Intelligent water wave analysis method and system for water conservancy project

Abstract

The invention discloses an intelligent water wave analysis method and system for a water conservancy project, and relates to the technical field of image processing, and the method comprises the steps: collecting a water surface image through a multispectral camera and a laser radar, extracting static environment features based on a target detection algorithm, and generating a water surface two-dimensional static environment map; capturing a dynamic image by using a high-speed camera, recognizing the movement direction of the water ripples at the moment when the object is in contact with the water surface according to the timestamp and the recognized dynamic water ripple shape, and generating a dynamic direction data set; performing multi-dimensional training on the neural network model according to a time sequence and a fluid mechanics constraint condition in combination with an object form parameter and a natural condition parameter, outputting a water wave event classification label, and obtaining event label classifications corresponding to different water wave features; water surface and underwater conditions are comprehensively identified, water waves are comprehensively analyzed, the water waves are efficiently analyzed according to physical constraint characteristics of fluid mechanics, a multi-dimensional prediction model is constructed, and the water waves are efficiently, comprehensively and comprehensively classified and predicted.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an intelligent water pattern analysis method and system for a hydraulic engineering project. Background Art

[0002] Water conservancy projects are a series of construction projects carried out for the purpose of managing, utilizing, protecting water resources, and flood control and drainage. With the development of society and technological progress, the scale of water conservancy projects has gradually increased, and the complexity and variability of water conservancy systems have also increased. In this context, how to accurately and effectively analyze the dynamic behavior of water flow in water conservancy projects, especially the changes in water ripples in water flow, has become an urgent problem to be solved.

[0003] At present, the Chinese invention patent with application number CN201910197655.5 discloses a water pollution emission source database and its establishment method. This invention first proposed the concept of three-dimensional molecular weight watermarks and incorporated it into the database. It has the characteristics of large amount of water information, mature and simple selection index analysis and testing methods, less equipment required, low cost, strong operability and timeliness, etc. It can quickly and low-costly realize pollution source tracing, which is conducive to large-scale promotion and has important significance for water pollution tracing. However, the existing technology cannot fully identify the water surface and underwater conditions, conduct a comprehensive analysis of watermarks, cannot efficiently analyze watermarks according to the physical constraint characteristics of fluid mechanics, cannot build a multi-dimensional prediction model, and cannot efficiently and comprehensively classify and predict watermarks. Summary of the invention

[0004] The technical problem solved by the present invention is that the existing technology cannot fully identify the water surface and underwater conditions, conduct a comprehensive analysis of water ripples, cannot efficiently analyze water ripples according to the physical constraint characteristics of fluid mechanics, cannot build a multi-dimensional prediction model, and cannot efficiently and comprehensively classify and predict water ripples.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for intelligent water pattern analysis of a hydraulic engineering project, comprising the following steps:

[0006] Step S1: Collect water surface images through a multispectral camera and a laser radar, extract static environment features based on a target detection algorithm, and generate a two-dimensional static environment map of the water surface;

[0007] Step S2: using a high-speed camera to capture dynamic images, identifying the direction of water ripple movement at the moment when the object touches the water surface according to the timestamp and the identified dynamic water ripple shape, and generating a dynamic direction data set;

[0008] Step S3: According to the time series and fluid mechanics constraints, the neural network model is trained in multiple dimensions in combination with the object morphology parameters and natural condition parameters, and the water ripple event classification labels are output to obtain event label classifications corresponding to different water ripple features;

[0009] Step S4: Automatically generate a dynamic water ripple propagation simulation animation according to the state label selected by the user, or automatically display the predicted state label and its probability according to the water ripple type.

[0010] Preferably, the step S1 comprises:

[0011] Multispectral cameras and lidar collect different bands of water surface images and precise depth information. The different bands of data include visible light, infrared spectrum and ultraviolet spectrum.

[0012] The convolutional neural network (CNN) is used to extract static environmental features in the water surface image, including floating objects, rocks, bridge piers and plants, identify static objects and background environment, and use OpenCV to draw a two-dimensional static environment map of the water surface;

[0013] The semantic segmentation model is used to separate the water surface area and the non-water surface area, and the image distortion is corrected through the point cloud data fusion technology, so as to optimize the two-dimensional static environment map of the water surface into a high-precision two-dimensional static environment topology map of the water surface.

