matrix.

[0012] Further, the coupling model of the shallow water equation and the vorticity constraint in the physical rule deduction branch simultaneously solves the mass conservation equation, the momentum conservation equation and the vorticity transport equation to establish the physical constraint relationship of the river surface flow field; the physical model calculation unit adopts the finite difference method to discretely solve the coupling model, wherein the terrain data are used to determine the elevation distribution of the calculation grid, the boundary condition data are used to set the boundary constraints of the river bank and the import and export, and the real-time hydrological measurement data are used to provide the initial conditions for model calculation.

[0013] Further, the adaptive weighted algorithm in the data fusion module is implemented as follows: in the area where the visual uncertainty matrix shows high reliability, a higher fusion weight is given to the initial surface flow field; in the area where the visual uncertainty matrix shows low reliability, a higher fusion weight is given to the physical reference flow field; the distribution of the fusion weight adopts a continuous function form to avoid sudden changes at the region boundary.

[0014] Furthermore, after the physical rule deduction branch receives the first version of the reconstructed flow field, the physical model calculation unit uses this flow field as the new initial condition to re-execute the solution process of the shallow water equation and vortex constraint coupled model. By comparing the difference between the second calculation result and the first version of the reconstructed flow field, flow field correction parameters are generated. The data fusion module corrects the first version of the reconstructed flow field according to the flow field correction parameters.

[0015] Furthermore, the spatiotemporal continuity verification unit identifies anomalous vectors that do not conform to the fluid motion characteristics by analyzing the variation patterns of the direction and magnitude of the flow field vectors in the time series and spatial grid. For the identified anomalous vectors, the interpolation results of adjacent normal vectors are used to replace them, ensuring the continuity and rationality of the output flow field in the spatiotemporal dimension.

[0016] Furthermore, it also includes a result display module, which is connected to the result output module, to visualize the final surface flow field in the form of a vector diagram, and to use different colors or markers in the visualization interface to distinguish the reliability level of each location in the flow field.

[0017] Furthermore, the vortex constraint term in the shallow water equation coupled with the vortex constraint model ensures that the physical reference flow field conforms to the vortex motion law of the actual river by restricting the generation and dissipation process of vortex in the flow field. The physical model calculation unit checks the satisfaction of the vortex constraint after each iteration calculation and makes local adjustments to the regions that do not meet the constraint.

[0018] Furthermore, the allocation of the fusion weights adopts a continuous weight allocation strategy based on Gaussian functions. The reliability values ​​in the visual uncertainty matrix are used as input parameters of the Gaussian function. By adjusting the mean and variance of the Gaussian function, the sensitivity of weight allocation and the width of the transition zone are controlled.

[0019] Furthermore, it also includes a data storage module, which connects the visual observation branch, the physical rule deduction branch, the data fusion module, and the result output module. It is used to store intermediate data and final results generated during system operation. The data storage module organizes the data in chronological order and establishes a data index for subsequent querying and analysis.

[0020] This invention provides a system for reconstructing the physical rules of river surface flow fields. It has the following beneficial effects:

[0021] This system for reconstructing the physical rules of river surface flow fields effectively overcomes the problem of flow field measurement failure caused by visual information degradation in complex environments, which is a problem of traditional methods. It employs a closed-loop reconstruction mechanism that combines visual observation and physical rule deduction in parallel with bidirectional verification. Utilizing a weighted fusion strategy dynamically guided by the visual uncertainty matrix, the system can fully preserve visual measurement details when image quality is good, and automatically enhance the constraints of the physical model when visual information is unreliable, thus ensuring the spatial integrity and continuity of the output flow field. Furthermore, multiple safeguards, including secondary physical verification and spatiotemporal continuity checks, improve the overall performance of the reconstruction results in terms of physical rationality and data reliability.

[0022] This system for reconstructing the physical rules of the river surface flow field deeply embeds river hydrodynamics knowledge into the visual computing process, forming an intelligent analysis system with self-verification and correction capabilities. It can provide stable data support for hydrological monitoring, flood control scheduling, and aquatic ecological research without being strictly limited by meteorological conditions, solving the problem of insufficient adaptability in practical applications and realizing a technological leap from relying on single sensor data to multi-source information fusion cognition. Attached Figure Description

[0023] Figure 1 This is a system component diagram of a river surface flow field physical rule reconstruction system according to the present invention;

[0024] Figure 2 This is a flowchart illustrating the uncertainty handling process of a river surface flow field physical rule reconstruction system according to the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Please see Figure 1 and Figure 2 The present invention provides a technical solution: a system for reconstructing the physical rules of the river surface flow field, including a visual sensor, a visual observation branch, a physical rule deduction branch, a data fusion module and a result output module;

[0027] Visual sensors are installed in the river observation area to continuously acquire video image sequences of the river surface;

[0028] The visual observation branch connects to a visual sensor, including a feature extraction unit and a flow field calculation unit. The feature extraction unit identifies water surface feature points from the video image sequence and generates the motion trajectory of the feature points. The flow field calculation unit generates an initial surface flow field based on the motion trajectory of the feature points and simultaneously generates a visual uncertainty matrix to characterize the reliability of each position in the initial surface flow field.

[0029] The physical rule derivation branch includes a hydrological data receiving unit and a physical model calculation unit. The hydrological data receiving unit acquires river topographic data, boundary condition data, and real-time hydrological measurement data. The physical model calculation unit, based on a coupled model of shallow water equations and eddy current constraints, independently calculates and generates the physical reference flow field according to the received data.

