A Dam Deformation Monitoring and Early Warning Method and System Based on Beidou and Machine Vision

Abstract

The present invention relates to the technical field of dam deformation monitoring, and proposes a dam deformation monitoring and early warning method and system based on Beidou and machine vision, including: collecting the original displacement data of the dam and the original image data of the dam surface; performing RTK differential processing and weighted averaging of multi-point positioning data on the original displacement data to obtain the dam displacement data, and performing multi-spectral analysis and deep learning processing on the original image data to obtain the dam surface structure feature data; performing localization processing on the dam displacement data to obtain real-time displacement change data, and performing parallel processing on the dam surface structure feature data to obtain real-time deformation feature data; performing dynamic fusion on the real-time displacement change data and the real-time deformation feature data to obtain comprehensive deformation data; establishing a dam deformation trend prediction model based on the comprehensive deformation data, and pushing hierarchical early warning information to the monitoring center when abnormal deformation is detected. The present invention improves the real-time performance and early warning reliability of dam deformation monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of dam deformation monitoring, and in particular to a dam deformation monitoring and early warning method and system based on Beidou and machine vision. Background Technique

[0002] With the continuous expansion of the scale of water conservancy project construction, the importance of dam safety monitoring has become increasingly prominent. Traditional dam deformation monitoring mainly relies on a single monitoring method, such as using displacement sensors or visual detection equipment alone for monitoring. These monitoring methods often have problems such as limited data acquisition accuracy and insufficient monitoring coverage.

[0003] Existing technologies usually use fixed monitoring equipment for dam deformation monitoring, and analyze deformation data through simple data processing methods. A single monitoring method is difficult to comprehensively reflect the deformation state of the dam, and various monitoring data are often processed separately, resulting in low correlation of monitoring information. Due to the lack of intelligent data fusion capabilities, it is difficult to effectively integrate monitoring data from different sources, affecting the accuracy of monitoring results. The monitoring and early warning mechanism is relatively fixed and cannot be dynamically optimized and adjusted according to the actual monitoring effect, resulting in insufficient timeliness and accuracy of early warning. There is a lack of effective fusion methods for multi-source heterogeneous monitoring data, and it is impossible to realize the collaborative analysis of displacement monitoring data and visual monitoring data, making it difficult to accurately grasp the overall deformation trend of the dam, thus affecting the reliability of monitoring and early warning. Summary of the Invention

[0004] In view of this, the present invention proposes a dam deformation monitoring and early warning method and system based on Beidou and machine vision, which solves the problem that the existing technology lacks effective fusion methods for multi-source heterogeneous monitoring data and cannot realize the collaborative analysis of displacement monitoring data and visual monitoring data.

[0005] The technical solution of the present invention is realized as follows: In the first aspect, the present invention provides a dam deformation monitoring and early warning method based on Beidou and machine vision, including the following steps:

[0006] Real-time collect the original displacement data of the dam through a Beidou receiver, and collect the original image data on the surface of the dam by using monitoring equipment;

[0007] Perform RTK differential processing and weighted average of multi-point positioning data on the original displacement data to obtain dam displacement data, and perform multi-spectral analysis and deep learning processing on the original image data to obtain dam surface structure feature data;

[0008] Perform localization processing on the dam displacement data by using the edge computing method to obtain real-time displacement change data, and perform parallel processing on the dam surface structure feature data by using the distributed computing method to obtain real-time deformation feature data;

[0009] Dynamically fuse the real-time displacement change data and the real-time deformation feature data based on the Kalman filtering algorithm to obtain comprehensive deformation data;

[0010] Establish a dam deformation trend prediction model according to the comprehensive deformation data, set multi-level early warning thresholds, and when abnormal deformation is detected, push hierarchical early warning information to the monitoring center.

[0011] Based on the above technical solutions, preferably, the original displacement data of the dam is collected in real time through a Beidou receiver, and the original image data on the surface of the dam is collected by monitoring equipment, specifically including:

[0012] Deploy multiple Beidou receivers in the monitoring area of the dam to construct a Beidou monitoring network. The Beidou monitoring network includes a reference station and multiple monitoring stations. The reference station is set in a stable area outside the deformation influence range of the dam, and the multiple monitoring stations are respectively set at multiple key parts of the dam. The multiple key parts include the dam crest, the dam body, and the dam foundation. The three-dimensional coordinate data of the dam is collected in real time through the Beidou monitoring network as the original displacement data;

[0013] Deploy multiple cameras around the dam to construct an all-round monitoring network, and collect multi-angle images on the surface of the dam as the original image data;

[0014] The multiple cameras include visible light cameras and infrared cameras. The visible light cameras are used to collect images on the surface of the dam during the day, and the real-time infrared cameras are used to collect thermal imaging images on the surface of the dam at night and under low light conditions.

[0015] Based on the above technical solutions, preferably, perform RTK differential processing and weighted average of multi-point positioning data on the original displacement data to obtain dam displacement data, and perform multi-spectral analysis and deep learning processing on the original image data to obtain dam surface structure feature data, specifically including:

[0016] Perform RTK differential processing on the original displacement data using a dynamic baseline adjustment algorithm to obtain optimized displacement data, and use a weighted average algorithm to fuse the optimized displacement data to obtain dam displacement data;

[0017] Integrate infrared image data and visible light image data through a multi-spectral image fusion algorithm to obtain a fused multi-spectral image, and train and identify the fused multi-spectral image based on a convolutional neural network model to obtain the dam surface structure feature data.

