A three-dimensional deformation monitoring and early warning method for reservoir dams based on image recognition

Abstract

The present invention relates to the technical field of reservoir dam deformation monitoring, and specifically to a three-dimensional deformation monitoring and early warning method for reservoir dams based on image recognition. First, multi-view two-dimensional image data of the reservoir dam is collected, and a three-dimensional model of the reservoir dam is constructed and the surface deformation area is extracted through image processing technology; then the deformation data of the deformation area and the environmental data are fused to obtain the instantaneous deformation of the reservoir dam; time series analysis is performed based on the instantaneous deformation and deformation data to predict the probability of future deformation. Finally, the predicted deformation and the instantaneous deformation are comprehensively analyzed to obtain the deformation early warning index, thereby realizing the automated deformation monitoring and early warning of the reservoir dam. The present invention combines image processing technology and deep learning methods to improve the accuracy and real-time performance of reservoir dam deformation monitoring, thereby improving the quality and efficiency of reservoir dam operation management.

Description

Technical Field

[0001] The invention relates to the technical field of reservoir dam deformation monitoring, and in particular to a reservoir dam three-dimensional deformation monitoring and early warning method based on image recognition. Background Art

[0002] Compared with dams in other scenarios, reservoir dams have certain particularities in design, function and use environment. The core function of reservoir dams is to store water, that is, to intercept and store water flow to form a reservoir, so as to realize functions such as hydropower generation, flood control irrigation and water supply regulation. Due to the large amount of water involved in reservoir dams, their pressure bearing capacity, dam structure and anti-seepage measures are also higher than other types of dams. In terms of dam scale, reservoir dams are usually larger than other dams, especially large reservoir dams, which can reach tens of meters or even hundreds of meters in height. In summary, reservoir dams mainly focus on water storage and water flow regulation. They are usually designed to be large in scale, diverse in functions, withstand high water pressure and focus on long-term stability. According to the above characteristics and needs, it is particularly important to accurately monitor the deformation of reservoir dams.

[0003] The existing technology for dam deformation monitoring integrates a variety of data processing technologies, including multi-source information fusion, deep learning and machine learning. Use multiple types of sensors to collect monitoring data of the dam, adopt neural network models to predict the deformation of the dam; combine automatic data collection with manual input to monitor the operating status of the dam in real time. The existing technology analyzes the geometric properties and environmental factors of the dam from multiple angles and dimensions to achieve the assessment of the safety of the dam. However, due to the particularity of the scale and function of reservoir dams, the existing technology that relies on sensors is not suitable for deformation monitoring of reservoir dams. The deployment and maintenance of a large number of sensors will bring new obstacles to the deformation monitoring of reservoir dams and the safety management of reservoir dams. At the same time, the quality of data collected by sensors installed on the structure of reservoir dams is greatly affected by factors such as the environment and instrument accuracy, resulting in the accuracy and real-time performance of reservoir dam deformation monitoring still need to be improved.

[0004] In order to improve the accuracy and real-time performance of reservoir dam deformation monitoring, thereby improving the quality and efficiency of reservoir dam operation management, the present invention proposes a reservoir dam three-dimensional deformation monitoring and early warning method based on image recognition. Summary of the invention

[0005] The purpose of the present invention is to provide a three-dimensional deformation monitoring and early warning method for a reservoir dam based on image recognition. First, multi-view two-dimensional image data of the reservoir dam is collected, and a three-dimensional model of the reservoir dam is constructed and the surface deformation area is extracted through image processing technology; then, the deformation data of the deformation area and the environmental data are fused to obtain the instantaneous deformation of the reservoir dam; based on the instantaneous deformation and deformation data, time series analysis is performed to predict the probability of future deformation. Finally, the deformation early warning index is obtained by comprehensively analyzing the predicted deformation and the instantaneous deformation, thereby realizing the automated deformation monitoring and early warning of the reservoir dam.

[0006] To achieve the above-mentioned purpose, the present invention provides a reservoir dam three-dimensional deformation monitoring and early warning method based on image recognition, comprising: obtaining two-dimensional image data of the reservoir dam; processing the two-dimensional image data using computer vision technology to obtain standard image data; generating three-dimensional model data of the reservoir dam using a stereo matching algorithm for the standard image data, wherein the three-dimensional model data includes the geometric shape and size of the reservoir dam;

[0007] For the three-dimensional model data, a deep convolutional neural network is used to identify the deformation area of ​​the surface of the reservoir dam, and extract the three-dimensional deformation data of the surface of the reservoir dam;

[0008] Collecting environmental data of the reservoir dam, fusing the environmental data with the three-dimensional deformation data into comprehensive deformation data, optimizing the comprehensive deformation data by a multi-source data fusion method, and generating a comprehensive deformation assessment model; and assessing the instantaneous deformation of the reservoir dam by using the comprehensive deformation assessment model;

[0009] Extracting the temporal correlation between the comprehensive deformation data and the instantaneous deformation, and establishing a deformation prediction regression model according to the temporal correlation; the deformation prediction regression model is used to obtain the predicted deformation of the reservoir dam;

[0010] Through the predicted deformation and the instant deformation, the reservoir dam is monitored in real time and a deformation warning index is calculated. When the deformation warning index exceeds a preset safety threshold, a reservoir dam safety warning is triggered.