[0030] The data fusion module is connected to the visual observation branch and the physical rule deduction branch respectively. It includes a weight allocation unit and a fusion calculation unit. The weight allocation unit dynamically adjusts the fusion weights of the initial surface flow field and the physical reference flow field at corresponding positions according to the reliability values ​​of each position in the visual uncertainty matrix. The fusion calculation unit uses an adaptive weighting algorithm to perform fusion calculations on the two flow fields and generate the first version of the reconstructed flow field.

[0031] The data fusion module also feeds back the first version of the reconstructed flow field to the physical rule deduction branch. The physical model calculation unit uses the first version of the reconstructed flow field as the initial condition to perform secondary calculations and generate a secondary verification flow field.

[0032] The results output module is connected to the data fusion module, which includes a spatiotemporal continuity verification unit. The spatiotemporal continuity verification unit performs a continuity analysis of the secondary verification flow field in both time and space dimensions, removes abnormal vectors that do not conform to the laws of fluid motion, and outputs the final surface flow field.

[0033] It should be further explained that a highly robust reconstruction of the river surface flow field is achieved by constructing a closed-loop system with parallel visual observation and physical rule deduction, and bidirectional verification. The system starts with visual sensors deployed in the river observation area, which continuously acquire raw video image sequences. After receiving this sequence, the visual observation branch first identifies natural or artificial feature points on the water surface by the feature extraction unit and tracks their trajectories; the flow field calculation unit then calculates the initial surface flow field based on these trajectories. The key innovation lies in the simultaneous generation of a visual uncertainty matrix corresponding to the initial flow field space. This matrix quantitatively evaluates the reliability of each flow field vector point based on image quality and the coherence of feature point tracking. Among them, image quality includes the identification of low-contrast areas caused by raindrop ripples, specular reflections, and fog, and the coherence of feature point tracking includes the trajectory break frequency.

[0034] Meanwhile, the physical rule derivation branch operates independently. Its hydrological data receiving unit acquires real-time hydrological data, including river topography, bank boundary conditions, and upstream flow velocity or water level. The physical model calculation unit is based on a coupled physical model that integrates shallow water equations and eddy constraints. It uses the finite difference method to perform discrete solutions and outputs a physical reference flow field derived purely from physical laws. In this model, the shallow water equations describe the conservation of mass and momentum, and the eddy constraints characterize the fluid rotation characteristics.

[0035] The core innovation lies in the data fusion module: the weight allocation unit of this module dynamically assigns fusion weights to the initial surface flow field and the physical reference flow field in each calculation unit based on the visual uncertainty matrix. Specifically, it adopts a continuous strategy based on Gaussian function, giving higher weights to the physical reference flow field in areas with low visual reliability. The fusion calculation unit then performs adaptive weighted calculation to generate the first version of the reconstructed flow field.

[0036] To achieve in-depth verification, the system feeds back this first version of the reconstructed flow field to the physical rule deduction branch. The physical model calculation unit uses it as a new and better initial condition for secondary calculation. By comparing the difference between the secondary calculation result and the initial fusion result, flow field correction parameters are generated and the first version of the reconstructed flow field is finely adjusted.

[0037] Finally, the spatiotemporal continuity verification unit in the results output module analyzes the flow field after secondary verification, identifies and removes anomalous vectors that abruptly change in the time series or are isolated in the spatial distribution. This is typically achieved using interpolation methods based on surrounding valid vectors, resulting in a physically reasonable and data-reliable final river surface flow field. This scheme, through deep fusion and iterative verification of visual information and physical rules, ensures that even in complex environments where single visual data fails, the system can still output stable and continuous flow field reconstruction results based on the constraints of the physical model.

[0038] The generation of the visual uncertainty matrix includes: the feature extraction unit performs quality assessment on the video image sequence, sets different reliability values ​​for raindrop interference areas, water surface reflection areas and low contrast areas, and the flow field calculation unit further corrects the reliability values ​​at each position based on the success rate of feature point tracking and trajectory continuity, forming a complete visual uncertainty matrix.

[0039] It should be further explained that the system first performs a multi-dimensional quality assessment on the input video image sequence. For raindrop interference areas, the system quantifies the interference intensity by detecting the density and distribution of high-frequency ripples in the image. For water surface reflection areas, the system identifies the range and degree of overexposure by analyzing the distribution characteristics of image pixel brightness values. For low-contrast areas, the system evaluates texture clarity by calculating the gradient magnitude of local image regions. Based on these image quality assessment results, the system sets corresponding initial reliability values ​​for different interference types, with lower initial reliability values ​​set for raindrop-dense areas, strong reflective core areas, and low-gradient areas. Furthermore, the system analyzes the success rate of feature point tracking and trajectory continuity, dynamically correcting the initial reliability values ​​by statistically analyzing the success rate of feature point matching between consecutive frames and the breakage of motion trajectories. Finally, the system organizes these reliability values, which have undergone comprehensive multi-factor evaluation, into a two-dimensional matrix that perfectly corresponds to the spatial location of the flow field—the visual uncertainty matrix. Each element in this matrix directly reflects the reliability of the flow field data at the corresponding location, providing a precise weight allocation basis for subsequent data fusion. Flow field data includes flow field vectors (x-direction velocity u, y-direction velocity v), water depth h, vertical vorticity ω, reliability r, and fusion weights (w). vis w phy A comprehensive dataset.