[0018] Based on the above technical solutions, preferably, the displacement data of the dam is locally processed by the edge computing method to obtain real-time displacement change data, and the surface structure feature data of the dam is processed in parallel by the distributed computing method to obtain real-time deformation feature data, specifically including:

[0019] Adopt a dynamic data screening method based on the model predictive control algorithm to filter and optimize the dam displacement data in real time to obtain optimized dam displacement data. Use the adaptive threshold detection algorithm to perform real-time anomaly detection on the optimized dam displacement data to obtain the real-time displacement change data;

[0020] Adopt the region growing algorithm to extract features from the surface structure feature data of the dam to obtain feature region change data, and use the spatio-temporal correlation analysis method based on the atlas neural network to comprehensively analyze the feature region change data to generate real-time deformation feature data.

[0021] Based on the above technical solutions, preferably, the calculation formula of the optimized dam displacement data is:

[0022] D 优 (t) = α2·D 实 (t) + (1 - α2)·D 预 (t);

[0023] Where, D 优 (t) is the optimized dam displacement data at time t, D 实 (t) is the dam displacement data at time t, D 预 (t) is the predicted displacement data at time t based on historical data, and α2 is the dynamic data adjustment coefficient;

[0024] The calculation formula of the real-time displacement change data is:

[0025]

[0026] Δd threshold = μ1 + β2·σ1;

[0027] Where, D 变 (t) is the real-time displacement change data at time t, D 优 (t) is the optimized dam displacement data at time t, D 优 (t - 1) is the optimized dam displacement data at time t - 1, Δd threshold is the adaptive threshold, μ1 is the moving average value of the optimized dam displacement data, σ1 is the moving standard deviation of the optimized dam displacement data, and β2 is the sensitivity adjustment coefficient;

[0028] The calculation formula of the feature region change data is:

[0029] S(p) = w1·ΔI(p) + w2·ΔT(p) + w3·C(p);

[0030]

[0031] Wherein, S(p) is the growth score of pixel point p in the surface structure feature data of the dam, ΔI(p) is the grayscale difference of pixel point p, ΔT(p) is the texture feature difference of pixel point p, C(p) is the spatial continuity constraint of pixel point p, w1, w2, and w3 are the weight coefficients of grayscale, texture feature, and spatial continuity constraint respectively, ΔF 区域 is the feature area change data, and R is the set of pixel points in the feature area;

[0032] The calculation formula of the real-time deformation feature data is:

[0033]

[0034] Wherein, FeatureFusion i is the fusion feature matrix of node i, N(i) is the set of nodes adjacent to node i, W j is the weight matrix of node j, Feature j is the feature matrix of node j, b2 is the neighbor node bias term, ReLU(·) is the ReLU activation function, F 形变 is the real-time deformation feature data, and N1 is the total number of feature nodes.

[0035] Based on the above technical solutions, preferably, the Kalman filter algorithm is used to dynamically fuse the real-time displacement change data and the real-time deformation feature data to obtain comprehensive deformation data, specifically including:

[0036] The Kalman filter algorithm is used to perform dynamic optimization processing on the real-time displacement change data and the real-time deformation feature data respectively, including: constructing a state space model, mapping the real-time displacement change data and the real-time deformation feature data into the state vector space respectively, and through the prediction-update iteration process, dynamically optimizing the state vector. The prediction step is based on historical data for state prediction, and the update iteration step combines the observed data to correct the prediction result, introducing an adaptive factor, and dynamically adjusting the system noise covariance matrix and the observed noise covariance matrix according to the time-varying characteristics of the data to obtain the optimized displacement change data and the optimized deformation feature data;

[0037] The calculation formula of the Kalman filter algorithm is:

[0038]

[0039] Wherein, is the state estimate value at time m, is the state estimation value at time m-1, and α m is the adaptive factor, and z m is the observation value at time m, and u m is the control input, A1 is the state transition matrix, B1 is the control matrix, K1 is the Kalman gain, and v m is the innovation sequence, and R m is the observation noise covariance matrix, and P m is the prediction error covariance matrix, and tr(·) is the trace of the matrix;

[0040] A sliding time window is established based on the data timestamp, and the optimized displacement change data and optimized deformation feature data are time-aligned within the window. The correlation coefficient and confidence level of the time-aligned data are calculated as the weight parameters for dynamic weighted fusion. A multi-scale weighted fusion method is adopted to combine local and global features to generate comprehensive deformation data;

[0041] The calculation formula for the time alignment is:

[0042] Δt sync = min(Δt pos , Δt def )·exp(-β3Δt pos -Δt def |);

[0043] where, Δt sync is the time synchronization window size, Δt pos is the displacement data time interval, Δt def is the deformation data time interval, and β3 is the time decay coefficient;

[0044] The calculation formula for the dynamic weighted fusion is:

[0045]

[0046] where, F 综 is the comprehensive deformation data, is the optimized displacement change data, is the optimized deformation feature data, and are respectively and 's weight coefficients, is the dynamic weight of the nth data source, is the correlation coefficient between the nth data source and the historical data, is the correlation coefficient between the qth data source and the historical data, is the data confidence level of the nth data source, is the data confidence of the q-th data source, where q = 1 and 2 are the optimized displacement change data and the optimized deformation feature data respectively.

