A dam leakage electrical anomaly identification method, system and electronic device

Abstract

The application provides a dam leakage electrical anomaly identification method, system and electronic equipment, and relates to the technical field of dam leakage electrical anomaly identification. The method comprises the following steps: acquiring an apparent resistivity image of a to-be-detected dam; inputting the apparent resistivity image of the to-be-detected dam into a dam leakage electrical anomaly identification model to determine the position and number of the leakage electrical anomaly area in the to-be-detected dam; and the dam leakage electrical anomaly identification model is obtained by training an end-to-end deep learning model using the labeled apparent resistivity image. The end-to-end deep learning model is constructed and trained, so that the recognition accuracy and efficiency of the dam leakage electrical anomaly area are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of dam leakage electrical anomaly identification, and particularly relates to a dam leakage electrical anomaly identification method and system and electronic equipment. BACKGROUND

[0002] With the development of water conservancy construction, the number of reservoir dams increases year by year, and the problem of reservoir dam leakage is increasingly prominent. On the one hand, old reservoir dams have leakage problems due to long-term disrepair. On the other hand, new reservoir dams also have leakage problems due to construction and material problems. Leakage problems seriously threaten the safe operation and maintenance of dams, so in-depth research on dam leakage diagnosis technology and safety evaluation is of great significance to ensure the safe and stable operation of dams. High-density electrical method is widely used in dam leakage anomaly detection as a non-destructive testing method with high precision, good reliability and low cost. In the traditional low-resistance anomaly identification of dams, researchers mainly observe the inversion apparent resistivity by visual observation to determine the target area range of low-resistance anomaly, but this method is too dependent on experience and time-consuming, and requires high professional requirements, is easily affected by subjective factors, and has relatively low precision and accuracy. With the increase of dam detection scale and the sharp increase of collected data, rapid and accurate positioning of dam leakage electrical anomaly area also becomes a problem to be solved. SUMMARY

[0003] The purpose of the present application is to provide a dam leakage electrical anomaly identification method, system and electronic equipment, which can improve the identification precision and efficiency of dam leakage electrical anomaly area.

[0004] To achieve the above purpose, the present application provides the following solutions:

[0005] A dam leakage electrical anomaly identification method comprises the following steps:

[0006] An apparent resistivity image of a dam to be measured is obtained.

[0007] The apparent resistivity image of the dam to be measured is input into a dam leakage electrical anomaly identification model to determine the leakage electrical anomaly condition in the dam to be measured. The dam leakage electrical anomaly identification model is obtained by training an end-to-end deep learning model using labeled apparent resistivity images. The leakage electrical anomaly condition includes the number of leakage electrical anomaly areas and the position of each leakage electrical anomaly area.

[0008] Optionally, the end-to-end deep learning model comprises a front end, a back end and a post-processing module connected in sequence.

[0009] The front end comprises an encoder, and the encoder is obtained by removing the full connection layer in the VGG16 deep convolutional neural network.

[0010] The back end comprises a split branch decoder and an embedded branch decoder; an input end of the split branch decoder and an input end of the embedded branch decoder are connected with an output end of the encoder; an output end of the split branch decoder and an output end of the embedded branch decoder are connected with an input end of the post-processing module;

[0011] The split branch decoder is used for outputting a seepage electrical abnormal area recognition result.

[0012] The embedded branch decoder is used for outputting a seepage electrical abnormal area mask image.

[0013] The post-processing module is used for performing clustering processing on the seepage electrical abnormal area mask image, combining a clustering result with an image output by the split branch decoder, obtaining a classification result of the seepage electrical abnormal area, determining whether seepage abnormal resistivities are the same according to visual features of the abnormal area, extracting seepage electrical abnormal features, and determining a seepage electrical abnormal condition in combination with the seepage electrical abnormal area mask image.

[0014] Optionally, before the apparent resistivity image of the to-be-tested dam is acquired, the method further comprises:

[0015] The end-to-end deep learning model is constructed.

[0016] A plurality of apparent resistivity historical images of the dam are acquired.