[0040] The shallow water equations and eddy constraint coupled model in the physical rule derivation branch establish the physical constraint relationship of the river surface flow field by simultaneously solving the mass conservation equation, momentum conservation equation and eddy transport equation. The physical model calculation unit uses the finite difference method to discretize and solve the coupled model. Topographic data is used to determine the elevation distribution of the calculation grid, boundary condition data is used to set the boundary constraints of the riverbank and inlet / outlet, and real-time hydrological measurement data is used to provide the initial conditions for model calculation.

[0041] It should be further explained that the implementation method, mathematical expression, discretization process, and key parameter determination method of the shallow water equation and eddy constraint coupled model are as follows:

[0042] The specific mathematical expressions for the coupling model include the following:

[0043] The system adopts a "shallow water equation and vortex constraint coupled model" based on the vertical average simplification of the three-dimensional incompressible fluid motion equations. The core governing equations include the continuity equation, the momentum conservation equation, and the vortex transport equation. Among them, the continuity equation is for mass conservation, the momentum conservation equation includes the x / y directions, and the vortex transport equation is a physical constraint, with the following specific form:

[0044] The continuity equation is as follows:

[0045] ;

[0046] Variable definitions: h(x,y,t): River surface water depth at position (x,y) at time t; u(x,y,t): Average vertical flow velocity along the x-direction at position (x,y) at time t; v(x,y,t): Average vertical flow velocity along the y-direction at position (x,y) at time t; t: time; x,y: spatial coordinates in a Cartesian coordinate system.

[0047] The momentum conservation equation in the x-direction is as follows:

[0048] ;

[0049] (3) The momentum conservation equation in the y-direction is as follows:

[0050] ;

[0051] The above x / y momentum conservation equation is derived from the "vertical average simplification of the three-dimensional incompressible fluid motion equation", which conforms to the shallow water flow assumption, that is, the depth of the river is smaller than the horizontal scale, and the vertical velocity gradient can be ignored.

[0052] Additional variable definition: g: acceleration due to gravity, taken as 9.81 m / s² 2 , is a constant; H: the elevation of the river surface at position (x,y) at time t, measured by a visual sensor or water level gauge; z b The elevation of the riverbed at location (x,y) is provided by topographic data and is a static value. The frictional force on the riverbed in the x / y direction is calculated using Manning's formula:

[0053] Where n is the Manning roughness coefficient and the substrate friction coefficient; ρ is the water density, taken as 1000 kg / m³. 3 ν is a constant; t Eddy viscosity coefficient, calculated using the Smagorinsky model: C s S is the Smagorinsky constant, taken as 0.1, ∆ is the computational grid step size, and S ij For the strain rate tensor components.

[0054] The vorticity transport equation, including vorticity constraint terms, includes the following:

[0055] ;

[0056] Variable definition: ω(x,y,t): Vertical vorticity at position (x,y) at time t, defined as... Characterizing the fluid rotation intensity; ▽ 2 The Laplace operator, i.e.: ;Dω : Eddy dissipation term, with a value of 10 -5 ~10 -4 s -2 The vorticity is dynamically adjusted according to the river flow velocity. The upper limit is used when the flow velocity is >1m / s and the lower limit is used when the flow velocity is <0.3m / s, to ensure that the vorticity does not surge abnormally.

[0057] Eddy constraint threshold: set according to river type; the absolute value of vertical eddy current in natural rivers typically does not exceed 0.5s. -1 Therefore, we set ω∈[-0.5, 0.5]s -1 When the value exceeds this range, a local adjustment is triggered.

[0058] The discretization process of the finite difference method is as follows:

[0059] The computational domain is discretized using a structured Cartesian grid. Let the grid step size in the x-direction be ∆x, the grid step size in the y-direction be ∆y, and the time step be ∆t, satisfying the CFL condition: To avoid numerical instability, the discretization schemes for each equation are as follows:

[0060] Discretization of the continuous equation, including central difference and forward difference, is as follows:

[0061] ;

[0062] Subscript definition: i, j are the x / y direction indices of the grid node, n is the current time step, and n+1 is the next time step;

[0063] Solution objective: From calculate .

[0064] The momentum equation is discretized using forward time difference and a spatial QUICK scheme to suppress numerical diffusion, as follows: Taking the momentum equation in the x-direction as an example, the discretized form is:

[0065] ;

[0066] The convection term Conv, QUICK format is:

[0067] ;

[0068] The pressure term is as follows:

[0069] ;

[0070] H: River surface elevation at position (x, y) at time t, measured by a visual sensor or water level gauge; z b The elevation of the riverbed at location (x,y) is provided by topographic data and is a static value.

[0071] The friction term Fric is as follows:

[0072] ;

[0073] The diffusion term Diff, with its central difference, is as follows:

[0074] The discrete scheme of the momentum equation in the y-direction is symmetric to that in the x-direction.

[0075] The vorticity transport equation is discretized, i.e., central difference + forward difference, as follows:

[0076] ; : The vorticity dissipation term at the (i,j) grid node at time step n; and the "D" defined above. ω "This is a spatiotemporal index representation of the same physical quantity, with the same value selection rules as D." ω Consistent; the solution logic is: first calculate using the momentum equation. , then calculate ,like Exceeding This will trigger a local adjustment.