[0047] Based on the above technical solutions, preferably, the dam deformation trend prediction model is established according to the comprehensive deformation data, and multiple warning thresholds are set. When abnormal deformation is detected, hierarchical warning information is pushed to the monitoring center, specifically including:

[0048] Construct a dam deformation trend prediction model based on the comprehensive deformation data, use the time series analysis method to predict the trend of the dam deformation data, and generate a future deformation trend curve;

[0049] Perform time series decomposition on the comprehensive deformation data, decompose the comprehensive deformation data into a trend term, a periodic term, and a random term. Based on the trend term, the periodic term, and the random term, use a long short-term memory network model for trend prediction, evaluate the error of the prediction result, and use a sliding window method to correct the prediction error to generate a future deformation trend curve;

[0050] According to the historical monitoring data and the dam structure characteristics, set multiple warning thresholds. The multiple warnings include normal, safety warning, risk warning, and emergency warning. Calculate the deviation between the predicted trend curve and the warning threshold. When any point on the predicted trend curve exceeds the safety warning threshold, trigger the corresponding level of warning information, and use the preset communication network to push the hierarchical warning information to the monitoring center, and attach a detailed deformation trend report according to the warning level.

[0051] In a second aspect, the present invention also provides a dam deformation monitoring and warning system based on Beidou and machine vision. The system includes:

[0052] A data acquisition module for real-time collecting the original displacement data of the dam through a Beidou receiver, and collecting the original image data on the surface of the dam by using monitoring equipment;

[0053] A data processing module for performing RTK differential processing and weighted averaging of multi-point positioning data on the original displacement data to obtain the dam displacement data, and performing multi-spectral analysis and deep learning processing on the original image data to obtain the dam surface structure feature data;

[0054] A deformation analysis module for locally processing the dam displacement data by using an edge computing algorithm to obtain real-time displacement change data, and performing parallel processing on the dam surface structure feature data by using a distributed computing method to obtain real-time deformation feature data;

[0055] A data fusion module for dynamically fusing the real-time displacement change data and the real-time deformation feature data based on the Kalman filter algorithm to obtain comprehensive deformation data;

[0056] The monitoring and early warning module is used to establish a dam deformation trend prediction model based on the comprehensive deformation data, set multiple levels of early warning thresholds, and push hierarchical early warning information to the monitoring center when abnormal deformation is detected.

[0057] In a third aspect, the present invention also provides an electronic device, including: at least one processor, at least one memory, a communication interface, and a bus;

[0058] Wherein, the processor, the memory, and the communication interface complete mutual communication through the bus, the memory stores program instructions executable by the processor, and the processor calls the program instructions to implement the steps of a dam deformation monitoring and early warning method based on Beidou and machine vision.

[0059] In a fourth aspect, the present invention also provides a computer-readable storage medium, and the computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the steps of a dam deformation monitoring and early warning method based on Beidou and machine vision.

[0060] The dam deformation monitoring and early warning method and system based on Beidou and machine vision of the present invention have the following beneficial effects compared with the prior art:

[0061] (1) By combining Beidou positioning and machine vision, real-time processing of dam monitoring data is carried out by using RTK differential processing, deep learning processing, edge computing, and distributed computing, and the dynamic fusion of multi-source data is realized by using the Kalman filter algorithm. A prediction model is established based on the fused comprehensive deformation data and hierarchical early warning is realized, which significantly improves the real-time performance and early warning reliability of dam deformation monitoring;

[0062] (2) By optimizing the dam displacement data and real-time displacement change data, combining historical data prediction, adaptive threshold, and sliding statistics methods, dynamically adjusting the data processing parameters, effectively reducing noise interference and random errors, and improving the sensitivity of displacement change detection, real-time monitoring of dam deformation is realized;

[0063] (3) By dynamically optimizing the real-time displacement change data and real-time deformation feature data through the Kalman filter algorithm, combining the state space model and the prediction-update iteration process, eliminating noise interference, dynamically adjusting the system noise and observation noise covariance matrices, and improving the stability of data optimization, real-time dynamic analysis of dam deformation is realized. Description of the Drawings

[0064] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0065] Figure 1 Flowchart of a dam deformation monitoring and early warning method based on Beidou and machine vision of the present invention;

[0066] Figure 2 Structural diagram of a dam deformation monitoring and early warning system based on Beidou and machine vision of the present invention. Specific embodiments

[0067] The following will combine 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 some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0068] Please refer to Figure 1 , the present invention provides a dam deformation monitoring and early warning method based on Beidou and machine vision, including the following steps:

[0069] Real-time acquisition of the original displacement data of the dam through Beidou receivers deployed at key parts of the dam, and acquisition of the original image data on the surface of the dam using monitoring equipment;

[0070] Perform RTK differential processing and weighted averaging of multi-point positioning data on the original displacement data to obtain the dam displacement data, and perform multi-spectral analysis and deep learning processing on the original image data to obtain the dam surface structure feature data;

[0071] Perform localization processing on the dam displacement data using the edge computing method to obtain real-time displacement change data, and perform parallel processing on the dam surface structure feature data through the distributed computing method to obtain real-time deformation feature data;

[0072] Based on the Kalman filter algorithm, dynamically fuse the real-time displacement change data and the real-time deformation feature data, establish a unified time synchronization mechanism to ensure the timeliness of data fusion, and obtain comprehensive deformation data;

[0073] Establish a dam deformation trend prediction model based on the comprehensive deformation data, set multiple levels of early warning thresholds, and when abnormal deformation is detected, push hierarchical early warning information to the monitoring center through a preset communication network.

[0074] Specifically, in this embodiment, by combining Beidou positioning and machine vision, real-time processing of dam monitoring data is carried out using RTK differential processing, deep learning processing, edge computing and distributed computing, and the Kalman filtering algorithm is used to realize the dynamic fusion of multi-source data. A prediction model is established based on the fused comprehensive deformation data and hierarchical early warning is realized, which significantly improves the real-time performance of dam deformation monitoring and the reliability of early warning.