[0014] Preferably, step S2 comprises:

[0015] Capturing a dynamic video of water ripples on a water surface by a high-speed camera, splitting the dynamic video of water ripples into a two-dimensional water ripple change line image sequence sorted in time series, wherein the two-dimensional water ripple change line image sequence includes water ripple lines that change in time and outer contour lines of an object from contacting the water surface to being completely submerged in the water surface;

[0016] When the two-dimensional water ripple change line image sequence corresponding to the time series includes an object touching the water surface, the water ripple lines in each frame in each time series are radially truncated with the midpoint of the object as the center, the midpoint of the truncated water ripple lines is obtained, the midpoints of the water ripple lines corresponding to the same time series are connected to obtain the direction trajectory of the water ripple moving in time, the midpoint of the outermost water ripple line of the last frame is used as the end point coordinate, the midpoint of the object is used as the original coordinate, and the water ripple direction coordinate is drawn;

[0017] When the two-dimensional water ripple change line image sequence corresponding to the time series does not include an object touching the water surface, taking the midpoint of the connecting line of the midpoints of the two closest water ripples in the first frame of the two-dimensional water ripple change line image as the center and the coordinate origin, randomly select points on the water ripples of the time series corresponding to the same curvature radius to connect, and obtain the direction trajectory of the water ripple moving in time;

[0018] The point coordinates included in the direction trajectory are subjected to abnormal data identification and cleaning by using a Gaussian filtering algorithm to obtain a smooth dynamic direction data set.

[0019] Preferably, k-means clustering is performed on the dynamic direction data set to obtain a first type of dynamic direction, a second type of dynamic direction, ... and a kth type of dynamic direction.

[0020] Preferably, the step S3 comprises:

[0021] The dynamic direction data set is input into the LSTM time series processing model to capture the characteristics of the dynamic evolution of water flow. The constraint module of the physical information neural network PINN is embedded in the fluid mechanics equation as a regular term.

[0022] Preferably, the object morphological parameters include the curvature distribution of the object's three-dimensional contour, the type, size and shape of the object, and the natural condition parameters include wind direction, wind speed, water flow, undercurrent, temperature, tide, terrain and ice and snow melting;

[0023] The object morphological parameters and natural condition parameters are respectively standardized, and the standardized data are label-encoded into neural network state labels as output feature values;

[0024] The state labels corresponding to various dynamic directions and object morphological parameters and the state labels corresponding to natural condition parameters are matched to obtain the state label sets corresponding to various dynamic directions.

[0025] Preferably, the time series is a dynamic directional data set, and the fluid mechanics constraints include water velocity, turbulence characteristics, density and viscosity;

[0026] The fluid mechanics constraints and time series are uniformly standardized and input into the embedding feature extraction layer to extract conditional features with the same scale, the number of nodes in the neural network input layer is reconstructed to be the total number of conditional features, and the conditional features are multi-classified through the Softmax activation function;

[0027] Acquire several dynamic direction data sets, use the dynamic direction data sets and corresponding conditional features as training sets to train the reconstructed neural network, obtain event feature label predictions corresponding to various dynamic directions, the event feature label predictions include state labels corresponding to various dynamic directions and event occurrence probabilities corresponding to the state labels, and obtain an intelligent water ripple analysis model for water conservancy projects.

[0028] Preferably, dynamically adjusting the hyperparameters of the intelligent water pattern analysis model for water conservancy projects using the Bayesian optimization algorithm includes:

[0029] The hyperparameters include the time step weight coefficient of the fluid mechanics equation and the feature selection threshold of the natural condition parameter, the time step weight coefficient range is set to α∈[0.1,1.0], α is the time step weight coefficient, and the feature selection threshold range is set to β∈[0.01,0.5], β is the feature selection threshold;

[0030] The mathematical expression of the objective function is:

[0031] F = L(α, β);

[0032] Wherein, F is the objective function, and L(α, β) is the loss function of the intelligent water ripple analysis model of the water conservancy project;

[0033] The loss function is minimized by using Bayesian optimization to maximize the performance of the intelligent water ripple analysis model for water conservancy projects. The performance indicators include the accuracy of the intelligent water ripple analysis model for water conservancy projects in predicting the water ripple movement law.

