A GNSS dam seepage safety monitoring method and system based on Beidou positioning and image recognition

Abstract

The embodiment of the present invention provides a GNSS dam seepage safety monitoring method and system based on Beidou positioning and image recognition, which relates to the field of data processing technology. The method includes: obtaining continuous Beidou positioning data and GNSS satellite images of the dam area; determining the dam body change information of the dam area according to the continuous Beidou positioning data; using the GNSS satellite image as the input of the pre-trained deep learning model to obtain the vegetation feature information contained in the GNSS satellite image, and the vegetation feature information includes vegetation type characteristics and vegetation distribution characteristics; clustering analysis is performed on the vegetation type characteristics and vegetation distribution characteristics to determine the vegetation change trend; based on the dam body change information and the vegetation change trend, the dam seepage problem is determined. Through the present invention, the problem of low accuracy of dam seepage safety monitoring is solved, thereby achieving the effect of improving the accuracy and efficiency of dam seepage safety monitoring.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of data processing technology, and in particular to a GNSS dam seepage safety monitoring method and system based on Beidou positioning and image recognition. Background Art

[0002] As an important water conservancy project facility, dams play a vital role in flood control, irrigation, power generation, etc. However, the safety of dams has always been the focus of attention in the engineering community, among which the seepage problem is particularly critical because it may lead to erosion, deformation or even destruction of the dam structure, thus causing safety accidents.

[0003] At present, dam safety monitoring mainly relies on regular manual inspections combined with some instruments and equipment, such as piezometers and strain gauges. These methods have problems such as limited monitoring range, poor real-time performance, and low efficiency, and cannot achieve comprehensive and real-time monitoring of dam seepage conditions.

[0004] There is currently no better solution to the above problems. Summary of the invention

[0005] The embodiment of the present invention provides a GNSS dam seepage safety monitoring method and system based on Beidou positioning and image recognition, so as to at least solve the problem of low accuracy of dam seepage monitoring in related technologies.

[0006] According to one embodiment of the present invention, a GNSS dam seepage safety monitoring method based on Beidou positioning and image recognition is provided, comprising:

[0007] Acquire continuous Beidou positioning data and GNSS satellite images of the dam area, wherein the Beidou positioning data is obtained by continuously positioning and monitoring monitoring stations set up in the dam area, and the GNSS satellite images are obtained by collecting images of the dam area through the GNSS navigation satellite system;

[0008] Determine dam body change information of the dam area according to the continuous Beidou positioning data;

[0009] Using the GNSS satellite image as an input of a pre-trained deep learning model to obtain vegetation feature information contained in the GNSS satellite image, wherein the vegetation feature information includes vegetation type features and vegetation distribution features;

[0010] Performing cluster analysis on the vegetation type characteristics and the vegetation distribution characteristics to determine the vegetation change trend;

[0011] The seepage problem of the dam is determined based on the dam body change information and the vegetation change trend.

[0012] In an exemplary embodiment, the deep learning model is a Swin Transformer model, the Swin Transformer model includes a block partition component, a linear embedding component, a multi-scale feature fusion component, a Swin Transformer block component and a block merging component, the Swin Transformer model includes at least one feature layer, and the training steps of the Swin Transformer model include:

[0013] In response to receiving the training image set, looping through the first step until a preset number of training times is reached;

[0014] Wherein, the first step comprises:

[0015] Determining initial parameters of the Swin Transformer model;

[0016] Using the block partitioning component to partition each image in the training image set into image blocks of fixed size;

[0017] Using the linear embedding component, each of the image blocks is mapped to a high-dimensional feature space through a linear layer to obtain a feature vector of each image block;

[0018] The multi-scale feature fusion component is used to fuse the feature vectors of different scales to extract multi-scale features;

[0019] Inputting the multi-scale features into a plurality of Swin Transformer block components, and merging blocks between different feature layers;

[0020] The initial parameters of the Swin Transformer model are updated according to a preset optimization algorithm.

[0021] In an exemplary embodiment, the Swin Transformer model includes a primary feature layer, a mid-level feature layer, a high-level feature layer, and a global semantic feature layer.

[0022] In an exemplary embodiment, determining the dam body change information of the dam area according to the continuous Beidou positioning data includes:

[0023] Performing real-time dynamic differentiation on the continuous Beidou positioning data to obtain continuous Beidou positioning data after differential positioning;

[0024] Performing time series analysis on the continuous Beidou positioning data after the differential positioning to identify the changing trend of the dam body position in the dam area and the sudden change positioning data of each monitoring station;

[0025] Calculate the displacement of each monitoring site according to the continuous Beidou positioning data after differential positioning, wherein the displacement includes horizontal displacement and vertical displacement;

[0026] The dam body change information of each of the monitoring sites in the dam area is determined according to the dam body position change trend, the mutation positioning data and the displacement.

[0027] In an exemplary embodiment, the dam body position change trend includes a dam body horizontal displacement trend and a dam body vertical displacement trend, and determining the dam body change information of each monitoring station in the dam area according to the dam body position change trend, the mutation positioning data and the displacement amount includes:

[0028] Determine the change of the dam body according to the horizontal displacement trend of the dam body and the vertical displacement trend of the dam body, wherein the change of the dam body includes uniform settlement, local sliding, foundation problems, crack generation, elevation of the infiltration line and deformation of the dam body;

[0029] According to the mutation positioning data and the displacement of each monitoring site, m suspicious dam body change situations of each monitoring site are determined in the dam body change situation, where m is a positive integer;

[0030] Summarizing the change conditions of the suspected dam body at all the monitoring sites and constructing a three-dimensional model of the dam body, wherein the three-dimensional model of the dam body is used to simulate the change conditions of the suspected dam body at each of the monitoring sites;

[0031] The mutation location data and the displacement amount of each monitoring site are corrected according to the simulation results of the dam body three-dimensional model, and the final dam body change situation is determined from the m suspected dam body change situations according to the corrected mutation location data and the displacement amount of each monitoring site, and the final dam body change situation is used to characterize the dam body change information.

[0032] In an exemplary embodiment, determining the dam seepage problem according to the dam body change information and the vegetation change trend includes:

[0033] Performing correlation analysis on the dam body change information and the vegetation change trend to determine correlation characteristics between the dam body change information and the vegetation change trend;

[0034] The associated features are input into a dam seepage prediction model to obtain the dam seepage problem.

[0035] In an exemplary embodiment, the dam seepage prediction model includes a logistic regression module and a linear regression module, the dam seepage problem includes dam seepage probability and seepage severity, and the inputting of the associated features into the dam seepage prediction model to obtain the dam seepage problem includes:

[0036] The associated features are input into the dam seepage prediction model to obtain the dam seepage probability and the seepage severity, wherein the dam seepage probability is obtained based on the logistic regression module, and the seepage severity is obtained based on the linear regression module.