[0075] The original displacement data of the dam is collected in real time through a Beidou receiver, and the original image data on the surface of the dam is collected by monitoring equipment, specifically including:

[0076] A plurality of Beidou receivers are arranged in the monitoring area of the dam to construct a Beidou monitoring network. The Beidou monitoring network includes a reference station and a plurality of monitoring stations. The reference station is set in a stable area outside the deformation influence range of the dam, and the plurality of monitoring stations are respectively set at a plurality of key parts of the dam. The plurality of key parts include the dam crest, the dam body and the dam foundation. High-precision positioning monitoring is realized through the cooperation of the reference station and the monitoring stations, and the three-dimensional coordinate data of the dam is collected in real time through the Beidou monitoring network as the original displacement data;

[0077] A plurality of cameras are arranged around the dam to construct an all-round monitoring network, and multi-angle images on the surface of the dam are collected as the original image data;

[0078] The plurality of cameras include visible light cameras and infrared cameras. The visible light cameras are used to collect images on the surface of the dam during the day, and the real-time infrared cameras are used to collect thermal imaging images on the surface of the dam at night and under low light conditions to realize all-weather monitoring.

[0079] Specifically, in this embodiment, by arranging a plurality of Beidou receivers at the key parts of the dam, including the dam crest, the dam body and the dam foundation, and combining the cooperation of the reference station and the monitoring stations, high-precision three-dimensional coordinate data collection is realized, which improves the collection accuracy of the dam displacement data. Especially under complex terrain conditions, stable and reliable displacement monitoring data can be provided.

[0080] A plurality of cameras, including visible light cameras and infrared cameras, are arranged around the dam to construct an all-weather and multi-angle monitoring network. The visible light cameras are used to collect images on the surface of the dam during the day, and the infrared cameras are used to collect thermal imaging images at night or under low light conditions, so as to realize all-weather monitoring ability. This multi-modal image collection method can comprehensively cover the surface of the dam and ensure the integrity of the monitoring data.

[0081] The introduction of infrared cameras enables normal operation at night or under low light conditions, overcomes the limitations of traditional monitoring equipment under insufficient light conditions, and significantly enhances the adaptability in complex environments.

[0082] The RTK differential processing is performed on the original displacement data and the weighted average of the multi-point positioning data is carried out to obtain the dam displacement data. The multi-spectral analysis and deep learning processing are performed on the original image data to obtain the dam surface structure feature data, specifically including:

[0083] The dynamic baseline adjustment algorithm is used to perform RTK differential processing on the original displacement data to improve the positioning accuracy under complex terrain conditions, and the optimized displacement data is obtained. The weighted average algorithm is used to fuse the optimized displacement data to reduce the random error of single-point data, and the dam displacement data is obtained;

[0084] The calculation formula of the dynamic baseline adjustment algorithm is:

[0085] B new = B current + α1·(D observed - D predicted );

[0086] Among them, B new is the new baseline value, B current is the current baseline value, α1 is the adjustment coefficient, D observed is the observed displacement data, D predicted is the predicted displacement data;

[0087] Through the multi-spectral image fusion algorithm, the infrared image data and the visible light image data are integrated to obtain the fused multi-spectral image. Based on the convolutional neural network model, the fused multi-spectral image is trained and recognized to achieve high-precision detection of the cracks and micro-deformations on the dam surface, and the dam surface structure feature data is obtained;

[0088] The calculation formula of the convolutional neural network model is:

[0089]

[0090] Among them, FeatureMap l+1 is the feature map of the (l + 1)-th layer, FeatureMap l is the feature map of the l-th layer, σ(·) is the activation function, W k is the k-th convolutional kernel, * is the convolution operation, b1 is the bias of the convolutional layer, β1 is the attention adjustment coefficient, and AttentionMap l is the attention map of the l-th layer.

[0091] Specifically, in this embodiment, RTK differential processing is performed through a dynamic baseline adjustment algorithm. According to the difference between real-time observation data and predicted data, the baseline value is dynamically adjusted, improving the positioning accuracy under complex terrain conditions. The weighted average algorithm is used to fuse the optimized displacement data, effectively reducing the random error of single-point data.

[0092] Through the multi-spectral image fusion algorithm, the infrared image and the visible light image are effectively integrated, making full use of the advantages of the two imaging methods, improving the information richness of the image, and enabling stable monitoring effects under different lighting and weather conditions.

[0093] Based on an improved convolutional neural network model, an attention mechanism is introduced for feature extraction and recognition, improving the detection accuracy of tiny cracks and deformations on the dam surface. The introduction of the attention mechanism enables the model to adaptively focus on important feature regions, improving the accuracy of recognition. By combining traditional image processing methods with deep learning, the intelligent processing of monitoring data is realized, greatly improving the ability to extract and analyze the structural characteristics of the dam surface.

[0094] The local processing of the dam displacement data by the edge computing method obtains real-time displacement change data, and the parallel processing of the dam surface structure feature data by the distributed computing method obtains real-time deformation feature data, specifically including:

[0095] A dynamic data screening method based on the model predictive control algorithm is used to filter and optimize the dam displacement data in real time, eliminating the influence of environmental noise on the displacement data, obtaining optimized dam displacement data. The adaptive threshold detection algorithm is used to perform real-time anomaly detection on the optimized dam displacement data, timely detecting and responding to the tiny displacement changes of the dam, and obtaining the real-time displacement change data;

[0096] The region growing algorithm is used to extract the features of the dam surface structure feature data, accurately identifying the feature regions such as cracks and deformations on the dam surface, obtaining the feature region change data. The spatio-temporal correlation analysis method based on the atlas neural network is used to comprehensively analyze the feature region change data, generating real-time deformation feature data.