[0034] Preferably, step S4 comprises:

[0035] Based on the three-dimensional hydrodynamic model, the dynamic water ripple propagation simulation animation is automatically generated according to the status label selected by the user, and AR technology is superimposed to realize the virtual and real fusion of environmental status display;

[0036] Based on the intelligent water ripple analysis model for water conservancy projects, the predicted state label and its probability are output according to the dynamic direction corresponding to the identified water ripple type.

[0037] A water conservancy project intelligent water pattern analysis system, the system is used to execute a water conservancy project intelligent water pattern analysis method, characterized in that it includes a static environment module, a dynamic water pattern recognition module, a prediction label module and a visualization module:

[0038] The static environment module is used to collect water surface images through a multispectral camera and a laser radar, extract static environment features based on a target detection algorithm, and generate a two-dimensional static environment map of the water surface;

[0039] The dynamic water ripple recognition module is used to capture dynamic images using a high-speed camera, identify the direction of water ripple movement at the moment when an object touches the water surface according to the timestamp and the identified dynamic water ripple shape, and generate a dynamic direction data set;

[0040] The prediction label module is used to perform multi-dimensional training on the neural network model according to the time series and fluid mechanics constraints, combined with the object morphology parameters and natural condition parameters, output the water ripple event classification labels, and obtain the event label classification corresponding to different water ripple features;

[0041] The visualization module is used to automatically generate a dynamic water ripple propagation simulation animation according to the state label selected by the user, or automatically display the predicted state label and its probability according to the water ripple type.

[0042] Beneficial effects of the present invention: The present invention provides more detailed and comprehensive environmental information of the water surface through multi-spectral cameras and laser radars, enhances the ability to identify the water surface environment, makes the static object identification more accurate, and uses the Gaussian filter algorithm to identify and clean abnormal data of the dynamic direction trajectory of the water ripples to obtain a smooth dynamic direction data set, further improving the accuracy and reliability of the data. Combined with the identification of the direction of the water ripple movement when the object contacts the water surface, a dynamic direction data set is formed, and K-means clustering is performed to further subdivide the water ripple types. The application of this method in water ripple analysis is more intelligent and efficient than traditional methods. Fluid mechanics constraints are combined in the neural network model, which enables the model to more accurately predict the occurrence and development of water ripple events when processing dynamic water ripples. Combined with object morphological parameters and natural conditions, the neural network is trained in multiple dimensions, which improves the adaptability and accuracy of the model in complex natural environments. The present invention outputs water ripple event classification labels through training based on time series and fluid mechanics constraints, increases the deduction of physics and fluid mechanics, and improves robustness, so that the model can provide more accurate predictions when facing different natural conditions and water ripple events. The application of AR technology makes the dynamic water ripple propagation simulation not only limited to static display, but also allows users to experience a more intuitive water ripple propagation process through interaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A schematic diagram of the basic flow of an intelligent water pattern analysis method for a water conservancy project provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0045] Reference Figure 1 , as an embodiment of the present invention, provides a method for intelligent water pattern analysis of a water conservancy project, comprising the following steps:

[0046] Step S1: Collect water surface images through a multispectral camera and a laser radar, extract static environment features based on a target detection algorithm, and generate a two-dimensional static environment map of the water surface;

[0047] Step S2: using a high-speed camera to capture dynamic images, identifying the direction of water ripple movement at the moment when the object touches the water surface according to the timestamp and the identified dynamic water ripple shape, and generating a dynamic direction data set;

[0048] Step S3: According to the time series and fluid mechanics constraints, the neural network model is trained in multiple dimensions in combination with the object morphology parameters and natural condition parameters, and the water ripple event classification labels are output to obtain event label classifications corresponding to different water ripple features;

[0049] Step S4: Automatically generate a dynamic water ripple propagation simulation animation according to the state label selected by the user, or automatically display the predicted state label and its probability according to the water ripple type.

[0050] Comprehensively identify water surface and underwater conditions, conduct comprehensive analysis of water ripples, efficiently analyze water ripples based on the physical constraint characteristics of fluid mechanics, build a multi-dimensional prediction model, and classify and predict water ripples efficiently and comprehensively.