[0017] Seepage electrical abnormal areas in the plurality of apparent resistivity historical images and labeled types of the seepage electrical abnormal areas are labeled to obtain a plurality of apparent resistivity historical labeled images.

[0018] The end-to-end deep learning model is trained by taking the apparent resistivity historical images as input and taking the plurality of apparent resistivity historical labeled images as output, to obtain a dam seepage electrical abnormality recognition model.

[0019] Optionally, the labeled type is an elliptical frame or a square frame.

[0020] A dam seepage electrical abnormality recognition system comprises:

[0021] An apparent resistivity image acquisition module is configured to acquire an apparent resistivity image of a to-be-tested dam.

[0022] A seepage electrical abnormal area recognition module is configured to input the apparent resistivity image of the to-be-tested dam into a dam seepage electrical abnormality recognition model to determine a seepage electrical abnormal condition in the to-be-tested dam; the dam seepage electrical abnormality recognition model is obtained by training an end-to-end deep learning model by using labeled apparent resistivity images; and the seepage electrical abnormal condition comprises a number of seepage electrical abnormal areas and a position of each seepage electrical abnormal area.

[0023] An electronic device includes a memory for storing a computer program and a processor for running the computer program to cause the electronic device to perform the dam leakage electrical anomaly identification method.

[0024] Optionally, the memory is a readable storage medium.

[0025] According to the specific embodiments of the present application, the following technical effects are disclosed:

[0026] The present application provides a dam leakage electrical anomaly identification method, system and electronic device, obtains the apparent resistivity image of the dam to be measured; the apparent resistivity image of the dam to be measured is input into the dam leakage electrical anomaly identification model to determine the leakage electrical anomaly area in the dam to be measured; the dam leakage electrical anomaly identification model is obtained by training the end-to-end deep learning model using the labeled apparent resistivity image. The present application can improve the recognition accuracy and efficiency of the dam leakage electrical anomaly area by constructing and training the end-to-end deep learning model. BRIEF DESCRIPTION OF DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0028] Figure 1 The dam leakage electrical anomaly identification method flowchart in the present application embodiment 1;

[0029] Figure 2 The mind map of the dam leakage electrical anomaly identification method based on end-to-end deep learning in the present application embodiment 1;

[0030] Figure 3 The data set schematic diagram in the present application embodiment 1;

[0031] Figure 4 The ARNet model structure schematic diagram in the present application embodiment 1;

[0032] Figure 5 The segmentation branch clustering effect schematic diagram in the present application embodiment 1;

[0033] Figure 6 The network training intersection over union schematic diagram in the present application embodiment 1;

[0034] Figure 7 The network training loss value schematic diagram in the present application embodiment 1;

[0035] Figure 8 For the ARNet model test identification effect in embodiment 1 of the application;

[0036] Figure 9 For the abnormal position predicted by the ARNet model in embodiment 1 of the application;

[0037] Figure 10 For the inversion result schematic diagram of the Res2Dinv inversion software in embodiment 1 of the application;

[0038] Figure 11 For the abnormal position and the drilling soil sample of the study area circled by the ARNet in embodiment 1 of the application. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the application.

[0040] The purpose of the application is to provide a dam leakage electrical anomaly identification method, system and electronic equipment, which can improve the identification accuracy and efficiency of the dam leakage electrical anomaly area.

[0041] In order to make the above-mentioned purposes, characteristics and advantages of the application more obvious and easy to understand, the application will be further described in detail below with reference to the drawings and specific embodiments.

[0042] Embodiment 1

[0043] As shown in the figure, the embodiment provides a dam leakage electrical anomaly identification method, which comprises: Figures 1-2

[0044] Step 101: obtaining the apparent resistivity image of the dam to be measured.

[0045] Step 102: inputting the apparent resistivity image of the dam to be measured into a dam leakage electrical anomaly identification model to determine the leakage electrical anomaly condition in the dam to be measured; the dam leakage electrical anomaly identification model is obtained by training an end-to-end deep learning model using the labeled apparent resistivity image; the leakage electrical anomaly condition comprises the number of leakage electrical anomaly areas and the position of each leakage electrical anomaly area.