[0097] Specifically, in this embodiment, a dynamic data screening method based on a model predictive control algorithm is used to filter and optimize the dam displacement data in real time, eliminate the interference of environmental noise, and adopt an adaptive threshold detection algorithm to dynamically adjust the detection threshold according to the actual monitoring data, so as to timely detect and respond to the small displacement changes of the dam, improve the sensitivity of anomaly detection, and extract the surface structure feature data of the dam through a region growing algorithm, which can accurately identify the feature regions such as cracks and deformations on the dam surface. By using a spatio-temporal correlation analysis method based on a graph neural network, the change data of the feature region can be comprehensively analyzed, realizing the intelligent identification and correlation analysis of deformation features.

[0098] The calculation formula for optimizing the dam displacement data is as follows:

[0099] D 优 (t) = α2·D 实 (t) + (1 - α2)·D 预 (t);

[0100] Where, D 优 (t) is the optimized dam displacement data at time t, D 实 (t) is the dam displacement data at time t, D 预 (t) is the predicted displacement data at time t based on historical data, and α2 is the dynamic data adjustment coefficient;

[0101] The calculation formula for the real-time displacement change data is as follows:

[0102]

[0103] Δd threshold = μ1 + β2·σ1;

[0104] Where, D 变 (t) is the real-time displacement change data at time t, D 优 (t) is the optimized dam displacement data at time t, D 优 (t - 1) is the optimized dam displacement data at time t - 1, Δd threshold is the adaptive threshold, μ1 is the moving average value of the optimized dam displacement data, σ1 is the moving standard deviation of the optimized dam displacement data, and β2 is the sensitivity adjustment coefficient;

[0105] The calculation formula for the change data of the feature region is as follows:

[0106] S(p) = w1·ΔI(p) + w2·ΔT(p) + w3·C(p);

[0107]

[0108] Among them, S(p) is the growth score of pixel point p in the dam surface structure feature data, ΔI(p) is the gray-scale difference of pixel point p, ΔT(p) is the texture feature difference of pixel point p, C(p) is the spatial continuity constraint of pixel point p, w1, w2, and w3 are the weight coefficients of gray-scale, texture feature, and spatial continuity constraint respectively, and ΔF 区域 is the feature region change data, and R is the set of pixel points in the feature region;

[0109] The calculation formula of the real-time deformation feature data is:

[0110]

[0111] Among them, FeatureFusion i is the fusion feature matrix of node i, N(i) is the set of nodes adjacent to node i, and W j is the weight matrix of node j, Feature j is the feature matrix of node j, b2 is the neighbor node bias term, ReLU(·) is the ReLU activation function, and F 形变 is the real-time deformation feature data, and N1 is the total number of feature nodes.

[0112] Specifically, in this embodiment, by optimizing the dam displacement data, combining historical data prediction and dynamically adjusting coefficients, the original displacement data is optimized, effectively improving the reliability of the data. By introducing prediction data as a reference, the optimization process is made more forward-looking.

[0113] The calculation of the real-time displacement change data uses an adaptive threshold method. Through the dynamic calculation of the sliding average and standard deviation, the detection sensitivity is automatically adjusted according to the data characteristics, improving the detection ability for small displacement changes. The calculation of the feature region change data comprehensively considers three key factors: gray-scale difference, texture feature, and spatial continuity constraint. Through the reasonable configuration of weight coefficients, the accurate identification and change analysis of the dam surface feature region are realized. The calculation of the real-time deformation feature data uses a graph-structure-based feature fusion method. Through the weight matrix between nodes and the ReLU activation function, the intelligent extraction and analysis of features are realized.

[0114] Performing dynamic fusion on the real-time displacement change data and the real-time deformation feature data based on the Kalman filter algorithm to obtain comprehensive deformation data, specifically including:

[0115] The Kalman filtering algorithm is used to perform dynamic optimization processing on the real-time displacement change data and the real-time deformation feature data respectively, including: constructing a state space model, mapping the real-time displacement change data and the real-time deformation feature data into the state vector space respectively, and through the prediction-update iterative process, dynamically optimizing the state vector. The prediction step performs state prediction based on historical data, and the update iterative step corrects the prediction result by combining the observation data. An adaptive factor is introduced to dynamically adjust the system noise covariance matrix and the observation noise covariance matrix according to the time-varying characteristics of the data, improving the optimization effect, and obtaining the optimized displacement change data and the optimized deformation feature data;

[0116] The calculation formula of the Kalman filtering algorithm is:

[0117]

[0118] where, is the state estimate value at time m, is the state estimate value at time m-1, α m is the adaptive factor, z m is the observation value at time m, u m is the control input, A1 is the state transition matrix, B1 is the control matrix, K1 is the Kalman gain, v m is the innovation sequence, R m is the observation noise covariance matrix, P m is the prediction error covariance matrix, tr(·) is the trace of the matrix;

[0119] A sliding time window is established based on the data timestamp, and the optimized displacement change data and the optimized deformation feature data are time-aligned within the window. The correlation coefficient and confidence level of the time-aligned data are calculated as the weight parameters for dynamic weighted fusion. The multi-scale weighted fusion method is used to generate the comprehensive deformation data by combining local and global features;

[0120] The calculation formula of the time alignment is:

[0121] Δt sync =min(Δt pos ,Δt def )·exp(-β3Δt pos -Δt def |);

[0122] where, Δt sync is the time synchronization window size, Δt pos is the time interval of the displacement data, Δt def is the time interval of the deformation data, β3 is the time decay coefficient;

[0123] The calculation formula of the dynamic weighted fusion is:

[0124]

[0125] Among them, F 综 is the comprehensive deformation data, is the optimized displacement change data, is the optimized deformation feature data, and are respectively and weight coefficients, is the dynamic weight of the nth data source, is the correlation coefficient between the nth data source and historical data, is the correlation coefficient between the qth data source and historical data, is the data confidence of the nth data source, is the data confidence of the qth data source, where q = 1, 2 are the optimized displacement change data and the optimized deformation feature data respectively.