[0051] Step S1 includes:

[0052] Multispectral cameras and lidar collect different bands of water surface image data and precise depth information. The different bands of data include visible light, infrared spectrum and ultraviolet spectrum. Multispectral images can help capture water surface details, while lidar data helps obtain precise water surface contour and height information.

[0053] Use convolutional neural network (CNN) to extract static environmental features from water surface images. Static environmental features include floating objects, rocks, bridge piers, and plants. Identify static objects and background environments, and use OpenCV to draw a two-dimensional static environment map of the water surface.

[0054] The semantic segmentation model is used to separate the water surface area from the non-water surface area, and the image distortion is corrected through point cloud data fusion technology. The two-dimensional static environment map of the water surface is optimized into a high-precision two-dimensional static environment topology map of the water surface.

[0055] Step S2 includes:

[0056] A high-speed camera is used to capture a dynamic video of water ripples on the water surface, and the dynamic video of the water ripples is split into a two-dimensional water ripple change line image sequence sorted in time series. The two-dimensional water ripple change line image sequence includes water ripple lines that change in time and outer contour lines of an object from contacting the water surface to being completely submerged in the water surface;

[0057] When the two-dimensional water ripple change line image sequence corresponding to the time series includes an object touching the water surface, the water ripple lines in each frame in each time series are radially truncated with the midpoint of the object as the center, the midpoint of the truncated water ripple lines is obtained, and the midpoints of the water ripple lines corresponding to the same time series are connected to obtain the direction trajectory of the water ripple moving in time, and the midpoint of the outermost water ripple line of the last frame is used as the end point coordinate, and the midpoint of the object is used as the original coordinate to draw the water ripple direction coordinate;

[0058] When the two-dimensional water ripple changing line image sequence corresponding to the time series does not include an object touching the water surface, the midpoint of the connecting line of the two closest water ripples in the first frame of the two-dimensional water ripple changing line image is taken as the center and the origin of the coordinates, and points corresponding to the same curvature radius on the water ripples in the time series are randomly selected and connected to obtain the direction trajectory of the water ripple moving in time;

[0059] The Gaussian filtering algorithm is used to identify and clean abnormal data of the point coordinates included in the direction trajectory to obtain a smooth dynamic direction data set. This vector set represents the direction and speed of all water ripples on the water surface.

[0060] The dynamic direction data set is clustered using k-means to obtain the first type of dynamic direction, the second type of dynamic direction, ... and the kth type of dynamic direction.

[0061] Step S3 includes:

[0062] The dynamic direction dataset is input into the LSTM time series processing model to capture the dynamic evolution characteristics of water flow. The constraint module based on the physical information neural network PINN is embedded in the fluid mechanics equation as a regular term.

[0063] Object morphological parameters include the curvature distribution of the object's three-dimensional contour, the object's type, size, and shape, and natural condition parameters include wind direction, wind speed, current, undercurrent, temperature, tide, terrain, and ice and snow melting;

[0064] The object morphological parameters and natural condition parameters are standardized respectively, and the standardized data are labeled and encoded into neural network state labels as output feature values;

[0065] The state labels corresponding to various dynamic directions and object morphological parameters and the state labels corresponding to natural condition parameters are matched to obtain the state label sets corresponding to various dynamic directions.

[0066] The time series is a dynamic directional data set, and the fluid mechanics constraints include the flow velocity, turbulence characteristics, density, and viscosity of the water;

[0067] The fluid mechanics constraints and time series are standardized uniformly and input into the embedding feature extraction layer to extract conditional features with the same scale. The number of nodes in the neural network input layer is reconstructed to be the total number of conditional features. The conditional features are multi-classified through the Softmax activation function, and the network output value is converted into the probability distribution of the category.

[0068] Several dynamic direction data sets are obtained, and the dynamic direction data sets and the corresponding conditional features are used as training sets to train the reconstructed neural network to obtain the event feature label predictions corresponding to each type of dynamic direction. The event feature label predictions include the state labels corresponding to each type of dynamic direction and the event occurrence probability corresponding to the state labels, and an intelligent water ripple analysis model for water conservancy projects is obtained.