[0077] The methods for determining key parameters include the following:

[0078] The Manning roughness coefficient, n, also known as the substrate friction coefficient, is determined based on the riverbed material and surface roughness. Specific values ​​are referenced in the "Hydrological Measurement Standard" (GB50179-2015), as shown in the following example:

[0079] Bed material Manning coefficient n range Typical values Smooth sandy bed 0.012～0.015 0.013 Gravel bed (5 to 20 cm particle size) 0.020～0.025 0.022 Boulder bed (> 20 cm particle size) 0.030～0.040 0.035 Weedy silt bed 0.035～0.050 0.040

[0080] In practical applications, the value of n can be calculated by back-calculating the flow velocity and water level data on site, ensuring that the error is less than 5%.

[0081] The adjustment logic for the vortex constraint term is as follows:

[0082] Each iteration calculates Then, perform the following verifications and adjustments:

[0083] like Decrease the v value of this node by 5%, because Decreasing v can reduce vorticity; recalculate. until ;

[0084] like Reduce the u value of this node by 5% and recalculate. until ;

[0085] The adjustment range should not exceed 20% of the original flow velocity to avoid disrupting the overall rationality of the flow field.

[0086] It should be further explained that the specific implementation of the shallow water equation and vortex constraint coupled model is as follows: The physical model calculation unit constructs the physical reference flow field by simultaneously solving the set of governing equations describing the macroscopic motion of the water body. This set of equations integrates the continuity equation based on mass conservation, the motion equation based on momentum conservation, and an additional vortex transport equation. The continuity equation is used to characterize the relationship between the water flow depth and the velocity field, the motion equation is used to describe the evolution of the velocity field under the action of gravity, bed friction, and pressure gradient, and the introduced vortex constraint equation is specifically used to characterize and restrict the generation, transport, and dissipation processes of fluid rotation. This constraint ensures that the generated physical reference flow field conforms to the vortex structure and shear flow characteristics commonly found in actual rivers by tracking and correcting the spatial distribution and temporal variation of vorticity in the flow field.

[0087] In terms of numerical solution, the physical model computation unit uses the finite difference method to discretize the above-mentioned coupled equations, dividing the continuous solution domain into a structured computational grid. The received topographic data is used to determine the elevation value at each grid node, thereby accurately characterizing the influence of the riverbed topography on the water flow. Boundary condition data is used to set the constraints at the edge of the computational domain, including the riverbanks being regarded as solid walls without slippage, and the velocity or water level boundary conditions set at the upstream inlet and downstream outlet based on measured data. Real-time hydrological measurement data, such as the initial velocity or water level at a specific cross section, is used as the initial field to start the model computation. By iteratively solving this discretized coupled system, the physical model computation unit finally outputs a physical reference flow field that is physically self-consistent and constrained by vorticity laws.

[0088] The adaptive weighting algorithm in the data fusion module is implemented as follows: in regions where the visual uncertainty matrix shows high reliability, the initial surface flow field is given a higher fusion weight; in regions where the visual uncertainty matrix shows low reliability, the physical reference flow field is given a higher fusion weight; the allocation of fusion weights adopts a continuous function form to avoid abrupt changes at the region boundaries.

[0089] It should be further explained that the adaptive weighting algorithm in the data fusion module is implemented as follows: The weight allocation unit receives the visual uncertainty matrix from the visual observation branch and uses each reliability value in the matrix as an input parameter of the Gaussian distribution function for mapping processing; by preset and adjusting the two key parameters of the Gaussian function, the system controls the sensitivity of the system to reliability changes and the smooth width of the weight transition zone between different reliability regions.

[0090] In the specific weight allocation, for regions with high reliability values ​​in the visual uncertainty matrix, the Gaussian function outputs a higher weight value and assigns it to the corresponding data in the initial surface flow field, while correspondingly reducing the weight of the physical reference flow field in that region; conversely, for regions with low reliability values, a lower weight value is output and assigned to the initial surface flow field, while increasing the contribution weight of the physical reference flow field in that region.

[0091] The fusion computing unit then performs weighted fusion calculations on each corresponding flow field vector in the initial surface flow field and the physical reference flow field according to this dynamically allocated weight mapping. The weight allocation adopts a continuously changing function to ensure that the fusion weights of the two flow field data present a smooth transition at the boundary between the high visual reliability region and the low reliability region. This avoids the abrupt changes or discontinuities in flow field data that may occur at the region boundary in the traditional threshold segmentation method, and finally achieves seamless fusion of two heterogeneous flow field data.

[0092] After the physical rule derivation branch receives the first version of the reconstructed flow field, the physical model calculation unit uses this flow field as the new initial condition to re-execute the solution process of the shallow water equation and vortex constraint coupled model. By comparing the difference between the second calculation result and the first version of the reconstructed flow field, flow field correction parameters are generated. The data fusion module corrects the first version of the reconstructed flow field according to the flow field correction parameters.