[0126] Specifically, in this embodiment, the Kalman filter algorithm is used to perform dynamic optimization processing on the real-time displacement change data and the real-time deformation feature data. Combining the state space model and the prediction-update iteration process, data noise is eliminated and data quality is improved. An adaptive factor is introduced to dynamically adjust the system noise covariance matrix and the observation noise covariance matrix, making the optimization process more adaptable. By establishing a sliding time window mechanism based on the data timestamp, precise time alignment of data from different sources is achieved, solving the problem of asynchronous time of multi-source data. The time alignment calculation formula takes into account the time decay factor. A multi-scale weighted fusion method is adopted, and by calculating the data correlation coefficient and confidence as weight parameters, effective combination of local and global features is achieved.

[0127] The dam deformation trend prediction model is established according to the comprehensive deformation data, and multi-level warning thresholds are set. When abnormal deformation is detected, hierarchical warning information is pushed to the monitoring center, specifically including:

[0128] Based on the comprehensive deformation data, a dam deformation trend prediction model is constructed, and the time series analysis method is used to predict the trend of dam deformation data, generating a future deformation trend curve;

[0129] The comprehensive deformation data is decomposed by time series, and the comprehensive deformation data is decomposed into a trend term, a periodic term, and a random term. Based on the trend term, the periodic term, and the random term, a long short-term memory network model is used for trend prediction. The long short-term memory network model includes introducing an attention mechanism to improve the prediction accuracy; the prediction result is evaluated for error, and the sliding window method is used to correct the prediction error, generating a future deformation trend curve;

[0130] The calculation formula for the time series decomposition is as follows:

[0131]

[0132] Among them, F 综 is the comprehensive deformation data, is the trend term, reflecting the long-term change trend of the deformation data, is the periodic term, reflecting the periodic fluctuation of the deformation data, is the random term, reflecting the random fluctuation in the deformation data;

[0133] The calculation formula for the error evaluation is as follows:

[0134]

[0135] Among them, is the prediction error, is the true deformation data, is the predicted deformation data, is the predicted data after correction, N2 is the sliding window size, is the current moment, s is the time step index;

[0136] According to the historical monitoring data and the dam structure characteristics, multi-level warning thresholds are set. The multi-level warnings include normal, safety warning, risk warning, and emergency warning. Calculate the deviation between the predicted trend curve and the warning threshold. When any point on the predicted trend curve exceeds the safety warning threshold, trigger the warning information of the corresponding level, and use the preset communication network to push the classified warning information to the monitoring center, and attach a detailed deformation trend report according to the warning level;

[0137] The judgment formula for multi-level warning is as follows:

[0138]

[0139] Among them, is the deviation between the predicted value and the warning threshold, is the predicted value at the moment of, is the warning threshold at the moment of, Δ1 is the first warning threshold, and Δ2 is the second warning threshold.

[0140] Specifically, in this embodiment, the comprehensive deformation data is decomposed into a trend term, a periodic term, and a random term through time series decomposition. The long short-term memory network model is adopted and an attention mechanism is introduced to improve the ability to capture long-term dependence relationships.

[0141] The prediction results are corrected in real time by the sliding window method, effectively reducing the prediction error. Through error assessment calculation, the dynamic optimization of the prediction results is realized, and the stability of the deformation trend prediction is improved.

[0142] Based on historical monitoring data and the dam structure characteristics, a multi-level early warning mechanism including normal, safety warning, risk warning and emergency warning is established. Through the setting of early warning thresholds, the accurate identification of deformation risks at different levels is realized, so that corresponding-level early warning information can be sent out in time, and a detailed deformation trend report is automatically generated. The early warning information includes a detailed analysis report of the deformation trend, providing comprehensive decision-making basis for management personnel. Through the preset communication network, the timely push of early warning information is realized, ensuring the timeliness of information transmission.

[0143] Please refer to Figure 2 , the present invention also provides a dam deformation monitoring and early warning system based on Beidou and machine vision, and the system includes:

[0144] A data acquisition module, configured to collect the original displacement data of the dam in real time through a Beidou receiver, and collect the original image data on the surface of the dam by using monitoring equipment;

[0145] A data processing module, configured to perform RTK differential processing and weighted average of multi-point positioning data on the original displacement data to obtain dam displacement data, and perform multi-spectral analysis and deep learning processing on the original image data to obtain dam surface structure feature data;

[0146] A deformation analysis module, configured to perform local processing on the dam displacement data by using an edge computing algorithm to obtain real-time displacement change data, and perform parallel processing on the dam surface structure feature data by using a distributed computing method to obtain real-time deformation feature data;

[0147] A data fusion module, configured to perform dynamic fusion on the real-time displacement change data and the real-time deformation feature data based on the Kalman filtering algorithm to obtain comprehensive deformation data;

[0148] A monitoring and early warning module, configured to establish a dam deformation trend prediction model according to the comprehensive deformation data, set multi-level early warning thresholds, and push hierarchical early warning information to the monitoring center when abnormal deformation is detected.