[0069] For example, assuming there are three categories: objects entering the water, wave impacts, and no obvious water ripple events, the output of the model may be:

[0070] [0.1,0.7,0.2]

[0071] This means that the event has a 10% probability of belonging to the object entering the water category, a 70% probability of belonging to the wave impact category, and a 20% probability of belonging to the no water ripple event category.

[0072] The Bayesian optimization algorithm is used to dynamically adjust the hyperparameters of the intelligent water pattern analysis model for water conservancy projects, including:

[0073] Hyperparameters include the time step weight coefficient of the fluid mechanics equation and the feature selection threshold of the natural condition parameter. By combining dynamic water ripple data with the theory of fluid mechanics, the motion law of water ripples is further analyzed. The time step weight coefficient range is set to α∈[0.1,1.0], where α is the time step weight coefficient. The feature selection threshold of the natural condition parameter is used to control which features should be included in the input of the neural network. By adjusting this threshold, features with little impact on model prediction can be removed to improve model performance. The feature selection threshold range is set to β∈[0.01,0.5], where β is the feature selection threshold.

[0074] The purpose is to evaluate the performance of the neural network model under given hyperparameter settings. The mathematical expression of the objective function is:

[0075] F = L(α, β);

[0076] Among them, F is the objective function, and L(α,β) is the loss function of the intelligent water ripple analysis model of water conservancy projects;

[0077] Bayesian optimization is used to minimize the loss function and maximize the performance of the intelligent water ripple analysis model for water conservancy projects. The performance indicators include the accuracy of the intelligent water ripple analysis model for water conservancy projects in predicting the movement laws of water ripples.

[0078] Step S4 includes:

[0079] Based on the three-dimensional hydrodynamic model, the dynamic water ripple propagation simulation animation is automatically generated according to the status label selected by the user, and AR technology is superimposed to realize the virtual and real fusion of environmental status display;

[0080] Based on the intelligent water ripple analysis model of water conservancy projects, the predicted state label and its probability are output according to the dynamic direction corresponding to the identified water ripple type.

[0081] A water conservancy project intelligent water ripple analysis system, which is used to execute a water conservancy project intelligent water ripple analysis method, including a static environment module, a dynamic water ripple recognition module, a prediction label module and a visualization module:

[0082] The static environment module is used to collect water surface images through multispectral cameras and lidar, extract static environment features based on target detection algorithms, and generate a two-dimensional static environment map of the water surface;

[0083] The dynamic water ripple recognition module is used to capture dynamic images using a high-speed camera, identify the direction of water ripple movement at the moment when an object touches the water surface based on the timestamp and the identified dynamic water ripple shape, and generate a dynamic direction data set;

[0084] The prediction label module is used to perform multi-dimensional training on the neural network model based on the time series and fluid mechanics constraints, combined with the object morphology parameters and natural condition parameters, output the water ripple event classification labels, and obtain the event label classification corresponding to different water ripple features;

[0085] The visualization module is used to automatically generate dynamic water ripple propagation simulation animations according to the state labels selected by the user, or automatically display the predicted state labels and their probabilities according to the water ripple types.

[0086] The present invention provides more detailed and comprehensive environmental information of the water surface through multi-spectral cameras and laser radars, enhances the recognition ability of the water surface environment, makes the static object recognition more accurate, and uses the Gaussian filtering algorithm to identify and clean the abnormal data of the dynamic direction trajectory of the water ripples, obtains a smooth dynamic direction data set, and further improves the accuracy and reliability of the data. Combined with the recognition of the direction of the water ripple movement when the object contacts the water surface, a dynamic direction data set is formed, and K-means clustering is performed to further subdivide the water ripple types. The application of this method in water ripple analysis is more intelligent and efficient than traditional methods. Fluid mechanics constraints are combined in the neural network model, which enables the model to more accurately predict the occurrence and development of water ripple events when processing dynamic water ripples. Combined with object morphological parameters and natural conditions, the neural network is trained in multiple dimensions, which improves the adaptability and accuracy of the model in complex natural environments. The present invention outputs water ripple event classification labels through training based on time series and fluid mechanics constraints, increases the deduction of physics and fluid mechanics, and improves robustness, so that the model can provide more accurate predictions when facing different natural conditions and water ripple events. The application of AR technology makes the dynamic water ripple propagation simulation not only limited to static display, but also allows users to experience a more intuitive water ripple propagation process through interaction.