[0093] It should be further explained that the specific implementation method for the secondary verification and correction process of the physical rule derivation branch for the first version of the reconstructed flow field is as follows:

[0094] After the data fusion module generates the first version of the reconstructed flow field, the system uses it as an optimized and more physically plausible initial condition, feeding it back to the physical model calculation unit of the physical rule derivation branch. This calculation unit replaces the original initial hydrological conditions with this first version of the reconstructed flow field, and, while maintaining the same topographic data and boundary conditions, re-executes the complete solution process of the shallow water equation and eddy current constraint coupled model, thus obtaining a secondary calculated flow field that is physically highly compatible with the fused flow field. Subsequently, the system initiates a difference analysis process, comparing the secondary calculated flow field with the first version of the reconstructed flow field point by point. The vector differences in velocity magnitude and direction are calculated, and their difference norm is used to generate a spatially distributed flow field correction parameter field. This correction parameter quantitatively characterizes the degree of deviation between the fused flow field and the physical model under higher accuracy requirements. Finally, the data fusion module receives this flow field correction parameter and refines the first version of the reconstructed flow field accordingly. Specifically, in areas where the correction parameter indicates a large deviation, the first version of the reconstructed flow field is directionally corrected based on the data from the secondary calculation of the flow field, while in areas with a small deviation, the original fusion result remains basically unchanged, thus completing the second physical enhancement and optimization of the reconstruction result.

[0095] It should be further explained that the difference between the above comparison of the secondary calculation results and the first version of the reconstructed flow field is specifically calculated using the L2 norm, i.e., the Euclidean norm. The formula and application logic are as follows:

[0096] The difference norm between the velocity components (u1, v1) of the first reconstructed flow field and the velocity components (u2, v2) of the second calculated flow field is:

[0097] ;

[0098] Variable definition: : The velocity difference vector between two flow fields at the same node; The L2 norm of the velocity difference vector represents the magnitude of the velocity difference.

[0099] Application logic: When When the velocity difference exceeds the normal fluctuation range of a natural river, flow field correction parameters are generated. The first version of the reconstructed flow field is corrected using the following formula: , .

[0100] The spatiotemporal continuity verification unit identifies anomalous vectors that do not conform to the fluid motion characteristics by analyzing the variation of the direction and magnitude of the flow field vectors in the time series and spatial grid. For the identified anomalous vectors, the interpolation results of adjacent normal vectors are used to replace them to ensure the continuity and rationality of the output flow field in the spatiotemporal dimensions.

[0101] It should be further explained that the final processing procedure of the spatiotemporal continuity verification unit for the secondary verification flow field is implemented as follows:

[0102] After receiving the flow field data that has undergone secondary verification by the physical model, this unit first calculates the difference in velocity magnitude and direction between each flow field vector and its eight adjacent directional vectors in the spatial dimension, and uses the Laplace operator to detect spatial smoothness, identifying anomalous vectors that show significant abrupt changes from the surrounding flow field patterns. Simultaneously, in the temporal dimension, it compares and analyzes the current flow field with the flow field sequence of several consecutive time steps, calculates the rate of change of each position vector on the time axis, and detects anomalous points that show non-physical jumps in the motion trajectory. For the anomalous vectors jointly identified by the above spatiotemporal analysis, the system does not directly discard them, but adopts an intelligent repair mechanism based on the continuity of fluid motion. That is, it uses the spatial distribution of effective vectors around the anomalous point to regenerate replacement vector values ​​through a bilinear interpolation algorithm, where the interpolation weight is determined by the spatial distance between the adjacent vector and the point and its own reliability. This processing method not only eliminates the interference of local outliers on the overall flow field, but also maintains the continuous structure of the flow field in spatial distribution and the smooth evolution in the time series, ultimately outputting a physically self-consistent and data-complete final river surface flow field.

[0103] It also includes a results display module, which is connected to the results output module, to visualize the final surface flow field in the form of a vector diagram, and to use different colors or markers in the visualization interface to distinguish the reliability level of each location in the flow field.

[0104] Further explanation is needed regarding the implementation process of the results display module, which is as follows: This module receives the final surface flow field data and its corresponding reliability information from the results output module, and drives the graphics display device to generate a composite visualization interface. During the visualization process, the flow field data is first converted into a vector map covering the river area, where the flow velocity and direction are represented by directional arrows, and the length and color tone of the arrows reflect the absolute value of the flow velocity. At the same time, the system maps the reliability information of each flow field vector location, which has undergone multi-level verification, into an independent visual channel. Specifically, a continuous color transition scheme from warm to cool tones is used to distinguish different reliability levels. High reliability areas are given a specific color, medium reliability areas are given another transitional color, and low reliability areas are given a third clearly distinguishable color. These colors representing reliability are superimposed under the flow field vector arrows in a semi-transparent fill manner, or used as the fill color of the arrow body. This allows users to intuitively and instantly identify the reliability of data in different areas of the flow field while observing the overall structure and movement trend of the flow field, providing an intuitive and reliable basis for subsequent data analysis and decision-making.

[0105] The vortex constraint term in the shallow water equation coupled model ensures that the physical reference flow field conforms to the vortex motion law of the actual river by restricting the generation and dissipation process of vortex in the flow field. After each iteration, the physical model calculation unit checks the satisfaction of the vortex constraint and makes local adjustments to the regions that do not meet the constraint.