[0149] Specifically, a dam deformation monitoring and early warning system based on Beidou and machine vision in this embodiment realizes the full-process intelligent processing of dam deformation monitoring through the collaborative work of five functional modules: the data acquisition module realizes all-weather and multi-angle data acquisition through Beidou receivers and monitoring devices; the data processing module uses RTK differential and deep learning technologies to improve data quality; the deformation analysis module uses edge computing and distributed computing to achieve real-time processing; the data fusion module realizes the dynamic fusion of multi-source data based on the Kalman filter algorithm; the monitoring and early warning module timely discovers potential risks through trend prediction and multi-level early warning mechanisms, improving the real-time performance and reliability of deformation monitoring.

[0150] The present invention also discloses an electronic device, including: at least one processor, at least one memory, a communication interface, and a bus: wherein, the processor, the memory, and the communication interface complete mutual communication through the bus; the memory stores program instructions executable by the processor, and the processor calls the program instructions to implement a method for dam deformation monitoring and early warning based on Beidou and machine vision.

[0151] The present invention also discloses a computer-readable storage medium, and the computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to implement all or part of the steps of the method for dam deformation monitoring and early warning based on Beidou and machine vision in the embodiment of the present invention. The storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory ROM, random access memory RAM, magnetic disks, or optical discs that can store program codes.

[0152] The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A dam deformation monitoring and early warning method based on Beidou and machine vision, characterized in that, It includes the following steps: Real-time collect the original displacement data of the dam through a Beidou receiver, and collect the original image data on the surface of the dam by using monitoring equipment; Perform RTK differential processing and weighted average of multi-point positioning data on the original displacement data to obtain the dam displacement data, and perform multi-spectral analysis and deep learning processing on the original image data to obtain the dam surface structure feature data; Perform local processing on the dam displacement data by using the edge computing method to obtain real-time displacement change data, and perform parallel processing on the dam surface structure feature data by using the distributed computing method to obtain real-time deformation feature data; Based on the Kalman filtering algorithm, dynamically fuse the real-time displacement change data and the real-time deformation feature data to obtain comprehensive deformation data; Establish a dam deformation trend prediction model according to the comprehensive deformation data, set multi-level warning thresholds, and when abnormal deformation is detected, push hierarchical warning information to the monitoring center.

2. The dam deformation monitoring and early warning method based on Beidou and machine vision according to claim 1, characterized in that The step of real-time collecting the original displacement data of the dam through a Beidou receiver and collecting the original image data on the surface of the dam by using monitoring equipment specifically includes: Deploy multiple Beidou receivers in the monitoring area of the dam to construct a Beidou monitoring network. The Beidou monitoring network includes a reference station and multiple monitoring stations. The reference station is set in a stable area outside the dam deformation influence range, and the multiple monitoring stations are respectively set at multiple key parts of the dam. The multiple key parts include the dam crest, the dam body, and the dam foundation. Collect the three-dimensional coordinate data of the dam as the original displacement data through the Beidou monitoring network; Deploy multiple cameras around the dam to construct an all-round monitoring network, and collect multi-angle images on the surface of the dam as the original image data; The multiple cameras include visible light cameras and infrared cameras. The visible light cameras are used to collect images on the surface of the dam during the day, and the real-time infrared cameras are used to collect thermal imaging images on the surface of the dam at night and under low light conditions.

3. The dam deformation monitoring and early warning method based on Beidou and machine vision according to claim 2, characterized in that The step of performing RTK differential processing and weighted average of multi-point positioning data on the original displacement data to obtain the dam displacement data, and performing multi-spectral analysis and deep learning processing on the original image data to obtain the dam surface structure feature data specifically includes: Perform RTK differential processing on the original displacement data by using the dynamic baseline adjustment algorithm to obtain optimized displacement data, and fuse the optimized displacement data by using the weighted average algorithm to obtain the dam displacement data; Integrate the infrared image data and the visible light image data through the multi-spectral image fusion algorithm to obtain a fused multi-spectral image, and train and identify the fused multi-spectral image based on a convolutional neural network model to obtain the dam surface structure feature data.

4. The dam deformation monitoring and early warning method based on Beidou and machine vision according to claim 1, characterized in that, ​ A dynamic data screening method based on a model predictive control algorithm is adopted to perform real-time filtering and optimization on the dam displacement data to obtain optimized dam displacement data. An adaptive threshold detection algorithm is used to perform real-time anomaly detection on the optimized dam displacement data to obtain the real-time displacement change data; A region growing algorithm is adopted to extract features from the dam surface structure feature data to obtain feature region change data. A spatio-temporal correlation analysis method based on an atlas neural network is used to comprehensively analyze the feature region change data to generate real-time deformation feature data.

5. The dam deformation monitoring and early warning method based on Beidou and machine vision according to claim 4, characterized in that, The calculation formula for the optimized dam displacement data is: D 优 D(t) = α2·D 实 (t) + (1 - α2)·D 预 (t); Among them, D 优 (t) is the optimized dam displacement data at time t, D 实 (t) is the dam displacement data at time t, D 预 (t) is the displacement data at time t predicted based on historical data, and α2 is the dynamic data adjustment coefficient; The calculation formula for the real-time displacement change data is: Δd threshold = μ1 + β2·σ1; Among them, D 变 (t) is the real-time displacement change data at time t, D 优 (t) is the optimized dam displacement data at time t, D 优 (t - 1) is the optimized dam displacement data at time t - 1, Δd threshold is the adaptive threshold, μ1 is the moving average of the optimized dam displacement data, σ1 is the moving standard deviation of the optimized dam displacement data, and β2 is the sensitivity adjustment coefficient; The calculation formula for the feature region change data is: S(p) = w1·ΔI(p) + w2·ΔT(p) + w3·C(p); Among them, S(p) is the growth score of pixel point p in the dam surface structure feature data, ΔI(p) is the gray-scale difference of pixel point p, ΔT(p) is the texture feature difference of pixel point p, C(p) is the spatial continuity constraint of pixel point p, w1, w2, and w3 are the weight coefficients of gray-scale, texture feature, and spatial continuity constraint respectively, and ΔF 区域 is the feature region change data, and R is the set of pixel points in the feature region; The calculation formula for the real-time deformation feature data is: Among them, FeatureFusion i is the fusion feature matrix of node i, N(i) is the set of nodes adjacent to node i, and W j is the weight matrix of node j, Feature j is the feature matrix of node j, b2 is the bias term of neighbor nodes, ReLU(·) is the ReLU activation function, and F 形变 is the real-time deformation feature data, and N1 is the total number of feature nodes.