[0037] In an exemplary embodiment, after taking the GNSS satellite image as an input of the pre-trained deep learning model to obtain vegetation feature information contained in the GNSS satellite image, the method further includes:

[0038] Based on the vegetation feature information, constructing a vegetation feature matrix according to a preset matrix construction rule;

[0039] Performing correlation calculation on the vegetation feature matrix to obtain a feature correlation value;

[0040] When the feature correlation value does not meet the preset threshold range condition, it is determined that the vegetation feature information is abnormal.

[0041] According to another embodiment of the present invention, a GNSS dam seepage safety monitoring system based on Beidou positioning and image recognition is provided, comprising:

[0042] A data acquisition module is used to obtain continuous Beidou positioning data and GNSS satellite images of the dam area, wherein the Beidou positioning data is obtained by continuously positioning and monitoring the monitoring stations set up in the dam area, and the GNSS satellite images are obtained by collecting images of the dam area through the GNSS navigation satellite system;

[0043] A data processing module, used for determining the dam body change information of the dam area according to the continuous Beidou positioning data;

[0044] A feature recognition module, used to use the GNSS satellite image as an input of a pre-trained deep learning model to obtain vegetation feature information contained in the GNSS satellite image, wherein the vegetation feature information includes vegetation type features and vegetation distribution features;

[0045] A feature analysis module, used to perform cluster analysis on the vegetation type features and the vegetation distribution features to determine the vegetation change trend;

[0046] The decision module is used to determine the seepage problem of the dam according to the dam body change information and the vegetation change trend.

[0047] According to yet another embodiment of the present invention, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the steps of any one of the above method embodiments when run.

[0048] According to yet another embodiment of the present invention, there is provided an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0049] Through the present invention, since the vegetation in the dam area will be affected when the dam seepage occurs, the dam body change situation determined by Beidou positioning and the vegetation change trend determined by the image data collected by the GNSS satellite are comprehensively judged, and the seepage problem of the dam can be more comprehensively evaluated, more accurate early warning information can be provided, and the dam body seepage situation can be accurately identified effectively, and all-weather real-time safety monitoring of the dam area can be realized. At the same time, the GNSS satellite images are preprocessed and analyzed using a deep learning model, and the vegetation characteristics can be automatically identified and the change trend analysis can be performed, which greatly improves the monitoring efficiency. Therefore, by integrating the dam body change information and the vegetation change trend, a more comprehensive seepage monitoring method is provided, which helps to improve the safety and reliability of the dam, solve the problem of low accuracy of dam seepage monitoring, and achieve the effect of improving the accuracy and efficiency of dam seepage safety monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of a GNSS dam seepage safety monitoring method based on Beidou positioning and image recognition according to an embodiment of the present invention;

[0051] Figure 2 is a schematic diagram of the architecture of a deep learning model according to an embodiment of the present invention;

[0052] Figure 3 It is a structural block diagram of a GNSS dam seepage safety monitoring system based on Beidou positioning and image recognition according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings and in combination with the embodiments.

[0054] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0055] In this embodiment, a GNSS dam seepage safety monitoring method based on Beidou positioning and image recognition is provided. Figure 1 : is a flow chart of a GNSS dam seepage safety monitoring method based on Beidou positioning and image recognition according to an embodiment of the present invention, such as Figure 1 As shown, the process includes the following steps:

[0056] S110, obtaining continuous Beidou positioning data and GNSS satellite images of the dam area, wherein the Beidou positioning data is obtained by performing continuous positioning monitoring on monitoring stations set up in the dam area, and the GNSS satellite images are obtained by collecting images of the dam area through the GNSS navigation satellite system;

[0057] S120, determining dam body change information of the dam area according to continuous Beidou positioning data;

[0058] S130, using the GNSS satellite image as an input of the pre-trained deep learning model to obtain vegetation feature information contained in the GNSS satellite image, where the vegetation feature information includes vegetation type features and vegetation distribution features;

[0059] S140, performing cluster analysis on vegetation type characteristics and vegetation distribution characteristics to determine vegetation change trends;

[0060] S150. Determine the seepage problem of the dam based on the dam body change information and vegetation change trends.

[0061] In this embodiment, the Beidou positioning data can be used to determine the displacement, deformation, and changes in the surrounding terrain of the dam, thereby determining whether seepage has occurred; and GNSS satellite images can be used to determine changes in vegetation and landforms around the dam, thereby indirectly determining whether seepage has occurred in the dam and the area or impact range of the seepage.

[0062] Among them, multiple monitoring stations set up in the dam area are continuously positioned and monitored through the Beidou satellite positioning system. These monitoring stations are usually installed at key locations of the dam, such as the dam crest, dam abutment and dam foot, to ensure that the entire dam area can be fully covered. The Beidou positioning system accurately calculates the geographical coordinates of each monitoring station by receiving satellite signals. This process is carried out continuously, and the location information of the monitoring station can be obtained in real time to form continuous Beidou positioning data. Positioning monitoring of the monitoring stations set up in the dam area includes arranging multiple monitoring stations in the key areas of the dam, each station is equipped with a Beidou positioning receiver to ensure coverage of the entire dam area, and then receiving signals from Beidou navigation satellites in real time through the Beidou positioning receiver, and recording the precise location coordinates of each monitoring station.

[0063] The GNSS satellite system not only provides positioning services, but also regularly captures satellite images of the dam area through the high-resolution camera equipment it carries. The collected GNSS satellite images cover the entire dam area and can reflect the detailed conditions of the surface, including vegetation, terrain, and water bodies. These image data will be transmitted to the ground station for subsequent processing and analysis.

[0064] GNSS satellite images are acquired by remote sensing equipment carried by the GNSS navigation satellite system to collect images of the dam area. The collected raw satellite images are then preprocessed, including radiation correction, atmospheric correction, and geometric correction, to improve image quality. Image enhancement techniques, such as contrast enhancement and edge detection, are then used to facilitate subsequent identification and classification of related images.

[0065] The GNSS satellite system can regularly capture satellite images of the dam area through the high-resolution camera equipment it carries. The collected GNSS satellite images cover the entire dam area and can reflect the detailed conditions of the surface, including vegetation, terrain, and water bodies. These image data will be transmitted to the ground station for subsequent processing and analysis.

[0066] By analyzing the continuous Beidou positioning data, the dam body change information in the dam area is determined. Specifically, the continuous position coordinates of the monitoring stations obtained by the Beidou positioning system are used to calculate the displacement of these monitoring points at different time points. By comparing the position information at different times, the displacement change trend of the monitoring stations can be obtained.