[0046] ​The end-to-end deep learning model comprises: a front end, a rear end and a post-processing module connected in sequence; the front end comprises an encoder; the encoder is obtained by removing the full connection layer in the VGG16 deep convolutional neural network; the rear end comprises a segmentation branch decoder and an embedding branch decoder; the input end of the segmentation branch decoder and the input end of the embedding branch decoder are connected with the output end of the encoder; the output end of the segmentation branch decoder and the output end of the embedding branch decoder are connected with the input end of the post-processing module; and the segmentation branch decoder is used for outputting a leakage electrical abnormal area recognition result.

[0047] The embedding branch decoder is used for outputting a leakage electrical abnormal area mask image.

[0048] The post-processing module is used for performing clustering processing on the leakage electrical abnormal area mask image, combining the clustering result with an image output by the segmentation branch decoder, obtaining a classification result of the leakage electrical abnormal area, determining whether the leakage abnormal resistivity is the same according to the visual features of the abnormal area, extracting the leakage electrical abnormal features and combining the leakage electrical abnormal area mask image to determine the leakage electrical abnormal condition.

[0049] Before step 101, further comprising:

[0050] Step 103: constructing an end-to-end deep learning model.

[0051] Step 104: acquiring multiple visual resistivity history images of the dam.

[0052] Step 105: labeling the leakage electrical abnormal area and the labeled type of the leakage electrical abnormal area in the multiple visual resistivity history images to obtain multiple visual resistivity history labeled images; the labeled type is an elliptical frame or a square frame. Collecting reservoir leakage monitoring information, analyzing the data, and summarizing the resistivity value range interval of the dam leakage abnormal area, the surrounding rock resistivity value range interval, the background value resistivity value, and the relative size and shape of the leakage target area. Referring to the dam leakage measured data characteristics, a resistivity model of a single abnormal body and a double abnormal body is respectively batch-established through an open-source forward simulation program (such as Figure 3 ); the visual resistivity value rho generated in the resistivity model is extracted, and the measurement point position coordinates (X, Z) of the corresponding set electrical method device are determined; the interpolation visualization processing of (X, Z, rho) is performed by using a programming language to batch generate visual resistivity image data sets; in order to make the trained model have better generalization performance, sample data is randomly extracted from the model data in proportion, and finally forms a sample data set and is used as a training set, a validation set and a test set in proportion for network training, validation and testing. The preset model of the model is set as a training label, and the abnormal feature information of the dam model is corresponded.

[0053] Step 106: training an end-to-end deep learning model with the apparent resistivity history image as input and multiple apparent resistivity history labeled images as output to obtain a dam leakage electrical anomaly identification model.

[0054] An end-to-end deep learning algorithm is used to train the apparent resistivity image dataset, and hyperparameter optimization is performed to obtain the best training learning classifier. The specific process is as follows:

[0055] 1) The end-to-end deep learning algorithm directly learns the mapping from the input (apparent resistivity data) to the output (abnormal resistivity model) through a convolutional neural network. The network structure mainly consists of three parts: front end, back end and post-processing part. The front end is the encoding and decoding operation of the input apparent resistivity features, the back end is the use of loss function and gradient descent function to update the model parameters, so that it continuously evolves in the expected direction, and the post-processing part uses the connected and clustering algorithm to circle all the anomalies in the apparent resistivity image, such as Figure 4 .

[0056] a. The front end of the model is composed of a multi-layer convolutional neural network, which is divided into encoding and decoding parts. The encoder uses a VGG16 deep convolutional neural network without full connection layer as a shared feature extractor, which contains 13 convolutional hidden layers and extracts feature maps of different sizes in 5 stages.

[0057] b. The back end includes the configuration optimization function of the network model. The segmentation branch uses a binary cross-entropy loss function L1, which can classify different types of abnormal areas. The prediction probabilities of the target anomaly body and the background are y and 1-y, respectively. The cross-entropy loss function is shown in equation (1):

[0058]

[0059] where y represents the label of the target anomaly body, with positive class as 1 and negative class as 0; y represents the probability of the target anomaly body being predicted as a positive class, the probability of all positive class items in the label being predicted as true, the probability of all negative class items in the label being predicted as true, the greater the product of the two probabilities, the smaller the loss value, and the more consistent the prediction result with the actual value.