6. The dam deformation monitoring and early warning method based on Beidou and machine vision according to claim 1, characterized in that Based on the Kalman filter algorithm, the real-time displacement change data and the real-time deformation feature data are dynamically fused to obtain comprehensive deformation data, which specifically includes: The Kalman filter algorithm is used to perform dynamic optimization processing on the real-time displacement change data and the real-time deformation feature data respectively, including: constructing a state space model, mapping the real-time displacement change data and the real-time deformation feature data into the state vector space respectively, and through the prediction-update iteration process, dynamically optimizing the state vector. The prediction step performs state prediction based on historical data, and the update iteration step corrects the prediction result by combining the observation data. An adaptive factor is introduced to dynamically adjust the system noise covariance matrix and the observation noise covariance matrix according to the time-varying characteristics of the data to obtain optimized displacement change data and optimized deformation feature data; The calculation formula for the Kalman filter algorithm is: Among them, is the state estimation value at time m, is the state estimation value at time m - 1, α m is the adaptive factor, z m is the observation value at time m, u m is the control input, A1 is the state transition matrix, B1 is the control matrix, K1 is the Kalman gain, v m is the innovation sequence, R m is the observation noise covariance matrix, P m is the prediction error covariance matrix, tr(·) is the trace of the matrix; A sliding time window is established based on the data timestamp. The optimized displacement change data and the optimized deformation feature data are time-aligned within the window, and the correlation coefficient and confidence level of the time-aligned data are calculated as the weight parameters for dynamic weighted fusion. A multi-scale weighted fusion method is adopted to combine local and global features to generate comprehensive deformation data; The calculation formula for the time alignment is: Δt sync = min(Δt pos , Δt def ) · exp(-β3Δt pos -Δt def |); where, Δt sync is the time synchronization window size, Δt pos is the displacement data time interval, Δt def is the deformation data time interval, and β3 is the time decay coefficient; The calculation formula for the dynamic weighted fusion is: Among them, F 综 is the comprehensive deformation data, is the optimized displacement change data, is the optimized deformation feature data, and are respectively and weight coefficients, is the dynamic weight of the nth data source, is the correlation coefficient between the nth data source and historical data, is the correlation coefficient between the qth data source and historical data, is the data confidence of the nth data source, is the data confidence of the qth data source, where q = 1, 2 are the optimized displacement change data and the optimized deformation feature data respectively.

7. The dam deformation monitoring and early warning method based on Beidou and machine vision according to claim 6, characterized in that, Based on the comprehensive deformation data, a dam deformation trend prediction model is established, and multi-level warning thresholds are set. When abnormal deformation is detected, hierarchical warning information is pushed to the monitoring center, which specifically includes: Based on the comprehensive deformation data, a dam deformation trend prediction model is constructed, and a time series analysis method is used to predict the trend of the dam deformation data to generate a future deformation trend curve; The comprehensive deformation data is decomposed by time series, and the comprehensive deformation data is decomposed into a trend term, a periodic term, and a random term. Based on the trend term, the periodic term, and the random term, a long short-term memory network model is used for trend prediction, and the prediction result is evaluated for error. The sliding window method is used to correct the prediction error to generate a future deformation trend curve; Based on historical monitoring data and the characteristics of the dam structure, multi-level warning thresholds are set. The multi-level warnings include normal, safety alert, risk warning, and emergency warning. The deviation between the predicted trend curve and the warning threshold is calculated. When any point on the predicted trend curve exceeds the safety alert threshold, the corresponding level of warning information is triggered. Using a preset communication network, the classified warning information is pushed to the monitoring center, and a detailed deformation trend report is attached according to the warning level.

8. A dam deformation monitoring and early warning system based on Beidou and machine vision, characterized in that, The system includes: A data acquisition module for collecting the original displacement data of the dam in real time through a Beidou receiver and collecting the original image data on the surface of the dam using monitoring equipment. A data processing module for performing RTK differential processing and weighted averaging of multi-point positioning data on the original displacement data to obtain the dam displacement data, and performing multi-spectral analysis and deep learning processing on the original image data to obtain the dam surface structure feature data. A deformation analysis module for locally processing the dam displacement data using an edge computing algorithm to obtain real-time displacement change data, and performing parallel processing on the dam surface structure feature data through a distributed computing method to obtain real-time deformation feature data. A data fusion module for dynamically fusing the real-time displacement change data and the real-time deformation feature data based on the Kalman filter algorithm to obtain comprehensive deformation data. A monitoring and warning module for establishing a dam deformation trend prediction model based on the comprehensive deformation data, setting multi-level warning thresholds, and pushing classified warning information to the monitoring center when abnormal deformation is detected.

9. An electronic device, characterized in that, It includes: At least one processor, at least one memory, a communication interface, and a bus; Wherein, the processor, the memory, and the communication interface complete communication with each other through the bus. The memory stores program instructions executable by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, and the computer instructions enable the computer to implement the method according to any one of claims 1 to 7.

Images