[0011] Furthermore, the acquisition process of standard image data includes image preprocessing and feature point matching;

[0012] Acquire two-dimensional image data of the reservoir dam from multiple angles, covering the dam top, both sides of the dam body and the dam foundation area of ​​the reservoir dam; pre-process the two-dimensional image data to obtain a two-dimensional processed image;

[0013] Extract key feature points from the two-dimensional processed image, perform feature matching on the two-dimensional processed images at different angles according to the key feature points, and acquire the standard image data.

[0014] Furthermore, the acquisition of the three-dimensional model data includes:

[0015] For the standard image data, the parallax between the image pairs is calculated, and the three-dimensional spatial coordinates of each pixel are calculated in combination with the parameters of the image acquisition device, and all the generated three-dimensional points constitute the point cloud data of the reservoir dam;

[0016] The point cloud data is converted into a three-dimensional surface model by using triangulation, and the three-dimensional model data of the reservoir dam is obtained by a simplified algorithm; the three-dimensional model data includes the geometric shape and size of the reservoir dam.

[0017] Furthermore, the process of identifying the deformation area from the three-dimensional model data includes:

[0018] The three-dimensional coordinates of each point in the three-dimensional model data are mapped into a feature map, which is input into the convolution layer after standardization to extract the local features of the three-dimensional model data; for each input feature map , through the convolution kernel Perform a convolution operation:

[0019] ;

[0020] in, is the coordinate of the input feature map, The output feature map is at position The value of For the input feature map at position The pixel value of is the value of the convolution kernel, and is the offset of the convolution kernel, and They are and The value range of

[0021] right Extract key features through pooling operation, and then output the position through the fully connected layer deformation probability; identifying the position coordinates where the deformation probability exceeds a preset value as the deformation area of ​​the reservoir dam.

[0022] Further, extracting the three-dimensional deformation data of the surface of the reservoir dam;

[0023] Extracting three-dimensional coordinates and corresponding displacement values ​​of the deformation area; the displacement value is calculated by the deformation three-dimensional coordinates of each deformation position and the three-dimensional coordinates of the original point;

[0024] Deformation Position The three-dimensional coordinates of the original point are , The three-dimensional coordinates of the deformation are , the deformation position The displacement value The calculation formula is:

[0025] ;

[0026] The three-dimensional coordinates of the deformation area and the displacement value of the deformation position are output as the three-dimensional deformation data.

[0027] Furthermore, the step of generating the comprehensive deformation assessment model includes:

[0028] The environmental data includes temperature and humidity data, air pressure data, precipitation and flow data of the area where the reservoir dam is located; the environmental data and the three-dimensional deformation data are aligned in time and optimized by a filtering algorithm; the spatial features of the optimized three-dimensional deformation data are extracted by a multi-layer convolutional neural network;

[0029] The spatial features are spliced ​​with the optimized environmental data, and a multi-layer fully connected layer is used to integrate the features and generate the result of the comprehensive deformation assessment through the output layer; the generation process of the comprehensive deformation assessment is optimized through a loss function to obtain the comprehensive deformation assessment model.

[0030] Furthermore, the step of evaluating the instantaneous deformation of the reservoir dam includes: obtaining real-time environmental data and real-time three-dimensional deformation data of the reservoir dam, aligning the real-time environmental data and the real-time three-dimensional deformation data in time and inputting them into the comprehensive deformation evaluation model to obtain the instantaneous deformation of the reservoir dam; the instantaneous deformation includes a real-time deformation position and a corresponding evaluation deformation value.

[0031] Further, the predicted deformation of the reservoir dam is obtained by combining the comprehensive deformation data with the instantaneous deformation;

[0032] A long short-term memory network is used to extract the temporal association between the comprehensive deformation data and the instant deformation, and a deformation prediction regression model is established:

[0033] ;

[0034] in, for The predicted deformation value at time for the temporal association of moments; and is the regression coefficient, is the total number of moments;

[0035] The predicted deformation of the reservoir dam is outputted through the deformation prediction regression model.

[0036] Furthermore, the deformation warning index calculation process includes: for each deformation position ,pass Calculating the difference between the instantaneous deformation and the predicted deformation; wherein, is the deformation position exist The deformation difference at time, is the deformation position exist The instant deformation of the moment, is the deformation position exist The predicted deformation at time instant;

[0037] Sum the deformation differences of all deformation positions to obtain the deformation of the reservoir dam at The overall deformation difference at each moment ,right Normalization is performed to obtain the deformation warning index.