[0106] It should be further explained that the implementation mechanism of vortex constraint in the coupled model of shallow water equations and vortex constraint is as follows:

[0107] During each iteration of the physical model calculation unit, after updating the flow field variables based on the finite difference method, the system simultaneously calculates the vorticity field distribution across the entire computational grid. This vorticity field is obtained through curl calculation of the flow field velocity vector, and its magnitude directly characterizes the rotational intensity of local fluid micro-clusters. Subsequently, the system compares the calculated vorticity field with a preset vorticity threshold range based on river hydrological characteristics, identifying grid nodes with abnormally high or low vorticity values. For these identified abnormal vorticity regions, the system does not simply employ global smoothing but initiates a localized physical adjustment process: first, it analyzes the source terms of abnormal vorticity based on the vorticity transport equation, and then corrects the vorticity value by fine-tuning the tangential component of the local velocity field, bringing it back to a reasonable physical range. This adjustment process is repeatedly executed in each iteration, forming a dynamic vorticity constraint closed loop embedded in the main flow field solution process, thereby ensuring that the final generated physical reference flow field not only satisfies the conservation of mass and momentum but also that the spatial distribution and intensity of its vortex structure better match the observed kinematic characteristics of river fluids.

[0108] The preset vorticity threshold range is selected based on the following criteria:

[0109] Eddy constraint threshold: Determined based on the fluid motion characteristics of natural rivers and industry-measured data. The specific selection method is as follows:

[0110] 1. Referring to the "Characteristics of Eddy Motion in Natural Rivers" section of the industry standard "River Dynamics" (GB / T50179-2015), the measured absolute value of vertical vorticity in surface water flow of natural rivers is within the range of [-0.3, 0.3] s. -1 ;

[0111] 2. Considering vorticity fluctuations under extreme hydrological conditions (such as rapids and eddies), a safety factor of 1.67 is reserved, and the final vorticity threshold range is set as Ω∈[-0.5,0.5]s. -1 ;

[0112] 3. If applied to specific types of rivers (such as artificial canals or mountain rapids), the eddy current data of more than 3 sets of typical working conditions can be measured on-site, and the upper limit of the threshold can be taken as 1.2 times the maximum value and the lower limit of the threshold can be taken as 1.2 times the minimum value to ensure that the threshold is adapted to the specific application scenario.

[0113] The allocation of fusion weights adopts a continuous weight allocation strategy based on Gaussian function. The reliability values ​​in the visual uncertainty matrix are used as input parameters of Gaussian function. By adjusting the mean and variance of Gaussian function, the sensitivity of weight allocation and the width of the transition zone are controlled.

[0114] It should be further explained that the specific mathematical form of the Gaussian function is as follows:

[0115] The system uses a univariate Gaussian function to calculate the fusion weight w of the initial surface flow field. vis The weight w of the physical reference flow field phy Determined by the "complementarity principle", The specific form is as follows:

[0116] ;

[0117] Variable definition: r i,j : The reliability value at grid node (i,j) in the (x,y) matrix of the visual uncertainty matrix, ranging from [0,1], where r=1 indicates complete reliability and r=0 indicates complete unreliability; μ: The mean of the Gaussian function, the reliability baseline threshold, controlling the center of the weight sensitivity interval; σ: The variance of the Gaussian function, controlling the sensitivity of the weights to changes in reliability. The smaller σ is, the more sensitive the weights are to changes in r; w vis (r i,j ): The initial surface flow field fusion weight at node (i,j), with a value range of [0,1].

[0118] The mean μ and variance σ are determined as follows:

[0119] Both μ and σ are dynamically calculated based on the reliability statistical characteristics of the visual uncertainty matrix. The specific steps are as follows:

[0120] Calculate the reliability statistic: the reliability r of all nodes in the visual uncertainty matrix. i,j Statistical analysis yields the median reliability, Med(r). r i,j The median value after sorting reflects the overall reliability level; the reliability standard deviation Std(r) reflects the dispersion of reliability. , The mean reliability is N, and the total number of grid nodes is N.

[0121] The dynamic setting of the mean μ is as follows: μ is adjusted according to the reliability median Med(r) to ensure that the weight-sensitive interval matches the actual reliability distribution:

[0122] When Med(r) ≥ 0.7, indicating high overall visual reliability: take μ = 0.8, so that the area where r ≥ 0.7 obtains a higher weight, w vis ≥ 0.8;

[0123] When 0.3 < Med(r) < 0.7, indicating medium overall visual reliability: take μ = Med(r), so that the weight-sensitive interval coincides with the center of the overall reliability;

[0124] When Med(r) ≤ 0.3, indicating low overall visual reliability: take μ = 0.2, only the area where r ≥ 0.3 obtains a lower weight, that is, w vis ≤ 0.5 to avoid unreliable visual data from affecting the fusion result.

[0125] The dynamic setting of the variance σ is as follows: σ is adjusted according to the reliability standard deviation Std(r) to control the width of the weight transition zone:

[0126] When Std(r) ≥ 0.2, the reliability dispersion degree is large, and there is an obvious high / low reliability boundary: take σ = 0.15, so that the weight changes rapidly with r, strengthening the effect of "emphasize vision in the high reliability area and physics in the low reliability area";

[0127] When Std(r) < 0.2, the reliability dispersion degree is small, and the overall reliability is uniform: take σ = 0.25, so that the weight changes slowly with r, avoiding sudden weight changes at the regional boundary.

[0128] The weight is calculated as follows:

[0129] Taking the visual uncertainty matrix of the river section as an example, its reliability statistical characteristics are Med(r) = 0.6, Std(r) = 0.18, then:

[0130] The mean μ = Med(r) = 0.6;

[0131] The variance σ = 0.25, because Std(r) = 0.18 < 0.2;

[0132] For r i,j = 0.9, indicating the high reliability area: ​​​​​​​​​​​​​​For r i,j =0.6 indicates a medium reliability zone: w phy =0.0, currently focusing on visual flow field, will be corrected through secondary verification later.