[0060] The embedded branch uses a discriminative loss function L emb , which is shown in equation (2):

[0061]

[0062] where ||·|| is the two-norm, [*]+=max(0,*);Lvar For the variance component, m c The cluster center of a single anomaly body, i.e., the average value of all pixel points in the anomaly region, and the radius of the cluster is δ v , C is the number of anomaly regions, N c is the number of pixel points in the cth anomaly region, and ||m c -x i represents the Euclidean distance of the pixel point to the cluster center of the anomaly body. If the distance of the pixel point to the cluster center is greater than the radius of the cluster δ v , it is considered that the pixel point is not within the cluster, and vice versa. The smaller the variance component, the closer the pixel point is to the cluster center, i.e., the smaller the difference between the pixel values in the anomaly region, which is visually manifested as the similar colors of the anomaly region. dist For the distance component, L represents the Euclidean distance between different anomaly cluster centers. If the distance between two cluster centers is less than 2δ d , the two clusters are considered to be of the same class, and vice versa. The smaller the distance component, the greater the distance between the anomaly body clusters, which is visually manifested as the significant difference in color between different anomaly regions. reg The regularization term constrains all cluster centers to prevent the cluster center from being too large and exceeding the pixel threshold. emb is the weighted sum of the first three components, and α, β, γ are used to balance the magnitudes of the three.

[0063] The model training goal is to minimize the loss function value to improve the performance of the model. To achieve efficient computation of the model, the stochastic gradient descent algorithm is used, which can quickly optimize the parameters of the model in large-scale data sets. The gradient value is calculated by the back propagation algorithm and the weight is iteratively updated to effectively find the global optimal solution.

[0064] c. Post-processing part: In order to distinguish different anomaly body regions in dam leakage, the model adds a post-processing method. This method first takes the output of the segmentation branch as a mask and extracts the corresponding anomaly region from the result of the embedding branch. Then the anomaly region is divided into different clusters by the DBSCAN clustering algorithm, and the clustering result is returned to the segmentation branch output, so as to further optimize the segmentation image. The surrounding rock region is presented in black, while other anomaly bodies are presented differently, with different colors representing different resistivities of the anomaly bodies. Figure 5 .

[0065] 2) Network model parameter preliminary setting Batch Size, Learning Rate, Weight Decay, Momentum, etc. The model takes the intersection over union and loss value as the evaluation index of network training. The intersection over union represents the overlap rate of the expected result and the label, that is, the ratio of their intersection to union. The loss value is mainly used to measure the difference between the model prediction value and the true value. After each round of training, the intersection over union and loss value of the model are recorded and plotted into a change curve, such as Figures 6-7 .

[0066] The trained classifier first performs identification test on the test set of the forward model, compares the test results with the prediction model, and tests its accuracy, such as Figure 8 . Then, the application test of the dam measured data is applied to prove the effectiveness and robustness of the network, such as Figures 9-11 .

[0067] Embodiment 2

[0068] In order to perform the method corresponding to the above-mentioned embodiment 1 to realize the corresponding functions and technical effects, the following provides a dam leakage electrical anomaly identification system, comprising:

[0069] The apparent resistivity image acquisition module is configured to acquire an apparent resistivity image of a dam to be measured.

[0070] The leakage electrical anomaly area identification module is configured to input the apparent resistivity image of the dam to be measured into a dam leakage electrical anomaly identification model to determine a leakage electrical anomaly condition in the dam to be measured. The dam leakage electrical anomaly identification model is obtained by training an end-to-end deep learning model using the labeled apparent resistivity image. The leakage electrical anomaly condition includes the number of leakage electrical anomaly areas and the position of each leakage electrical anomaly area.