[0038] Furthermore, the safety warning trigger mechanism of the reservoir dam includes: dividing the risk threshold range and setting the deformation risk level of the reservoir dam; judging the risk threshold range in which the deformation warning index is located and outputting the corresponding deformation risk level; when the deformation warning index exceeds the warning reminder threshold, triggering a safety warning reminder.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] 1. The present invention first obtains high-resolution two-dimensional image data of the reservoir dam from multiple angles, including the dam top, both sides of the dam body and the dam foundation area, and forms standardized two-dimensional image data after preprocessing and geometric correction; then the standard image data is processed by a stereo matching algorithm to generate a three-dimensional model of the reservoir dam; finally, a deep convolutional neural network is used to analyze the three-dimensional model data, identify the deformation area on the surface of the reservoir dam and extract the three-dimensional deformation data of the area. This method can obtain accurate three-dimensional deformation data without relying on a large number of sensors through image recognition technology, reduce the complexity of sensor deployment, and cover a larger range of the dam surface; the deep learning network can identify the specific area where the reservoir dam surface is deformed, thereby providing basic data support for improving the accuracy and real-time performance of subsequent deformation assessment and risk prediction.

[0041] 2. The present invention first collects environmental data of the area where the reservoir dam is located, including temperature and humidity, air pressure, precipitation and flow; then aligns the environmental data and three-dimensional deformation data in time to generate comprehensive deformation data; for the three-dimensional deformation data in the comprehensive deformation data, a convolutional neural network is used to extract spatial features; the spatial features and the optimized environmental data are spliced, and the features are integrated through a fully connected layer, and finally the result of the comprehensive deformation is output; finally, a comprehensive deformation assessment model is obtained through loss function optimization. By fusing environmental data and three-dimensional deformation data, the present invention can consider the influence of environmental factors on deformation while obtaining the deformation data of the reservoir dam, thereby obtaining more accurate and dynamic instant deformation; and provide more accurate phased data support for subsequent deformation assessment and early warning of the reservoir dam.

[0042] 3. The present invention conducts time series analysis on the comprehensive deformation data and the instant deformation data, uses the long short-term memory network to extract the time series correlation information between the two, and establishes a regression prediction model based on this to predict the deformation of the reservoir dam; for each deformation position, the difference between the predicted deformation and the instant deformation is calculated, and the deformation differences of all deformation positions are summed and normalized to generate a deformation warning index. This method combines the instant deformation and comprehensive deformation data, and predicts the future deformation of the reservoir dam through the time series prediction model, thereby improving the accuracy of the reservoir dam deformation prediction; the predicted deformation and the instant deformation are fused and calculated to obtain the deformation warning index of the reservoir dam, and then combined with the risk classification mechanism to achieve real-time monitoring and early warning of the deformation state of the reservoir dam, thereby providing a scientific basis for the operation and management of the reservoir. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of a reservoir dam three-dimensional deformation monitoring and early warning method based on image recognition;

[0044] Figure 2 This is a structural diagram of a reservoir dam three-dimensional deformation monitoring model based on image recognition proposed by the present invention;

[0045] Figure 3 This is a data flow diagram of the three-dimensional deformation monitoring and early warning method proposed in the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] The present invention provides a reservoir dam three-dimensional deformation monitoring and early warning method based on image recognition. The specific method flow chart is shown in Figure 1 For ease of understanding, the present invention also provides Figure 2 , Figure 2 The structural diagram of the reservoir dam three-dimensional deformation monitoring model based on image recognition proposed by the present invention includes a deformation area recognition model, a comprehensive deformation recognition model and a deformation prediction regression model; and Figure 3 , Figure 3 This is a data flow diagram of the three-dimensional deformation monitoring and early warning method proposed by the present invention. Figure 1 , Figure 2 and Figure 3 The specific implementation method of the reservoir dam three-dimensional deformation monitoring and early warning method proposed in the present invention is described.

[0048] Embodiment 1

[0049] As an embodiment of the present invention, refer to Figure 1 S10 in the embodiment is used to obtain high-resolution two-dimensional image data of a reservoir dam; and the two-dimensional image data is processed using computer vision technology to obtain standard image data. Specifically:

[0050] Acquire high-resolution 2D image data of the reservoir dam from multiple angles to ensure coverage of all important areas of the reservoir dam, including acquiring dam crest images from the top area of ​​the reservoir dam, taking images of both sides of the dam from different sides, and taking images of the dam foundation area from the bottom or downstream side to capture the morphology of the dam foundation. Collect multiple images at each angle to ensure that the clarity of the acquired 2D image data meets the analysis requirements.

[0051] Gaussian filtering is used to denoise the collected two-dimensional image data to remove the noise in the image caused by shooting equipment or environmental factors. Then the brightness, contrast and grayscale of the image are adjusted to ensure that the details in the image are clearly visible. According to the internal and external parameters of the image acquisition device, the image is corrected to remove distortion; the background information in the image that is not related to the reservoir dam is removed by image segmentation to obtain a two-dimensional processed image. In this step, this embodiment uses openCV to implement the preprocessing and geometric correction of the two-dimensional image data.