[0067] For example, suppose the position coordinates of a monitoring station at the initial time are (x0, y0, z0), and the position coordinates at the subsequent time are (x t ,y t ,z t ), then the displacement of the monitoring station at time t can be calculated by the following formula:

[0068] Δx=x t -x0; Δy = y t -y0; Δz = z t -z0;

[0069]

[0070] Among them, Δx, Δy, and Δz represent the displacement of the monitoring station in three directions, respectively, and Δd represents the total displacement of the monitoring station.

[0071] By statistically analyzing the displacement data of all monitoring stations, a displacement distribution map of the dam area can be drawn to intuitively show the deformation of the dam. If the displacement of some areas is significantly greater than that of other areas, it may indicate that there are potential structural problems in the area, such as soil settlement or dam deformation caused by seepage.

[0072] The collected GNSS satellite images are input into the pre-trained deep learning model to extract vegetation feature information in the image. Specifically, the GNSS satellite images are first pre-processed, including image denoising, enhancement, and standardization steps to improve image quality and model recognition accuracy. Then, the pre-processed images are input into the deep learning model to extract low-level features (such as edges, textures, etc.) and high-level features (such as vegetation type, coverage, etc.) in the image.

[0073] In order to improve the recognition accuracy of the deep learning model, this embodiment uses a large amount of labeled satellite image data for model training and verification. Through repeated training and optimization, the deep learning model can accurately identify and classify different types of vegetation features and achieve high recognition accuracy.

[0074] Cluster analysis is performed on the extracted vegetation type characteristics and vegetation distribution characteristics to determine the vegetation change trend. Cluster analysis is an unsupervised learning method that identifies vegetation change patterns in different regions by grouping similar vegetation characteristics. Vegetation type characteristics include (but are not limited to) vegetation index, color characteristics, etc., and vegetation distribution characteristics include (but are not limited to) vegetation coverage (or vegetation coverage percentage), density, and distribution uniformity.

[0075] Specifically, the vegetation feature information at each time point is first represented as a feature vector, including vegetation type, coverage area, and distribution location, etc. Then, clustering algorithms (such as K-means, hierarchical clustering, etc.) are used to cluster these feature vectors and classify similar vegetation features into the same category.

[0076] For example, if the vegetation characteristics of a dam area show significant changes at different time points, cluster analysis can be used to identify which areas have changed vegetation types and which areas have increased or decreased vegetation coverage. By comparing the clustering results at different time points, the trend of vegetation changes can be determined. For example, if the vegetation in some areas gradually decreases, it may indicate that there is a seepage problem in the area, which leads to increased soil moisture and affects vegetation growth.

[0077] In order to further analyze the trend of vegetation changes, a vegetation change trend map can be drawn to intuitively display the vegetation changes in different areas. For example, a heat map can be used to show the changes in vegetation coverage, and the increase or decrease of vegetation can be reflected by the change in color. In this way, it is possible to intuitively observe which areas in the dam area have the most significant vegetation changes, so as to focus on the seepage problems in these areas.

[0078] The cluster analysis method can effectively identify and analyze the vegetation change trend in the dam area, providing an important reference for dam seepage monitoring. By timely discovering abnormal vegetation changes, measures can be taken in advance to prevent the seepage problem from further deteriorating and ensure the safe operation of the dam.

[0079] Finally, the dam body change information and vegetation change trends are combined to comprehensively analyze and determine the dam seepage problem. Specifically, by comparing the displacement changes of the dam body and the change trends of the vegetation, potential seepage problem areas can be identified.

[0080] For example, if the displacement of monitoring stations in a certain area increases significantly, and at the same time the vegetation coverage in the area decreases significantly, it can be inferred that there may be a seepage problem in the area. Seepage causes soil moisture to increase, affecting vegetation growth, and causing deformation and displacement of the dam structure. By comprehensively analyzing these two types of information, the specific location and severity of the seepage problem can be determined more accurately.

[0081] It should be noted that in addition to observing the changes in vegetation characteristics, the distribution and changes of surrounding water bodies can also be observed to determine possible seepage areas. For example, if the water area near the reservoir is found to increase abnormally, this may be a sign of seepage; or landform changes such as pits or uplifts may also be changes caused by seepage; and before calculating the dam body changes and recognizing vegetation images, the relevant data needs to be filtered to reduce interference, and the filtering process can be achieved through a micro-filtration model, which can be:

[0082] CDFM(D,T,P,C1,C2,...,Cn)=TM(REM(CFM(D,C1,C2,...,Cn),P),T)

[0083] Among them, CDFM is the first data set, D is the collection of Beidou positioning data and GNSS satellite images, T is the threshold of each data point, P is the preset category, C1, C2, ..., Cn are preset logical conditions, TM is the data point in the data set that is greater than or equal to the threshold, REM is the data point that meets the preset category, and CFM is the data point that meets the preset conditions.

[0084] Through the above steps, since the vegetation in the dam area will be affected when the dam seepage occurs, the dam body changes determined by Beidou positioning and the vegetation change trends determined by the image data collected by the GNSS satellite can be comprehensively judged to more comprehensively evaluate the seepage problem of the dam, provide more accurate early warning information, effectively identify the seepage situation of the dam body, and realize all-weather real-time safety monitoring of the dam area. At the same time, the deep learning model is used to pre-process and analyze the GNSS satellite images, which can automatically identify the vegetation characteristics and analyze the change trend, greatly improving the monitoring efficiency. Therefore, by integrating the dam body change information and the vegetation change trend, a more comprehensive seepage monitoring method is provided, which helps to improve the safety and reliability of the dam, solve the problem of low accuracy of dam seepage monitoring, and achieve the effect of improving the accuracy and efficiency of dam seepage safety monitoring.

[0085] The execution subject of the above steps may be a base station, a terminal, etc., but is not limited thereto.

[0086] Figure 2 A schematic diagram of the architecture of a deep learning model provided in an embodiment of the present application is shown, Figure 2 As shown, in an optional embodiment, the deep learning model is a Swin Transformer model, the Swin Transformer model includes a patch partitioning component (Patch Partitioning), a linear embedding component (Linear Embedding), a multi-scale feature fusion component (Multi-scale Feature Fusion), a Swin Transformer block component (Swin TransformerBlocks) and a block merging component (Patch Merging), the Swin Transformer model includes at least one feature layer, and the training steps of the Swin Transformer model include:

[0087] S210, in response to receiving the training image set, looping through the first step until a preset number of training times is reached;

[0088] The first step includes:

[0089] S220, determining initial parameters of the Swin Transformer model;

[0090] S230, using a block partitioning component to divide each image in the training image set into image blocks of a fixed size;

[0091] S240, using a linear embedding component to map each image block to a high-dimensional feature space through a linear layer to obtain a feature vector for each image block;

[0092] S250, using a multi-scale feature fusion component to fuse feature vectors of different scales to extract multi-scale features;

[0093] S260, input the multi-scale features into multiple Swin Transformer block components, and merge the blocks between different feature layers;

[0094] S270, updating the initial parameters of the Swin Transformer model according to a preset optimization algorithm.