[0087] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program codes. Among them, the storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, magnetic disk or optical disk. These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0088] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for intelligent water pattern analysis in water conservancy projects, characterized in that: The following steps are involved: Step S1: Collect water surface images through a multispectral camera and a laser radar, extract static environment features based on a target detection algorithm, and generate a two-dimensional static environment map of the water surface; Step S2: using a high-speed camera to capture dynamic images, identifying the direction of water ripple movement at the moment when the object touches the water surface according to the timestamp and the identified dynamic water ripple shape, and generating a dynamic direction data set; Step S3: According to the time series and fluid mechanics constraints, the neural network model is trained in multiple dimensions in combination with the object morphology parameters and natural condition parameters, and the water ripple event classification labels are output to obtain event label classifications corresponding to different water ripple features; Step S4: automatically generating a dynamic water ripple propagation simulation animation according to the state label selected by the user, or automatically displaying the predicted state label and its probability according to the water ripple type; The step S1 comprises: Multispectral cameras and lidar collect different bands of water surface images and precise depth information. The different bands of data include visible light, infrared spectrum and ultraviolet spectrum. The convolutional neural network (CNN) is used to extract static environmental features in the water surface image, including floating objects, rocks, bridge piers and plants, identify static objects and background environment, and use OpenCV to draw a two-dimensional static environment map of the water surface; The semantic segmentation model is used to separate the water surface area and the non-water surface area, and the image distortion is corrected through the point cloud data fusion technology, so as to optimize the two-dimensional static environment map of the water surface into a high-precision two-dimensional static environment topology map of the water surface.

2. The intelligent water pattern analysis method for water conservancy projects according to claim 1, characterized in that: The step S2 comprises: Capturing a dynamic video of water ripples on a water surface by a high-speed camera, splitting the dynamic video of water ripples into a two-dimensional water ripple change line image sequence sorted in time series, wherein the two-dimensional water ripple change line image sequence includes water ripple lines that change in time and outer contour lines of an object from contacting the water surface to being completely submerged in the water surface; When the two-dimensional water ripple change line image sequence corresponding to the time series includes an object touching the water surface, the water ripple lines in each frame in each time series are radially truncated with the midpoint of the object as the center, the midpoint of the truncated water ripple lines is obtained, the midpoints of the water ripple lines corresponding to the same time series are connected to obtain the direction trajectory of the water ripple moving in time, the midpoint of the outermost water ripple line of the last frame is used as the end point coordinate, the midpoint of the object is used as the original coordinate, and the water ripple direction coordinate is drawn; When the two-dimensional water ripple change line image sequence corresponding to the time series does not include an object touching the water surface, taking the midpoint of the connecting line of the midpoints of the two closest water ripples in the first frame of the two-dimensional water ripple change line image as the center and the coordinate origin, randomly select points on the water ripples of the time series corresponding to the same curvature radius to connect, and obtain the direction trajectory of the water ripple moving in time; The point coordinates included in the direction trajectory are subjected to abnormal data identification and cleaning by using a Gaussian filtering algorithm to obtain a smooth dynamic direction data set.

3. The intelligent water pattern analysis method for water conservancy projects according to claim 2, characterized in that: The dynamic direction data set is subjected to k-means clustering to obtain a first type of dynamic direction, a second type of dynamic direction, ... and a kth type of dynamic direction.

4. The intelligent water pattern analysis method for water conservancy projects according to claim 3, characterized in that: The step S3 comprises: The dynamic direction data set is input into the LSTM time series processing model to capture the characteristics of the dynamic evolution of water flow. The constraint module of the physical information neural network PINN is embedded in the fluid mechanics equation as a regular term.