[0052] A feature point detection algorithm is used to extract key feature points from two-dimensional image data, and a descriptor is generated for each key feature point. The two-dimensional image data at different angles are matched with the help of the descriptors. After matching images at different angles, multiple images are spliced ​​according to the geometric relationship of the matching point pairs to generate complete standard image data. This embodiment uses the SIFT algorithm to complete the extraction and matching of key feature points, and introduces the RANSAC algorithm to achieve the splicing of images at different angles.

[0053] This embodiment first obtains all-round, multi-view high-resolution two-dimensional image data of the reservoir dam to ensure the data integrity of the important areas of the reservoir dam; then, with the help of a variety of image processing algorithms and tools, the two-dimensional image data is pre-processed and geometrically corrected to ensure the accuracy and reliability of subsequent image analysis. Finally, the key feature points in images of different angles are extracted through the feature point detection algorithm, and the multi-view images are matched and spliced ​​to generate complete standard image data, which provides accurate basic data for the subsequent three-dimensional modeling and deformation analysis of the reservoir dam.

[0054] Further, refer to Figure 1 S20 in the method is applied to the three-dimensional model data, using a deep convolutional neural network to identify the deformation area of ​​the surface of the reservoir dam, and extracting the three-dimensional deformation data of the surface of the reservoir dam. Specifically:

[0055] For the standard image data, the disparity between the image pairs is calculated. First, the image pairs are selected for matching, and the disparity calculation formula is used. Calculate the position difference of the pixel in the left and right image pairs; in the parallax calculation formula, is parallax, and are the corresponding horizontal coordinates of the pixel points in the left image and the right image of the image pair respectively.

[0056] According to the parallax, combined with the internal and external parameters of the image acquisition device, the three-dimensional spatial coordinates of each pixel are calculated by triangulation. :

[0057] ;

[0058] in, and are the focal length and baseline distance of the image acquisition device, respectively. is the coordinate of the pixel point in the standard image data, is the optical center coordinate of the image acquisition device.

[0059] The three-dimensional spatial coordinate data calculated above is used to generate point cloud data of the reservoir dam. The point cloud is processed by a triangulation algorithm and converted into a three-dimensional surface model. The three-dimensional surface model is then simplified by Laplacian smoothing to obtain a three-dimensional model of the reservoir dam. The three-dimensional model includes the geometric shape and size information of the reservoir dam.

[0060] In this embodiment, DJI Mavic 3pro is used to collect multi-view high-resolution two-dimensional image data of reservoir dam A. The parameters of the image acquisition equipment are baseline 1000mm and focal length 50mm. Standard image data is obtained through the above-mentioned image preprocessing, geometric correction, feature point matching and stitching. Through the processing processes such as parallax calculation and triangulation, the data example of the three-dimensional model of the reservoir dam is given in Table 1.

[0061] Table 1. Examples of data of the three-dimensional model of the reservoir dam

[0062]

[0063] This embodiment performs parallax calculation on the image pairs in the standard image data, and then calculates the three-dimensional coordinates of each pixel point through triangulation based on the parallax and the internal and external parameters of the image acquisition device to generate three-dimensional point cloud data of the reservoir dam; the two-dimensional image data is converted into three-dimensional space coordinate data through triangulation, thereby ensuring the accuracy of the three-dimensional model of the reservoir dam and providing a data basis for the subsequent identification of the surface deformation of the reservoir dam.

[0064] Furthermore, a deformation area recognition model is constructed to identify the deformation area of ​​the reservoir dam A. The construction process of the deformation area recognition model includes: firstly, deformation annotation of the acquired three-dimensional model data of the reservoir dam is performed, deformation is calculated by comparing the three-dimensional model data of the reservoir dam in the historical time series, and annotation is generated by the difference of the three-dimensional data at the previous and next time points;

[0065] The 3D coordinates of each point in the 3D model data are mapped to a feature map. Each 3D coordinate contains and Three values, then each 3D point can be represented as a pixel position in the feature map, and the pixel value corresponds to the 3D point Coordinates; for three-dimensional coordinates The values ​​are standardized to adapt them to the input requirements of the neural network. The standardization method used in this embodiment is ,in, After standardization value, and They are The mean and standard deviation of the values.

[0066] The standardized 3D model data is input into the local features extracted by the convolution layer; for each input feature map , through the convolution kernel Perform a convolution operation:

[0067] ;

[0068] in, is the coordinate of the input feature map, is the value of the output feature map, For the input feature map at position The pixel value of is the value of the convolution kernel, and is the offset of the convolution kernel, and They are and The value range of .

[0069] For the output feature map , extract important information through maximum pooling and reduce the size of the feature map:

[0070] ;

[0071] In the above formula, is the output after maximum pooling, To offset the output feature map to The pixel value of is the size of the pooling window. In this embodiment, .