[0135] It should be further explained that the specific implementation process of the continuous weight allocation strategy based on Gaussian functions is as follows:

[0136] The weight allocation unit inputs each reliability value in the visual uncertainty matrix into a preset Gaussian distribution function. The mathematical expression of this function determines the continuous response characteristics of its output value as the input reliability changes. By setting the mean parameter of the Gaussian function, the system essentially defines a baseline threshold for reliability. When the reliability value approaches this mean, the weight allocation enters a sensitive transition range. Adjusting the variance parameter of the Gaussian function controls the width of the transition region from high to low weight; a larger variance results in a smoother transition, while a smaller variance results in a steeper transition. In practical operation, this Gaussian... The output value of the function is directly mapped to the fusion weight of the initial surface flow field, while its complementary value is used as the weight of the physical reference flow field. This achieves the mutual increase and decrease of the two flow field data during the fusion process. This allocation mechanism based on continuous mathematical functions ensures that the contribution weights of the two flow fields do not change abruptly at the boundary between areas with good visual observation quality and areas with poor quality. Instead, they present a smooth and natural gradual transition. This fundamentally avoids the introduction of artificial false boundaries or discontinuities in the final fused flow field due to abrupt changes in weight allocation, and improves the spatial consistency and physical rationality of the reconstructed flow field.

[0137] It also includes a data storage module, which connects the visual observation branch, the physical rule deduction branch, the data fusion module, and the result output module. It is used to store intermediate data and final results generated during system operation. The data storage module organizes data in chronological order and establishes data indexes for subsequent querying and analysis.

[0138] It should be further explained that the data storage module is implemented as follows:

[0139] This module establishes a structured time-series database to systematically record and manage multi-source heterogeneous data generated throughout the reconstruction process. During the data entry phase, the module not only stores the final surface flow field generated by the output module, but also records, in chronological order, the initial surface flow field and visual uncertainty matrix generated by the visual observation branch, the physical baseline flow field calculated by the physical rule deduction branch, the intermediate fused flow fields of various versions generated by the data fusion module, and the corresponding fusion weight distribution maps. All stored data is appended with precise timestamps and data processing stage tags, and by establishing cross-data table association indexes, users can trace the original visual data, intermediate processing results, and their verification process corresponding to any final flow field result. The database adopts a hierarchical storage architecture, storing frequently updated real-time data separately from historical baseline data, and has a dedicated data retrieval interface that supports combined queries by time range, spatial region, or data processing stage. This systematic data management mechanism not only ensures the full traceability of the reconstruction process but also provides a complete data foundation for subsequent analysis of flow field evolution patterns and optimization of model parameters.

[0140] It should be further explained that the data indexing implementation method and data organization logic of the data storage module are as follows:

[0141] The data storage module uses a two-dimensional index of "timestamp + spatial coordinates" to organize data, with the specific structure as follows:

[0142] Time Index: A unique timestamp is assigned to the data stream at each time step in the format "YYYY-MM-DD-HH-MM". For example, 2024-05-20-14-30 represents the flow field data at 14:30 on May 20, 2024. All data are sorted in ascending order by timestamp.

[0143] Spatial Indexing: For the flow field data at each time step, a grid node index is established according to the "x-coordinate - y-coordinate". For example, the node index for x=100m and y=50m is "100-50". Each spatial index corresponds to the velocity components (u, v), reliability r, and fusion weight (w) of that node. vis ,w phy );

[0144] Query interface: Supports "time range query", such as querying flow field data from 14:00 to 15:00 on May 20, 2024; "spatial region query", such as querying flow field data for x∈[50,150]m and y∈[30,70]m; and "spatiotemporal combination query", such as querying historical data for nodes x=100m and y=50m at 14:30 on May 20, 2024. The query response time is ≤1s, and it is implemented using a B+ tree index of MySQL database.

[0145] The system effectively overcomes the flow field measurement failure problem caused by visual information degradation in complex environments by constructing a closed-loop reconstruction mechanism that combines visual observation and physical rule deduction in parallel with bidirectional verification. Utilizing a weighted fusion strategy dynamically guided by the visual uncertainty matrix, the system can fully preserve visual measurement details when image quality is good, and automatically enhance the constraints of the physical model when visual information is unreliable, thereby ensuring the spatial integrity and continuity of the output flow field. Furthermore, multiple safeguards, including secondary physical verification and spatiotemporal continuity checks, improve the overall performance of the reconstruction results in terms of physical rationality and data reliability.

[0146] By deeply embedding river hydrodynamics knowledge into the visual computing process, an intelligent analysis system with self-verification and correction capabilities has been formed. It can provide stable data support for hydrological monitoring, flood control scheduling, and aquatic ecological research without being strictly limited by meteorological conditions, solving the problem of insufficient adaptability in practical applications and realizing a technological leap from relying on single sensor data to multi-source information fusion cognition.