[0095] like Figure 2 As shown, the Swin Transformer model includes a patch partitioning component (Patch Partitioning), a linear embedding component (Linear Embedding), a multi-scale feature fusion component (Multi-scale Feature Fusion), a Swin Transformer block component (Swin Transformer Blocks) and a block merging component (Patch Merging). Among them, the multi-scale feature fusion component is intended to fuse features of different scales to enhance the feature representation capability of the model. This component can be inserted between each Stage. Specifically, in this embodiment, the multi-scale feature fusion component is inserted between Stage1 and Stage2, and the specific connection relationship is:

[0096] The output feature map F1 of Stage 1 is obtained by extracting features using convolution kernels 1×1, 3×3 and 5×5. 1,1 , F 1,3 , and F 1,5 ; F 1,1 , F 1,3 and F 1,5 Perform concatenation or weighted summation to obtain the fusion feature F fusion ; for F fusion Perform channel attention processing to obtain the weighted feature F attention ; F attention As input, passed to Stage2.

[0097] By introducing a multi-scale feature fusion component, Swin Transformer can more effectively capture image details and global information at different scales, thereby improving the model's feature representation ability and image recognition accuracy. Especially when processing images with complex backgrounds and details, the multi-scale feature fusion component can significantly enhance the performance of the model.

[0098] In the process of training a deep learning model, you first need to receive a training image set, which contains a large amount of labeled image data for training the model. The training process is an iterative process called a training cycle. Each training cycle consists of a series of steps that are repeated until a preset number of training times is reached or other stopping conditions are met (such as achieving the expected accuracy). In each training cycle, the model gradually improves its ability to recognize image features by continuously adjusting parameters. The preset value of the number of training times is usually determined based on experience or experimental results, in order to ensure that the model can fully learn the feature information in the image, so as to show good recognition performance in practical applications.

[0099] Before starting training, you need to set the initial parameters for the Swin Transformer model. The initial parameters include the weights and biases in the model, which are usually randomly initialized. Random initialization can use methods such as Gaussian distribution and uniform distribution to ensure that the parameters have a certain diversity in the initial state, which helps the training of the model. Specifically, assuming that the size of a weight matrix W is m×n, the initial value can be randomly generated in the interval [-a, a] using uniform distribution, where a is a small constant.

[0100] Each image in the training image set is divided into fixed-size image blocks. This is achieved through the block partitioning component. Assuming that the size of the image is H×W and the size of each image block is P×P, the entire image can be divided into (H / P)×(W / P) image blocks. The purpose of image block partitioning is to decompose large images into smaller units, which are easier to process and analyze. Each image block can be regarded as a small local area that contains the feature information in the area. In this way, the model can better capture the local features in the image and improve the accuracy of feature extraction.

[0101] After the image block division is completed, the next step is to map each image block to a high-dimensional feature space. This is achieved through the linear embedding component. The linear embedding component contains one or more linear layers, each of which can be regarded as a fully connected layer. Assuming that the size of each image block is P×P, it can be flattened into a vector with a length of P2. Then, this vector is mapped to a high-dimensional feature space through a linear layer to obtain a feature vector. Assuming that the length of the feature vector is D, the mapping process can be expressed as: v=Wx+b, where x is the flattened image block vector, W is the weight matrix, b is the bias vector, and v is the mapped feature vector. In this way, each image block is converted into a high-dimensional feature vector, which is convenient for subsequent feature fusion and processing.

[0102] After obtaining the feature vector of each image block, feature fusion is required to extract multi-scale features. The role of the multi-scale feature fusion component is to fuse feature vectors of different scales to obtain a richer feature representation. Multi-scale feature fusion can be achieved through multi-layer convolution operations, and each layer of convolution operations can extract feature information of different scales. For example, the first layer of convolution can extract local detail features, the second layer of convolution can extract features of a larger range, and so on. In this way, multi-scale features containing information of different scales can be obtained. Specifically, assuming that the size of the input feature vector is D, after multi-layer convolution operations, the size of the obtained multi-scale feature vector is D′. These multi-scale feature vectors contain feature information at different levels in the image, providing a rich foundation for subsequent feature processing.

[0103] The extracted multi-scale features are input into multiple Swin Transformer block components for processing. The Swin Transformer block component is the core part of the Swin Transformer model. It captures global and local features in the image through self-attention mechanism and window operation. Each Swin Transformer block component contains multiple self-attention layers and feedforward neural network layers. Through multi-level feature processing, it gradually extracts and fuses feature information in the image. In order to further improve the effect of feature extraction, block merging operations are also performed between different feature layers. The block merging operation can be implemented through a pooling layer or a convolutional layer to fuse the feature vectors of different feature layers to obtain a higher-level feature representation. In this way, the feature extraction capability of the model can be improved to capture more details and global information in the image.

[0104] At the end of each training cycle, the initial parameters of the model need to be updated according to the preset optimization algorithm. The role of the optimization algorithm is to improve the performance of the model by adjusting the parameters of the model so that the prediction results of the model gradually approach the true value. Commonly used optimization algorithms include stochastic gradient descent (SGD), Adam, etc. The optimization algorithm guides the update direction and step size of the parameters by calculating the gradient of the loss function. Assuming the loss function is L(θ), where θ is the parameter vector of the model, the optimization algorithm updates the parameters through the following formula:

[0105]

[0106] Where η is the learning rate, is the gradient of the loss function at the parameter. Through repeated iterations and parameter updates, the performance of the model will gradually improve and eventually achieve the desired effect.

[0107] In an optional embodiment, the Swin Transformer model includes a primary feature layer, a mid-level feature layer, a high-level feature layer, and a global semantic feature layer.

[0108] In this implementation, the Swin Transformer model can gradually learn and extract multi-scale features in the image, and effectively capture global and local information through the self-attention mechanism. First, the training image set is mapped to a high-dimensional feature space through block partitioning and linear embedding. Then, features of different scales are extracted through the multi-scale feature fusion component and input into the Swin Transformer block component for deep processing. Block merging is performed between different feature layers to further improve the richness and accuracy of feature expression. Finally, the preset optimization algorithm is used to continuously adjust the model parameters and gradually improve the recognition performance of the model, which can effectively improve the feature extraction capability and enhance the generalization performance.