[0072] Will pass Flattened to a one-dimensional vector Then input it to the fully connected layer and the output layer to obtain the deformation probability of each pixel position; is the flattening operation. Each pixel position The deformation probability The formula for obtaining is:

[0073] ;

[0074] in, is the pixel position The output vector of the corresponding fully connected layer. Set the deformation preset threshold , the deformation probability is greater than The position coordinates are identified as the deformation area of ​​the reservoir dam .

[0075] ;

[0076] This embodiment collects real-time 3D model data of reservoir dam A and sets the deformation preset threshold It is set to 0.5, and the deformation area data of reservoir dam A are identified by the deformation area identification model and are given in Table 2.

[0077] Table 2 Partial data of deformation area of ​​reservoir dam A

[0078]

[0079] This embodiment maps the three-dimensional coordinates of each point in the three-dimensional model data of the reservoir dam into a feature map and performs conversion processing, thereby converting the three-dimensional data into a form suitable for deep learning model input; automatically learns and extracts local features in the three-dimensional model through the convolution layer, and then aggregates important information, which helps to understand complex deformation patterns; generates the deformation probability of each pixel position through the fully connected layer and the output layer, and identifies the deformation area of ​​the reservoir dam by setting a threshold, thereby improving the accuracy of deformation area detection.

[0080] Extracting three-dimensional coordinates and corresponding displacement values ​​of the deformation area; the displacement value is calculated by the deformation three-dimensional coordinates of each deformation position and the three-dimensional coordinates of the original point;

[0081] Deformation Position The three-dimensional coordinates of the original point are , The three-dimensional coordinates of the deformation are , the deformation position The displacement value ;

[0082] ;

[0083] The three-dimensional coordinates of the deformation area and the displacement value of the deformation position are output as the three-dimensional deformation data.

[0084] This embodiment gives an example of the three-dimensional deformation data of the reservoir dam part A in Table 3;

[0085] Table 3 Partial example of 3D deformation data of reservoir dam A

[0086]

[0087] This embodiment extracts the three-dimensional coordinates of each deformation position and calculates the coordinate difference before and after the deformation, so as to obtain an accurate deformation value and quantitatively evaluate the deformation degree of each deformation position; quantifying the deformation degree can provide an accurate data basis for intuitive understanding of the deformation and subsequent further analysis and learning.

[0088] Reference Figure 1 S30 in the embodiment is used to collect the environmental data of the reservoir dam, fuse the environmental data with the three-dimensional deformation data into comprehensive deformation data, optimize the comprehensive deformation data through a multi-source data fusion method, and generate a comprehensive deformation assessment model; the instantaneous deformation of the reservoir dam is assessed through the comprehensive deformation assessment model. Specifically:

[0089] The deformation of a reservoir dam is not only affected by its structure itself, but also by environmental factors. For example, temperature and humidity data will affect the physical properties of the reservoir dam building materials such as expansion and contraction. Air pressure data is a meteorological factor that affects the stability of the reservoir dam. Precipitation directly affects the reservoir water level. Flow data reflects the dynamic changes of the reservoir water flow, affecting the stress distribution and deformation of the reservoir dam body. For example, in extreme weather conditions, hurricanes, heavy precipitation or high temperatures may directly cause the deformation of the dam. In this case, relying solely on three-dimensional deformation data cannot fully reflect the cause and nature of the deformation. Moreover, environmental factors may cause short-term deformation changes, and relying solely on deformation data may mistakenly believe that the reservoir dam has structural problems or potential risks. By combining environmental factors, it is possible to more clearly distinguish between normal deformation caused by environmental factors and abnormal deformation caused by structural problems, thereby effectively reducing false alarms and misjudgments and improving the efficiency of reservoir dam management and maintenance.

[0090] The training process of the comprehensive deformation assessment model includes: standardizing the environmental data and aligning it with the three-dimensional deformation data in time, and optimizing it through a filtering algorithm. When the acquisition frequency of the environmental data and the three-dimensional deformation data is inconsistent or there are abnormalities in the data at some time points, linear interpolation is used to complete them. In this embodiment, Kalman filtering is used to denoise and optimize the environmental data and the three-dimensional deformation data.

[0091] The optimized three-dimensional deformation data is used to extract spatial features through a multi-layer convolutional neural network. The specific steps are:

[0092] The three-dimensional deformation data is input into a three-layer convolutional neural network to extract spatial features; the number of channels of the input data is the number of coordinate axes 3. The number of convolution kernels of the three layers of convolution is 64, 128 and 256 respectively; the convolution kernel sizes are 3×3×3, 3×3×3 and 5×5×5 respectively; the step sizes are all 1. After each layer of convolution, maximum pooling is used to achieve downsampling. The pooling window size of the three layers of maximum pooling is 2×2×2, and the step size is 2.