[0147] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0148] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A system for reconstructing the physical rules of river surface flow field, characterized in that, It includes a visual sensor, a visual observation branch, a physical rule deduction branch, a data fusion module, and a result output module; The visual sensor is installed in the river observation area to continuously acquire video image sequences of the river surface; The visual observation branch is connected to a visual sensor and includes a feature extraction unit and a flow field calculation unit. The feature extraction unit identifies water surface feature points from the video image sequence and generates the motion trajectory of the feature points. The flow field calculation unit generates an initial surface flow field based on the motion trajectory of the feature points and simultaneously generates a visual uncertainty matrix to characterize the reliability of each position in the initial surface flow field. The physical rule deduction branch includes a hydrological data receiving unit and a physical model calculation unit. The hydrological data receiving unit acquires river topographic data, boundary condition data, and real-time hydrological measurement data. The physical model calculation unit, based on the coupled model of shallow water equations and eddy constraint, independently calculates and generates a physical reference flow field according to the received data. The data fusion module is connected to the visual observation branch and the physical rule deduction branch respectively. It includes a weight allocation unit and a fusion calculation unit. The weight allocation unit dynamically adjusts the fusion weights of the initial surface flow field and the physical reference flow field at corresponding positions according to the reliability values ​​of each position in the visual uncertainty matrix. The fusion calculation unit uses an adaptive weighting algorithm to perform fusion calculations on the two flow fields to generate the first version of the reconstructed flow field. The data fusion module also feeds back the first version of the reconstructed flow field to the physical rule deduction branch. The physical model calculation unit performs secondary calculations using the first version of the reconstructed flow field as the initial condition to generate a secondary verification flow field. The result output module is connected to the data fusion module, including a spatiotemporal continuity verification unit. The spatiotemporal continuity verification unit performs a continuity analysis of the secondary verification flow field in terms of time and space, removes abnormal vectors that do not conform to the laws of fluid motion, and outputs the final surface flow field. The generation of the visual uncertainty matrix includes: the feature extraction unit performs quality assessment on the video image sequence, sets different reliability values ​​for raindrop interference areas, water surface reflection areas and low contrast areas respectively, and the flow field calculation unit further corrects the reliability values ​​at each position based on the success rate of feature point tracking and trajectory continuity to form a complete visual uncertainty matrix. The shallow water equations and eddy constraint coupled model in the physical rule derivation branch establish the physical constraint relationship of the river surface flow field by simultaneously solving the mass conservation equation, momentum conservation equation, and eddy transport equation. The physical model calculation unit uses the finite difference method to discretize and solve the coupled model. Topographic data is used to determine the elevation distribution of the calculation grid, boundary condition data is used to set the boundary constraints of the riverbank and inlet / outlet, and real-time hydrological measurement data is used to provide the initial conditions for model calculation.

2. The system for reconstructing the physical rules of river surface flow field according to claim 1, characterized in that: The adaptive weighting algorithm in the data fusion module is implemented as follows: in regions where the visual uncertainty matrix shows high reliability, the initial surface flow field is assigned a higher fusion weight; in regions where the visual uncertainty matrix shows low reliability, the physical reference flow field is assigned a higher fusion weight. The allocation of fusion weights adopts a continuous function form to avoid abrupt changes at the region boundaries.

3. The system for reconstructing the physical rules of river surface flow field according to claim 1, characterized in that: After the physical rule derivation branch receives the first version of the reconstructed flow field, the physical model calculation unit uses this flow field as the new initial condition to re-execute the solution process of the shallow water equation and vortex constraint coupled model. By comparing the difference between the second calculation result and the first version of the reconstructed flow field, flow field correction parameters are generated. The data fusion module corrects the first version of the reconstructed flow field according to the flow field correction parameters.

4. The system for reconstructing the physical rules of river surface flow field according to claim 1, characterized in that: The spatiotemporal continuity verification unit analyzes the variation patterns of the direction and magnitude of the flow field vectors in the time series and spatial grid to identify abnormal vectors that do not conform to the fluid motion characteristics. For the identified abnormal vectors, the interpolation results of adjacent normal vectors are used to replace them to ensure the continuity and rationality of the output flow field in the spatiotemporal dimension.

5. The system for reconstructing the physical rules of river surface flow field according to claim 1, characterized in that: It also includes a results display module, which is connected to the results output module, to visualize the final surface flow field in the form of a vector diagram, and to use different colors or markers in the visualization interface to distinguish the reliability level of each location in the flow field.

6. The system for reconstructing the physical rules of river surface flow field according to claim 1, characterized in that: The vortex constraint term in the shallow water equation coupled with the vortex constraint model ensures that the physical reference flow field conforms to the vortex motion law of the actual river by restricting the generation and dissipation process of vortex in the flow field. The physical model calculation unit checks the satisfaction of the vortex constraint after each iteration and makes local adjustments to the regions that do not meet the constraint.

7. The system for reconstructing the physical rules of river surface flow field according to claim 1, characterized in that: The fusion weights are allocated using a continuous weight allocation strategy based on a Gaussian function. The reliability values ​​in the visual uncertainty matrix are used as input parameters for the Gaussian function. By adjusting the mean and variance of the Gaussian function, the sensitivity of the weight allocation and the width of the transition zone are controlled.

8. The system for reconstructing the physical rules of river surface flow field according to claim 1, characterized in that: It also includes a data storage module, which connects the visual observation branch, the physical rule deduction branch, the data fusion module, and the result output module. It is used to store intermediate data and final results generated during system operation. The data storage module organizes the data in chronological order and establishes a data index for subsequent querying and analysis.

Images