[0109] In an optional embodiment, determining the dam body change information of the dam area according to the continuous Beidou positioning data includes:

[0110] S310, performing real-time dynamic differentiation on the continuous Beidou positioning data to obtain continuous Beidou positioning data after differential positioning;

[0111] S320, performing time series analysis on the continuous Beidou positioning data after differential positioning, and identifying the change trend of the dam body position in the dam area and the sudden change positioning data of each monitoring station;

[0112] S330, calculating the displacement of each monitoring site according to the continuous Beidou positioning data after differential positioning, wherein the displacement includes horizontal displacement and vertical displacement;

[0113] S340, determining the dam body change information of each monitoring station in the dam area according to the dam body position change trend, mutation positioning data and displacement.

[0114] In dam monitoring, continuous Beidou positioning data can provide high-precision location information. However, due to the influence of factors such as the atmosphere, ionosphere, and satellite orbit errors, the original Beidou positioning data may have certain errors. In order to improve the positioning accuracy, real-time dynamic differential technology is used to process the continuous Beidou positioning data. Real-time dynamic differential (RTK) is a high-precision satellite navigation technology that eliminates or reduces positioning errors by using differential correction information provided by one or more reference stations. In specific implementation, the reference station is located at a known fixed position, continuously receives Beidou satellite signals, and calculates error information. The monitoring station corrects the received Beidou signal based on the error information provided by the reference station, thereby obtaining high-precision Beidou positioning data after differential positioning. In this way, the positioning error can be reduced from the meter level to the centimeter level or even higher, providing a reliable data basis for subsequent dam change monitoring.

[0115] Based on the continuous Beidou positioning data after differential positioning, time series analysis is an important step to identify the changing trend of the dam body position in the dam area. Time series analysis is a statistical method used to analyze data sequences that change over time to identify the laws and trends therein. In specific implementation, the positioning data can be smoothed by methods such as sliding average and weighted average to eliminate short-term fluctuations and highlight long-term trends. Then, the changing trend of the dam body position can be identified by methods such as trend analysis, seasonal analysis, and periodic analysis. In addition, anomaly detection techniques such as Z-score and DBSCAN can be used to identify the mutation positioning data of monitoring stations. These mutation data may indicate that some stations have experienced sudden displacements, such as displacement changes caused by earthquakes and landslides. Through time series analysis, the overall change trend and local mutation situation of the dam area can be fully understood, providing an important reference for dam safety monitoring.

[0116] After identifying the change trend and mutation positioning data of the dam area, it is necessary to further calculate the specific displacement of each monitoring station. The displacement calculation is based on the continuous Beidou positioning data after differential positioning, which mainly includes horizontal displacement and vertical displacement. The horizontal displacement can be obtained by calculating the position change of the monitoring station in the plane coordinate system. The vertical displacement represents the position change of the monitoring station in the vertical direction, which can be calculated by the change of elevation data.

[0117] According to the change trend obtained by time series analysis, the overall change pattern of the dam area can be determined, such as long-term settlement, tilt, etc. Secondly, by analyzing the mutation positioning data, abnormal change points in the dam area can be identified to determine whether there are sudden displacement events. Finally, combined with the specific displacement of each monitoring station, the changes in the dam area are comprehensively evaluated. In specific implementation, data fusion technology can be used to comprehensively process multi-source data to obtain detailed change information for each monitoring station. These change information can include displacement direction, displacement speed, change amplitude, etc., providing a comprehensive description of dam changes.

[0118] This implementation method can comprehensively and accurately monitor the dam body change information in the dam area through real-time dynamic differential processing, time series analysis, displacement calculation and comprehensive analysis of continuous Beidou positioning data. Real-time dynamic differential technology improves the accuracy of positioning data, time series analysis identifies the dam change trend and mutation point, displacement calculation provides specific displacement data, and comprehensive analysis integrates multi-source information to obtain a detailed description of dam body changes. This comprehensive monitoring process can not only detect potential safety hazards of the dam in a timely manner, but also provide a scientific basis for the maintenance and management of the dam, ensuring the long-term safe operation of the dam.

[0119] In an optional embodiment, the dam body position change trend includes the dam body horizontal displacement trend and the dam body vertical displacement trend. According to the dam body position change trend, the mutation positioning data and the displacement, the dam body change information of each monitoring station in the dam area is determined, including:

[0120] S410, determining the change of the dam body according to the horizontal displacement trend and the vertical displacement trend of the dam body, wherein the change of the dam body includes uniform settlement, local sliding, foundation problems, crack generation, elevation of the infiltration line and deformation of the dam body;

[0121] S420, determining m suspicious dam body change situations at each monitoring station in the dam body change situation according to the mutation positioning data and displacement of each monitoring station, where m is a positive integer;

[0122] S430, summarizing the suspicious dam body changes at all monitoring sites, and constructing a dam body three-dimensional model, the dam body three-dimensional model is used to simulate the suspicious dam body changes at each monitoring site;

[0123] S440. According to the simulation results of the dam body three-dimensional model, the mutation positioning data and displacement of each monitoring site are corrected, and according to the corrected mutation positioning data and displacement of each monitoring site, the final dam body change situation is determined from the m suspected dam body change situations. The final dam body change situation is used to characterize the dam body change information.

[0124] When determining the changes in the dam body, it is first necessary to analyze the horizontal displacement trend and vertical displacement trend of the dam body. The horizontal displacement trend can be determined by the change in the horizontal displacement of the monitoring station at different time points. For example, if multiple monitoring stations show consistent horizontal displacement in the same direction, this may indicate that the dam body as a whole has undergone horizontal displacement. The vertical displacement trend is determined by the elevation change of the monitoring station. If multiple monitoring stations show consistent elevation drop, this may indicate that the dam body has undergone uniform settlement. In specific implementation, a variety of statistical analysis methods, such as regression analysis and trend analysis, can be used to extract displacement trend information. By comprehensively analyzing the horizontal and vertical displacement trends, the changes in the dam body can be determined. For example, if the monitoring stations in a certain area show significant horizontal displacement and small vertical displacement, this may indicate that local sliding has occurred in the area. If the monitoring stations in a certain area show elevation drop accompanied by horizontal displacement, this may indicate that there is a problem with the foundation. In addition, by analyzing the mutation points of the displacement trend, problems such as crack generation and elevated infiltration line can be identified. In this way, the changes in the dam body can be fully understood, providing an important basis for subsequent safety assessment and risk management.