[0093] The spatial features and the standardized environmental data are spliced ​​and input into the fully connected layer. The fully connected layer further integrates and transforms the feature vectors, and outputs the estimated deformation value of each deformation position of the reservoir dam at each time point through the softmax of the output layer. The acquisition process of the estimated deformation value is optimized through the loss function to obtain a comprehensive deformation assessment model.

[0094] The real-time environmental data and real-time three-dimensional deformation data of the reservoir dam are obtained and aligned in time, and input into the comprehensive deformation assessment model to obtain the instantaneous deformation of the reservoir dam; the instantaneous deformation includes a real-time deformation position and a corresponding assessment deformation value.

[0095] This embodiment extracts spatial features from three-dimensional deformation data through a multi-layer convolutional neural network, and then integrates environmental data to build a comprehensive deformation assessment model, which can more comprehensively consider various factors affecting reservoir dams. The combination of environmental data and deformation data helps to identify the root cause of deformation, avoid misjudging environmental fluctuations caused by environmental changes, and provide more accurate reservoir dam deformation assessment results.

[0096] This embodiment obtains instant deformation through a comprehensive deformation assessment model for real-time environmental data and three-dimensional deformation data, and can timely capture external factors that affect the deformation of the reservoir dam. The impact of environmental factors on deformation is highly timely. Combined with real-time environmental data, it can timely determine whether the source of deformation is caused by environmental factors or structural problems, thereby providing a more scientific basis for response decision-making.

[0097] Reference Figure 1 S40 in the embodiment is used to extract the temporal correlation between the comprehensive deformation data and the instantaneous deformation, and obtain the predicted deformation of the reservoir dam by establishing a deformation prediction regression model. Specifically:

[0098] Collect the comprehensive deformation data and instant deformation data at a specified time, and align the two at the time point; use the long short-term memory network to extract the temporal association between the comprehensive deformation data and the instant deformation, and establish a deformation prediction regression model:

[0099] ;

[0100] in, for The predicted deformation value at time for the temporal association of moments; is the regression coefficient, is the total number of moments;

[0101] The mean square error is used as the loss function to minimize the difference between the predicted deformation and the actual deformation, and the optimized deformation prediction regression model is used to output the real-time predicted deformation of the reservoir dam.

[0102] This embodiment uses time series data of comprehensive deformation and instantaneous deformation for modeling, and can identify the deformation law of the reservoir dam in historical data; time series analysis is not limited to static data, but through continuous updating and learning, it can adjust the prediction according to real-time deformation and environment, thereby improving the accuracy and dynamic adaptability of reservoir dam deformation prediction.

[0103] Reference Figure 1S50 in the embodiment is used to monitor the reservoir dam in real time and calculate the deformation warning index through the predicted deformation and the instant deformation, and trigger the reservoir dam safety warning when the deformation warning index exceeds the preset safety threshold. Specifically:

[0104] For each deformation position ,pass Calculating a difference between the instantaneous deformation and the predicted deformation;

[0105] in, is the deformation position exist The deformation difference at the moment reflects the deviation between the instantaneous deformation and the predicted deformation at that position; is the deformation position exist The instant deformation of the moment, is the deformation position exist The predicted deformation at time instant;

[0106] Sum the deformation differences of all deformation positions to obtain the deformation of the reservoir dam at The overall deformation difference at each moment In order to ensure that the overall deformation can be expressed within a reasonable range and to facilitate subsequent comparison with the preset safety threshold, it is necessary to normalize the overall deformation difference and obtain the deformation warning index:

[0107] ;

[0108] in, for Deformation warning index at all times, is the maximum deformation difference of all deformation positions. Table 4 gives some deformation warning indexes of reservoir dam A. In Table 4, the units of instant deformation, predicted deformation, deformation difference and overall deformation difference are all meters (m).

[0109] Table 4 Example of deformation warning index of reservoir dam A

[0110]

[0111] This embodiment realizes real-time and accurate deformation monitoring of reservoir dams by calculating the difference between the instant deformation and the predicted deformation at the deformation position, and using the difference of all deformation positions to calculate the overall deformation difference of the reservoir dam. By analyzing the difference between the instant deformation and the predicted deformation, the deformation risk is dynamically evaluated and the deformation early warning index is generated, providing timely data support for the safety response decision of the reservoir dam, thereby improving the management efficiency of the reservoir dam.

[0112] Embodiment 2

[0113] Reference Figure 1 S60 in the embodiment is used to divide the risk threshold range, set the deformation risk level of the reservoir dam; determine the risk threshold range of the deformation warning index, and output the corresponding deformation risk level; when the deformation warning index exceeds the warning reminder threshold, trigger a safety warning reminder. Specifically:

[0114] Reference Figure 3 First, the deformation risk level is divided according to the safety management requirements of the reservoir dam A. This embodiment determines the preset threshold value through historical data combined with expert experience, and divides the deformation risk into 4 levels:

[0115] Low risk: deformation warning index range is (0-0.2);

[0116] Medium risk: deformation warning index range is [0.2-0.5);

[0117] High risk: deformation warning index range is [0.5-0.8);

[0118] Extremely high risk: deformation warning index range is [0.8-0.1).