[0071] Embodiment 3

[0072] The embodiment provides an electronic device, comprising a memory and a processor, the memory is used for storing a computer program, and the processor runs the computer program to make the electronic device execute the dam leakage electrical anomaly identification method of embodiment 1. The memory is a readable storage medium

[0073] In the present specification, each embodiment is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts of each embodiment can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the related parts can be referred to the method part.

[0074] The principles and implementation manners of the present application are described by using specific examples in the present application, and the above examples are only used to help understand the method of the present application and its core idea; meanwhile, for the general technical personnel in the art, the specific implementation manners and application ranges will be changed according to the idea of the present application. In conclusion, the content of the present specification should not be understood as the limitation of the present application.

Claims

1. A dam leakage electrical anomaly identification method, characterized in that, The method comprises the following steps: obtaining a visual resistivity image of a to-be-tested dam; inputting the visual resistivity image of the to-be-tested dam into a dam leakage electrical anomaly recognition model to determine a leakage electrical anomaly condition in the to-be-tested dam; the dam leakage electrical anomaly recognition model is obtained by training an end-to-end deep learning model using labeled visual resistivity images; the leakage electrical anomaly condition comprises a number of leakage electrical anomaly regions and a position of each leakage electrical anomaly region; the end-to-end deep learning model comprises a front end, a back end and a post-processing module connected in sequence; the front end comprises an encoder; the encoder is obtained by removing a fully connected layer in a VGG16 deep convolutional neural network; the back end comprises a segmentation branch decoder and an embedding branch decoder; an input end of the segmentation branch decoder and an input end of the embedding branch decoder are both connected to an output end of the encoder; an output end of the segmentation branch decoder and an output end of the embedding branch decoder are both connected to an input end of the post-processing module; the segmentation branch decoder is used to output a leakage electrical anomaly region recognition result; the embedding branch decoder is used to output a leakage electrical anomaly region mask image; the post-processing module is used to perform clustering processing on the leakage electrical anomaly region mask image, combine a clustering result with an image output by the segmentation branch decoder, obtain a classification result of the leakage electrical anomaly region, and determine whether the leakage electrical anomaly resistivity is the same according to a visual feature of the anomaly region; leakage electrical anomaly features are extracted and combined with the leakage electrical anomaly region mask image to determine the leakage electrical anomaly condition.

2. The method according to claim 1, characterized in that, Before obtaining the visual resistivity image of the to-be-tested dam, the method further comprises the following steps: constructing the end-to-end deep learning model; obtaining a plurality of visual resistivity historical images of the dam; labeling leakage electrical anomaly regions in the plurality of visual resistivity historical images and a labeling type of the leakage electrical anomaly regions to obtain a plurality of visual resistivity historical labeled images; training the end-to-end deep learning model by taking the visual resistivity historical images as input and taking the plurality of visual resistivity historical labeled images as output to obtain the dam leakage electrical anomaly recognition model.

3. The method according to claim 1, characterized in that, The labeling type is an elliptical frame or a square frame.

4. A dam leakage electrical anomaly identification system, characterized in that, The dam leakage electrical anomaly recognition system applies the dam leakage electrical anomaly recognition method according to any one of claims 1-3, and the dam leakage electrical anomaly recognition system comprises: a visual resistivity image acquisition module configured to obtain a visual resistivity image of a to-be-tested dam; a leakage electrical anomaly region identification module configured to input the visual resistivity image of the to-be-tested dam into a dam leakage electrical anomaly recognition model to determine a leakage electrical anomaly condition in the to-be-tested dam; the dam leakage electrical anomaly recognition model is obtained by training an end-to-end deep learning model using labeled visual resistivity images; the leakage electrical anomaly condition comprises a number of leakage electrical anomaly regions and a position of each leakage electrical anomaly region.

5. An electronic device, comprising: The electronic device comprises a memory and a processor, the memory is used for storing a computer program, and the processor runs the computer program to make the electronic device execute the method for identifying the electrical anomaly of dam leakage in any one of claims 1 to 3.

6. The electronic device of claim 5, wherein, The memory is a readable storage medium.

Images