[0125] When determining the suspicious dam changes at each monitoring site, it is necessary to combine the mutation positioning data and displacement for comprehensive analysis. Mutation positioning data refers to the significant displacement changes that occur at a monitoring site at a certain point in time. These data usually reflect the changes in the dam caused by sudden events such as earthquakes and landslides. The displacement includes horizontal displacement and vertical displacement, reflecting the cumulative displacement of the monitoring site over a period of time. In specific implementation, pattern matching and anomaly detection techniques can be used to match the mutation positioning data with the known dam change pattern to preliminarily determine the suspicious dam changes. For example, if a monitoring site shows significant horizontal and vertical displacements at a certain point in time, this may indicate that the site has local sliding or foundation problems. Then, by further analyzing the trend of displacement changes, multiple suspicious changes at the site can be determined. For example, a monitoring site shows a continuous drop in elevation and horizontal displacement, which may indicate that the site has uniform settlement and foundation problems.

[0126] After summarizing the suspected dam changes at all monitoring sites, constructing a three-dimensional model of the dam is an important step to further analyze and simulate the changes in the dam. The three-dimensional model can intuitively display the structure and changes of the dam, helping to identify potential safety hazards. In specific implementation, computer-aided design (CAD) software or geographic information system (GIS) software can be used to construct a three-dimensional model of the dam based on the location information and displacement data of the monitoring sites. First, the geographic coordinates and elevation data of all monitoring sites are input into the software to generate an initial three-dimensional model of the dam. Then, the model is adjusted according to the suspected dam changes at each monitoring site. For example, if a monitoring site shows significant horizontal and vertical displacements, the coordinates of the site can be adjusted accordingly in the model to simulate its displacement. In this way, the suspected dam changes at each monitoring site can be intuitively displayed in the three-dimensional model, helping to identify the change trend of the overall dam and local abnormalities. The three-dimensional model can also be used to simulate the dam structure under different changes, evaluate its stability and safety, and provide a scientific basis for the maintenance and management of the dam.

[0127] After constructing the three-dimensional model of the dam body and conducting preliminary simulations, it is necessary to correct the mutation positioning data and displacement of each monitoring station according to the simulation results. The purpose of the correction is to improve the accuracy of the data and ensure that the final changes in the dam body are more reliable. In specific implementation, the mutation positioning data and displacement of the monitoring station can be adjusted through an iterative optimization algorithm to make them more consistent with the simulation results of the three-dimensional model. For example, if the simulation results of a monitoring station show that its displacement is deviated from the actual observation data, the displacement of the station can be adjusted until the simulation results are consistent with the actual data. Then, based on the corrected mutation positioning data and displacement, the suspected dam body changes of each monitoring station are further analyzed to determine the final dam body changes. For example, if a monitoring station still shows significant horizontal and vertical displacements after correction, it can be finally determined that local sliding or foundation problems have occurred at the station.

[0128] For example, during a monitoring period, displacement data from all monitoring stations were collected. By analyzing these data, it was found that multiple monitoring stations showed significant horizontal and vertical displacements during a certain period of time. For example, monitoring stations A, B, and C showed downward vertical displacements and eastward horizontal displacements, while monitoring stations D, E, and F showed southward horizontal displacements and smaller vertical displacements.

[0129] Through further analysis, it was found that the vertical displacements of sites A, B, and C were relatively consistent, indicating that uniform settlement may have occurred in this area. At the same time, the horizontal displacement trends of sites D, E, and F were relatively consistent, but the vertical displacements were small, which may indicate that local sliding occurred in this area. Combining these trends, we can preliminarily judge the changes in the dam, including uniform settlement and local sliding.

[0130] Based on the mutation location data and displacement of the monitoring site, the suspected changes of each site are further analyzed. For example, the mutation location data of site A shows that a significant vertical displacement occurred at a certain point in time, and the displacement shows that its cumulative vertical displacement during that time period is large. Combining this information, it can be determined that the suspected changes of site A include uniform settlement and foundation problems.

[0131] Similarly, the mutation location data of site D indicate that a significant horizontal displacement occurred at a certain point in time, and the displacement amount shows that the cumulative horizontal displacement during this period is large, but the vertical displacement is small. It can be determined that the suspected changes at site D include local sliding and crack generation.

[0132] After the data from all monitoring stations are collected, a 3D model of the dam is constructed using CAD software. First, the geographic coordinates and elevation data of all monitoring stations are input to generate an initial 3D model. Then, the model is adjusted according to the suspected changes at each station. For example, the vertical displacement of stations A, B, and C is reflected in the model to simulate uniform settlement; the horizontal displacement of stations D, E, and F is reflected in the model to simulate local sliding.

[0133] According to the simulation results of the three-dimensional model, the mutation positioning data and displacement of each monitoring site are corrected. For example, if the simulation results of site A are found to be deviated from the actual data, its vertical displacement is adjusted through an iterative optimization algorithm until the simulation results are consistent with the actual data.

[0134] After correction, further analysis of the suspected changes at each monitoring site was conducted to determine the final changes in the dam body. For example, site A still showed significant vertical displacement after correction, which ultimately determined that it had uniform settlement and foundation problems. Site D showed significant horizontal displacement after correction, which determined that it had localized sliding and cracks.

[0135] This implementation method can comprehensively understand the changes in the dam area by analyzing the horizontal and vertical displacement trends of the dam body, combined with the mutation positioning data and displacement. By constructing a three-dimensional model of the dam body, the suspected dam body changes at each monitoring station are intuitively displayed and simulated, and the data is corrected according to the simulation results to ensure that the final dam body changes are more accurate and reliable. This comprehensive monitoring process can not only detect potential safety hazards of the dam in a timely manner, but also provide a scientific basis for the maintenance and management of the dam, ensuring the long-term safe operation of the dam. Through this systematic and refined monitoring and analysis method, the accuracy and reliability of dam safety monitoring can be effectively improved, providing comprehensive support for the safety management of dam projects.

[0136] In an optional embodiment, determining the seepage problem of the dam based on the dam body change information and the vegetation change trend includes:

[0137] S510, performing correlation analysis on the dam body change information and the vegetation change trend to determine the correlation characteristics between the dam body change information and the vegetation change trend;

[0138] S520, inputting the associated features into the dam seepage prediction model to obtain the dam seepage problem.

[0139] Statistical methods, such as the Pearson correlation coefficient and the Spearman rank correlation coefficient, are used to calculate the correlation between the dam change information and the vegetation change trend. These methods can quantify the degree of correlation between the two. For example, if it is found that the settlement of the dam body and the decrease in vegetation coverage in a certain period show a significant positive correlation, it can be preliminarily determined that there is a correlation between the two.