[0119] According to the deformation warning index calculated in real time, the risk level range to which the index belongs is determined. This embodiment provides the deformation warning index and corresponding risk level of the reservoir dam A at six time points, as shown in Table 5.

[0120] Table 5 Deformation warning index and corresponding risk level

[0121]

[0122] According to the deformation warning index, the deformation risk level of the reservoir dam at time points t2 to t6 in Table 5 is at medium risk or above. In this embodiment, the warning threshold of the deformation warning index is set to 0.5, and the safety warning reminder is automatically triggered at time point t4.

[0123] This embodiment calculates the deformation warning index by real-time monitoring of the deformation of the reservoir dam, which can effectively identify the potential risks of the dam, promptly discover the deformation beyond the normal range, and realize automated risk assessment and reminder. This method can automatically determine the risk level based on real-time data, trigger early warning reminders, and enable managers to take different emergency measures in time according to different risk levels. Combined with deformation data, reasonable response decisions can be made to improve the quality and efficiency of reservoir dam operation management.

[0124] In order to improve the accuracy and real-time performance of reservoir dam deformation monitoring, thereby improving the quality and efficiency of reservoir dam operation management, the present invention proposes a reservoir dam three-dimensional deformation monitoring and early warning method based on image recognition. In this method, the three-dimensional deformation of the reservoir dam is first monitored by using image recognition technology, which no longer relies too much on the deployment of a large number of sensors, reduces the workload of sensor maintenance for management personnel, and also reduces the problem of inaccurate monitoring data caused by sensor errors. Then, through the three-dimensional deformation data acquisition method based on image recognition, the deformation information of the reservoir dam can be captured more accurately, and a drone equipped with a high-resolution image acquisition device can be used to quickly collect high-quality and wide-coverage high-definition images of the reservoir dam, providing a high-quality data basis for subsequent image analysis; the deep learning methods such as multi-layer convolutional neural networks and long short-term memory networks used in this method can capture the spatial features and temporal features in the three-dimensional deformation data, effectively process complex multi-dimensional data, and improve the accuracy and real-time performance of data processing.

[0125] In acquiring the instant deformation of the reservoir dam, this method also comprehensively considers the environmental data of the reservoir dam area. Not only the deformation of the reservoir dam should be monitored, but also the causes and influencing factors of the deformation should be taken into account, which makes the deformation monitoring of the reservoir dam more comprehensive. This method obtains the predicted deformation of the reservoir dam in the future by performing time series analysis on the instant deformation and comprehensive deformation data, which helps to timely discover potential hidden dangers. Finally, the deformation warning index is calculated by combining the instant deformation and the predicted deformation, thereby realizing the automated deformation monitoring and early warning of the reservoir dam. The present invention combines image processing technology, deep learning methods and data processing means to improve the accuracy and real-time performance of reservoir dam deformation monitoring, and comprehensively improves the quality and efficiency of reservoir dam operation management, thereby providing basic guarantees and support for the development of related water conservancy projects.

[0126] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A three-dimensional deformation monitoring and early warning method for reservoir dams based on image recognition, characterized in that: include: Acquire two-dimensional image data of the reservoir dam; process the two-dimensional image data using computer vision technology to obtain standard image data; For the standard image data, the three-dimensional model data of the reservoir dam is generated by the stereo matching algorithm. The three-dimensional model data includes the geometric shape and size of the reservoir dam; For the three-dimensional model data, a deep convolutional neural network is used to identify the deformation area on the surface of the reservoir dam and extract the three-dimensional deformation data of the reservoir dam surface; Collect environmental data of reservoir dams, fuse environmental data with three-dimensional deformation data into comprehensive deformation data, optimize the comprehensive deformation data through multi-source data fusion method, and generate a comprehensive deformation assessment model; Assess the instantaneous deformation of reservoir dams through a comprehensive deformation assessment model; Extract the temporal correlation between comprehensive deformation data and instant deformation, and establish a deformation prediction regression model based on the temporal correlation; The deformation prediction regression model is used to obtain the predicted deformation of the reservoir dam; The steps of generating the comprehensive deformation assessment model include: environmental data including temperature and humidity data, air pressure data, precipitation and flow data of the area where the reservoir dam is located; aligning the environmental data and the three-dimensional deformation data in time and optimizing them through a filtering algorithm; extracting spatial features from the optimized three-dimensional deformation data through a multi-layer convolutional neural network; splicing the spatial features with the optimized environmental data, integrating the features using a multi-layer fully connected layer and generating the results of the comprehensive deformation assessment through the output layer; optimizing the generation process of the comprehensive deformation assessment through a loss function to obtain a comprehensive deformation assessment model; By predicting deformation and instant deformation, the reservoir dam is monitored in real time and the deformation warning index is calculated. When the deformation warning index exceeds the preset safety threshold, the reservoir dam safety warning is triggered.