5. The intelligent water pattern analysis method for water conservancy projects according to claim 4, characterized in that: The object morphological parameters include the curvature distribution of the object's three-dimensional contour, the type, size and shape of the object, and the natural condition parameters include wind direction, wind speed, water flow, undercurrent, temperature, tide, terrain and ice and snow melting; The object morphological parameters and natural condition parameters are respectively standardized, and the standardized data are label-encoded into neural network state labels as output feature values; The state labels corresponding to various dynamic directions and object morphological parameters and the state labels corresponding to natural condition parameters are matched to obtain the state label sets corresponding to various dynamic directions.

6. The intelligent water pattern analysis method for water conservancy projects according to claim 5, characterized in that: The time series is a dynamic directional data set, and the fluid mechanics constraints include water velocity, turbulence characteristics, density and viscosity; The fluid mechanics constraints and time series are uniformly standardized and input into the embedding feature extraction layer to extract conditional features with the same scale, the number of nodes in the neural network input layer is reconstructed to be the total number of conditional features, and the conditional features are multi-classified through the Softmax activation function; Acquire several dynamic direction data sets, use the dynamic direction data sets and corresponding conditional features as training sets to train the reconstructed neural network, obtain event feature label predictions corresponding to various dynamic directions, the event feature label predictions include state labels corresponding to various dynamic directions and event occurrence probabilities corresponding to the state labels, and obtain an intelligent water ripple analysis model for water conservancy projects.

7. The intelligent water pattern analysis method for water conservancy projects according to claim 6, characterized in that: The Bayesian optimization algorithm is used to dynamically adjust the hyperparameters of the intelligent water pattern analysis model for water conservancy projects, including: The hyperparameters include the time step weight coefficient of the fluid mechanics equation and the feature selection threshold of the natural condition parameter, the time step weight coefficient range is set to α∈[0.1,1.0], α is the time step weight coefficient, and the feature selection threshold range is set to β∈[0.01,0.5], β is the feature selection threshold; The mathematical expression of the objective function is: F = L(α, β); Wherein, F is the objective function, and L(α, β) is the loss function of the intelligent water ripple analysis model of the water conservancy project; The loss function is minimized by using Bayesian optimization to maximize the performance of the intelligent water ripple analysis model for water conservancy projects. The performance indicators include the accuracy of the intelligent water ripple analysis model for water conservancy projects in predicting the water ripple movement law.

8. The intelligent water pattern analysis method for water conservancy projects according to claim 7, characterized in that: The step S4 comprises: Based on the three-dimensional hydrodynamic model, the dynamic water ripple propagation simulation animation is automatically generated according to the status label selected by the user, and AR technology is superimposed to realize the virtual and real fusion of environmental status display; Based on the intelligent water ripple analysis model for water conservancy projects, the predicted state label and its probability are output according to the dynamic direction corresponding to the identified water ripple type.

9. A water conservancy project intelligent water pattern analysis system, the system is used to execute a water conservancy project intelligent water pattern analysis method, characterized in that: Including static environment module, dynamic water pattern recognition module, prediction label module and visualization module: The static environment module is used to collect water surface images through a multispectral camera and a laser radar, extract static environment features based on a target detection algorithm, and generate a two-dimensional static environment map of the water surface; Multispectral cameras and lidar collect different bands of water surface images and precise depth information. The different bands of data include visible light, infrared spectrum and ultraviolet spectrum. The convolutional neural network (CNN) is used to extract static environmental features in the water surface image, including floating objects, rocks, bridge piers and plants, identify static objects and background environment, and use OpenCV to draw a two-dimensional static environment map of the water surface; A semantic segmentation model is used to separate the water surface area from the non-water surface area, and image distortion is corrected by point cloud data fusion technology to optimize the water surface two-dimensional static environment map into a high-precision water surface two-dimensional static environment topology map; The dynamic water ripple recognition module is used to capture dynamic images using a high-speed camera, identify the direction of water ripple movement at the moment when an object touches the water surface according to the timestamp and the identified dynamic water ripple shape, and generate a dynamic direction data set; The prediction label module is used to perform multi-dimensional training on the neural network model according to the time series and fluid mechanics constraints, combined with the object morphology parameters and natural condition parameters, output the water ripple event classification labels, and obtain the event label classification corresponding to different water ripple features; The visualization module is used to automatically generate a dynamic water ripple propagation simulation animation according to the state label selected by the user, or automatically display the predicted state label and its probability according to the water ripple type.

Images