[0140] In order to further determine this correlation feature, a machine learning algorithm can be used to build a correlation model, with the dam change information as the input feature and the vegetation change trend as the output label. By training the correlation model, the complex relationship between the two can be identified. The correlation model can be a random forest model or a support vector machine, which will not be described here.

[0141] After completing the correlation analysis and determining the correlation characteristics between the dam body change information and the vegetation change trend, these characteristics were input into the dam seepage prediction model.

[0142] Specifically, in an optional embodiment, the dam seepage prediction model includes a logistic regression module and a linear regression module, the dam seepage problem includes the dam seepage probability and the seepage severity, and the associated features are input into the dam seepage prediction model to obtain the dam seepage problem, including:

[0143] S610, inputting the associated features into the dam seepage prediction model to obtain the dam seepage probability and seepage severity, wherein the dam seepage probability is obtained based on the logistic regression module, and the seepage severity is obtained based on the linear regression module.

[0144] The logistic regression module is used to predict the probability of dam seepage. Logistic regression is a widely used classification algorithm suitable for binary classification problems. In this model, the input variables are the associated features extracted previously, such as the settlement of the dam body, crack width, vegetation coverage, etc. Logistic regression calculates the linear combination of the input features and then maps them to a probability value between 0 and 1 through the Sigmoid function. The specific formula is as follows:

[0145]

[0146] Where P(y=1|X) represents the probability of seepage at the dam, β0 is the bias term, and β1, β2, ..., β n are model parameters, X1, X2, …, X n The model is trained by the maximum likelihood estimation method to obtain the optimal parameters. After training, the model can predict the probability of seepage in the dam based on the input features. For example, if the input features show that the settlement of the dam and the width of the cracks increase significantly over a certain period of time, the model may predict that the probability of seepage in the dam is high.

[0147] The linear regression module is used to predict the severity of dam seepage. Linear regression is a regression analysis method suitable for predicting continuous variables. In this model, the input variables are also the associated features extracted previously. Linear regression predicts by establishing a linear relationship between the input features and the output variables. The specific formula is as follows:

[0148] Y=α0+α1X1+α2X2+…+α n X n ;

[0149] Where Y represents the severity of seepage, α0 is the bias term, and α0, α1, ..., α n are model parameters, X1, X2, …, X n is the input feature.

[0150] The model is trained by the least squares method to obtain the optimal parameters. After training, the model can predict the severity of dam seepage based on the input features. For example, if the input features show that the settlement of the dam body and the width of the cracks have increased significantly over a certain period of time, the model may predict that the severity of the seepage is high. By inputting the associated features into the dam seepage prediction model, the probability and severity of the dam seepage can be obtained at the same time. This dual prediction method can provide a more comprehensive and accurate seepage risk assessment. For example, when the model predicts that the probability of dam seepage is high and the severity of seepage is large over a certain period of time, corresponding protective measures can be taken in time to prevent serious leakage accidents in the dam.

[0151] This implementation method inputs the correlation characteristics of dam body change information and vegetation change trends into the dam seepage prediction model, and obtains the probability and severity of dam seepage respectively. The logistic regression module predicts the probability of seepage in the dam by calculating the linear combination of input features, and provides a quantitative assessment of seepage risk. The linear regression module predicts the severity of seepage by establishing a linear relationship between input features and seepage severity, and provides a qualitative assessment of seepage risk. Combining these two prediction results can provide a more comprehensive and accurate risk assessment for the safety monitoring and management of the dam, help identify and prevent dam leakage risks, and ensure the long-term safe operation of the dam. Through this systematic and refined monitoring and analysis method, the accuracy and reliability of dam safety monitoring are effectively improved, providing comprehensive support for the safety management of dam projects.

[0152] In an optional embodiment, after using the GNSS satellite image as an input of the pre-trained deep learning model to obtain vegetation feature information contained in the GNSS satellite image, the method further includes:

[0153] S710, constructing a vegetation feature matrix based on vegetation feature information and according to a preset matrix construction rule;

[0154] S720, performing correlation calculation on the vegetation feature matrix to obtain a feature correlation value;

[0155] S730: When the feature correlation value does not meet the preset threshold range condition, it is determined that there is an abnormality in the vegetation feature information.

[0156] In this embodiment, under normal circumstances, the type of vegetation and its distribution are related. For example, when seepage occurs, the number of water-loving plants increases, and the distribution distance between the water-loving plants decreases. At the same time, the density of hydrophobic plants decreases, the distribution amount decreases, and the distribution distance between the hydrophobic plants increases. If the distribution distance between the water-loving plants and the distribution distance between the hydrophobic plants increase or decrease at the same time, it means that there is an abnormality in the identified data, and so on.

[0157] Among them, the preset matrix construction rules can be the vegetation feature types and distribution distance changes represented by the rows and columns of the matrix, or other construction rules, which can be adjusted according to actual needs; the correlation calculation can be based on the Pearson correlation coefficient, or it can be based on other correlation calculation formulas; before constructing the matrix, it is necessary to filter the relevant data to reduce the interference of other data. At the same time, for different feature matrices, the relevant features need to be normalized to ensure that the relevant data can be calculated in a unified data format; when filling in matrix elements, for blank data, zero filling can be used to fill it, or the relevant data can be directly eliminated, which can be adjusted according to actual needs;.

[0158] For example, the distribution distances between plants 1-3 in region 1 may vary as shown in the table:

[0159] Table 1

[0160] Plant 1 Plant 2 Plant 3 Plant 4 +10 -12 +5 Plant 5 +20 +15 -20 Plant 6 -30 +10 +14

[0161] Thus, the vegetation feature matrix H is obtained: Then, the vegetation feature matrix H is correlated. When the correlation result and the similarity results between different feature matrices meet the threshold condition, it means that the feature data corresponding to the matrix element is normal. Otherwise, it is judged to be abnormal, and the first model needs to be adjusted.

[0162] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods of various embodiments of the present invention.

[0163] In this embodiment, a GNSS dam seepage safety monitoring system based on Beidou positioning and image recognition is also provided. The system is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0164] Figure 3: is a structural block diagram of a GNSS dam seepage safety monitoring system based on Beidou positioning and image recognition according to an embodiment of the present invention, such as Figure 3 As shown, the system includes:

[0165] A data acquisition module is used to obtain continuous Beidou positioning data and GNSS satellite images of the dam area, wherein the Beidou positioning data is obtained by continuously positioning and monitoring the monitoring stations set up in the dam area, and the GNSS satellite images are obtained by collecting images of the dam area through the GNSS navigation satellite system;

[0166] The data processing module is used to determine the dam body change information in the dam area based on the continuous Beidou positioning data;

[0167] A feature recognition module is used to use the GNSS satellite image as the input of the pre-trained deep learning model to obtain vegetation feature information contained in the GNSS satellite image, where the vegetation feature information includes vegetation type features and vegetation distribution features;

[0168] The feature analysis module is used to perform cluster analysis on vegetation type characteristics and vegetation distribution characteristics to determine vegetation change trends;

[0169] The decision-making module is used to determine the seepage problem of the dam based on the dam body change information and vegetation change trend.