2. The method for monitoring and early warning of three-dimensional deformation of a reservoir dam based on image recognition according to claim 1 is characterized in that: The acquisition process of standard image data includes image preprocessing and feature point matching; Acquire two-dimensional image data of the reservoir dam from multiple angles, covering the dam top, both sides of the dam body and the dam foundation area of ​​the reservoir dam; pre-process the two-dimensional image data to obtain a two-dimensional processed image; Extract key feature points from the two-dimensional processed image, perform feature matching on the two-dimensional processed images at different angles according to the key feature points, and acquire the standard image data.

3. The method for monitoring and early warning of three-dimensional deformation of a reservoir dam based on image recognition according to claim 1 is characterized in that: The acquisition of the three-dimensional model data includes: For the standard image data, the parallax between the image pairs is calculated, and the three-dimensional spatial coordinates of each pixel are calculated in combination with the parameters of the image acquisition device, and all the generated three-dimensional points constitute the point cloud data of the reservoir dam; The point cloud data is converted into a three-dimensional surface model by using triangulation, and the three-dimensional model data of the reservoir dam is obtained by a simplified algorithm; the three-dimensional model data includes the geometric shape and size of the reservoir dam.

4. The method for monitoring and early warning of three-dimensional deformation of reservoir dams based on image recognition according to claim 1 is characterized in that: The process of identifying the deformation area from the three-dimensional model data includes: The three-dimensional coordinates of each point in the three-dimensional model data are mapped into a feature map, which is input into the convolution layer after standardization to extract the local features of the three-dimensional model data; for each input feature map , through the convolution kernel Perform a convolution operation: in, is the coordinate of the input feature map, The output feature map is at position The value of For the input feature map at position The pixel value of is the value of the convolution kernel, and is the offset of the convolution kernel, and They are and The value range of right Extract key features through pooling operation, and then output the position through the fully connected layer deformation probability; identifying the position coordinates where the deformation probability exceeds a preset value as the deformation area of ​​the reservoir dam.

5. The method for monitoring and early warning of three-dimensional deformation of reservoir dams based on image recognition according to claim 1 is characterized in that: Extracting the three-dimensional deformation data of the surface of the reservoir dam; Extracting three-dimensional coordinates and corresponding displacement values ​​of the deformation area; the displacement value is calculated by the deformation three-dimensional coordinates of each deformation position and the three-dimensional coordinates of the original point; Deformation Position The three-dimensional coordinates of the original point are , The three-dimensional coordinates of the deformation are , the deformation position The displacement value The calculation formula is: The three-dimensional coordinates of the deformation area and the displacement value of the deformation position are output as the three-dimensional deformation data.

6. The method for monitoring and early warning of three-dimensional deformation of reservoir dams based on image recognition according to claim 1 is characterized in that: The step of evaluating the instantaneous deformation of the reservoir dam comprises: Real-time environmental data and real-time three-dimensional deformation data of the reservoir dam are obtained, and the real-time environmental data and the real-time three-dimensional deformation data are aligned in time and then input into the comprehensive deformation assessment model to obtain the instantaneous deformation of the reservoir dam; the instantaneous deformation includes a real-time deformation position and a corresponding assessment deformation value.

7. The method for monitoring and early warning of three-dimensional deformation of reservoir dams based on image recognition according to claim 1 is characterized in that: Combining the comprehensive deformation data with the instantaneous deformation to obtain a predicted deformation of the reservoir dam; A long short-term memory network is used to extract the temporal association between the comprehensive deformation data and the instant deformation, and a deformation prediction regression model is established: in, for The predicted deformation value at time for the temporal association of moments; and is the regression coefficient, is the total number of moments; The predicted deformation of the reservoir dam is outputted through the deformation prediction regression model.

8. The method for monitoring and early warning of three-dimensional deformation of a reservoir dam based on image recognition according to claim 1 is characterized in that: The calculation process of the deformation warning index includes: For each deformation position ,pass Calculating the difference between the instantaneous deformation and the predicted deformation; wherein, is the deformation position exist The deformation difference at time, is the deformation position exist The instant deformation of the moment, is the deformation position exist The predicted deformation at the time; summing the deformation differences of all deformation positions to obtain the reservoir dam at The overall deformation difference at each moment ,right Normalization is performed to obtain the deformation warning index.

9. The method for monitoring and early warning of three-dimensional deformation of reservoir dams based on image recognition according to claim 1 is characterized in that: The safety early warning triggering mechanism of the reservoir dam includes: Divide the risk threshold range and set the deformation risk level of the reservoir dam; determine the risk threshold range in which the deformation warning index is located, and output the corresponding deformation risk level; when the deformation warning index exceeds the warning reminder threshold, trigger a safety warning reminder.

Images