[0170] It should be noted that the above modules can be implemented by software or hardware. For the latter, it can be implemented in the following ways, but not limited to: the above modules are all located in the same processor; or the above modules are located in different processors in any combination.

[0171] An embodiment of the present invention further provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0172] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0173] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0174] In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0175] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.

[0176] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0177] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A GNSS dam seepage safety monitoring method based on Beidou positioning and image recognition, characterized in that: include: Acquire continuous Beidou positioning data and GNSS satellite images of the dam area, wherein the Beidou positioning data is obtained by continuously positioning and monitoring monitoring stations set up in the dam area, and the GNSS satellite images are obtained by collecting images of the dam area through the GNSS navigation satellite system; Determine dam body change information of the dam area according to the continuous Beidou positioning data; Using the GNSS satellite image as an input of a pre-trained deep learning model to obtain vegetation feature information contained in the GNSS satellite image, wherein the vegetation feature information includes vegetation type features and vegetation distribution features; Performing cluster analysis on the vegetation type characteristics and the vegetation distribution characteristics to determine the vegetation change trend; The seepage problem of the dam is determined based on the dam body change information and the vegetation change trend.

2. The method according to claim 1, characterized in that The deep learning model is a Swin Transformer model, which includes a block partition component, a linear embedding component, a multi-scale feature fusion component, a Swin Transformer block component, and a block merging component. The Swin Transformer model includes at least one feature layer, and the training steps of the Swin Transformer model include: In response to receiving the training image set, looping through the first step until a preset number of training times is reached; Wherein, the first step comprises: Determining initial parameters of the Swin Transformer model; Using the block partitioning component to partition each image in the training image set into image blocks of fixed size; Using the linear embedding component, each of the image blocks is mapped to a high-dimensional feature space through a linear layer to obtain a feature vector of each image block; The multi-scale feature fusion component is used to fuse the feature vectors of different scales to extract multi-scale features; Inputting the multi-scale features into a plurality of Swin Transformer block components, and merging blocks between different feature layers; The initial parameters of the Swin Transformer model are updated according to a preset optimization algorithm.

3. The method according to claim 2, characterized in that The Swin Transformer model includes a primary feature layer, a mid-level feature layer, a high-level feature layer and a global semantic feature layer.

4. The method according to claim 1, characterized in that Determining the dam body change information of the dam area according to the continuous Beidou positioning data includes: Performing real-time dynamic differentiation on the continuous Beidou positioning data to obtain continuous Beidou positioning data after differential positioning; Performing time series analysis on the continuous Beidou positioning data after the differential positioning to identify the changing trend of the dam body position in the dam area and the sudden change positioning data of each monitoring station; Calculate the displacement of each monitoring site according to the continuous Beidou positioning data after differential positioning, wherein the displacement includes horizontal displacement and vertical displacement; The dam body change information of each of the monitoring sites in the dam area is determined according to the dam body position change trend, the mutation positioning data and the displacement.

5. The method according to claim 4, characterized in that The dam body position change trend includes a dam body horizontal displacement trend and a dam body vertical displacement trend. Determining the dam body change information of each monitoring station in the dam area according to the dam body position change trend, the mutation positioning data and the displacement amount includes: Determine the change of the dam body according to the horizontal displacement trend of the dam body and the vertical displacement trend of the dam body, wherein the change of the dam body includes uniform settlement, local sliding, foundation problems, crack generation, elevation of the infiltration line and deformation of the dam body; According to the mutation positioning data and the displacement of each monitoring site, m suspicious dam body change situations of each monitoring site are determined in the dam body change situation, where m is a positive integer; Summarizing the change conditions of the suspected dam body at all the monitoring sites and constructing a three-dimensional model of the dam body, wherein the three-dimensional model of the dam body is used to simulate the change conditions of the suspected dam body at each of the monitoring sites; The mutation location data and the displacement amount of each monitoring site are corrected according to the simulation results of the dam body three-dimensional model, and the final dam body change situation is determined from the m suspected dam body change situations according to the corrected mutation location data and the displacement amount of each monitoring site, and the final dam body change situation is used to characterize the dam body change information.

6. The method according to claim 5, characterized in that Determining the dam seepage problem according to the dam body change information and the vegetation change trend includes: Performing correlation analysis on the dam body change information and the vegetation change trend to determine correlation characteristics between the dam body change information and the vegetation change trend; The associated features are input into a dam seepage prediction model to obtain the dam seepage problem.

7. The method according to claim 6, characterized in that The dam seepage prediction model includes a logistic regression module and a linear regression module, the dam seepage problem includes dam seepage probability and seepage severity, and the associated features are input into the dam seepage prediction model to obtain the dam seepage problem, including: The associated features are input into the dam seepage prediction model to obtain the dam seepage probability and the seepage severity, wherein the dam seepage probability is obtained based on the logistic regression module, and the seepage severity is obtained based on the linear regression module.

8. The method according to claim 1, characterized in that After taking the GNSS satellite image as an input of the pre-trained deep learning model to obtain vegetation feature information contained in the GNSS satellite image, the method further includes: Based on the vegetation feature information, constructing a vegetation feature matrix according to a preset matrix construction rule; Performing correlation calculation on the vegetation feature matrix to obtain a feature correlation value; When the feature correlation value does not meet the preset threshold range condition, it is determined that the vegetation feature information is abnormal.

9. A GNSS dam seepage safety monitoring system based on Beidou positioning and image recognition, characterized in that: include: A data acquisition module is used to obtain continuous Beidou positioning data and GNSS satellite images of the dam area, wherein the Beidou positioning data is obtained by continuously positioning and monitoring the monitoring stations set up in the dam area, and the GNSS satellite images are obtained by collecting images of the dam area through the GNSS navigation satellite system; A data processing module, used for determining the dam body change information of the dam area according to the continuous Beidou positioning data; A feature recognition module, used to use the GNSS satellite image as an input of a pre-trained deep learning model to obtain vegetation feature information contained in the GNSS satellite image, wherein the vegetation feature information includes vegetation type features and vegetation distribution features; A feature analysis module, used to perform cluster analysis on the vegetation type features and the vegetation distribution features to determine the vegetation change trend; The decision module is used to determine the seepage problem of the dam according to the dam body change information and the vegetation change trend.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 8 when executed.

Images