Conductivity imaging method and system based on neural network, and storage medium
Patent Information
- Application Number
- CN202410983270.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-11-08
Smart Images

Figure CN118915169A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of conductivity detection technology, and in particular to a conductivity imaging method, system and storage medium based on a neural network. Background Art
[0002] Among the related technologies, DC conductivity detection is widely used in many fields such as environmental investigation, engineering exploration, hydrological investigation and mining applications. However, due to the inherent limitations of the observation data obtained by conductivity detection, it is often difficult to fully reveal the complex characteristics of the underground. In order to meet the needs of geological interpretation, the inversion method, as one of the important tools of geophysics, has been exploring more accurate and detailed ways to reconstruct underground properties.
[0003] In recent years, with the development of artificial neural networks (ANNs), ANNs have been introduced into conductivity detection. ANNs can directly establish the mapping relationship between apparent conductivity data and the real conductivity model, so that in conductivity inversion, the conductivity characteristics can be quickly inferred. However, the prediction accuracy is not high, which limits the application of this method in complex underground media. Therefore, how to improve the accuracy of conductivity detection in conductivity inversion has become a technical problem that needs to be solved urgently. Summary of the invention
[0004] The present application aims to solve at least one of the technical problems existing in the prior art. To this end, the present application proposes a conductivity imaging method, system and storage medium based on a neural network, which can improve the accuracy of conductivity detection in conductivity inversion by combining neural networks.
[0005] According to the first aspect of the present application, the conductivity imaging method based on a neural network includes:
[0006] According to various types of electrical conductivity models of two-dimensional underground distribution, a Poisson equation satisfied by the potential is established, and the Poisson equation is solved to obtain voltage data between observation electrodes, and a plurality of apparent conductivity color images corresponding to the electrical conductivity model are generated according to the voltage data;
[0007] Acquire a real conductivity model color image and a position model gray image that correspond one-to-one to the apparent conductivity color image;
[0008] Building a first initial neural network and a second initial neural network, wherein the first initial neural network is built based on a mapping relationship between the apparent conductivity color image and the real conductivity color image, and the second initial neural network is built based on a mapping relationship between the apparent conductivity color image and a position model gray image;
[0009] According to a plurality of apparent conductivity color images, corresponding real conductivity model color images, and corresponding position model gray images, respectively training the first initial neural network and the second initial neural network to obtain a first target neural network and a second target neural network;
[0010] Combining the first target neural network and the second target neural network to obtain a combined neural network;
[0011] The conductivity model imaging prediction result based on the apparent conductivity image is determined through the combined neural network.
[0012] According to the conductivity imaging method based on a neural network in the embodiment of the present application, there are at least the following beneficial effects: a first step, according to various types of conductivity models of two-dimensional underground distribution, a Poisson equation satisfied by the potential is established, the Poisson equation is solved, and the voltage data between the observation electrodes is obtained, thereby generating a corresponding plurality of apparent conductivity color images, and obtaining a real conductivity model color image and a position model gray image corresponding to the apparent conductivity color image; a second step, building a first initial neural network and a second initial neural network, wherein the first initial neural network is built based on the mapping relationship between the apparent conductivity color image and the real conductivity color image, and the second initial neural network is built based on the mapping relationship between the apparent conductivity color image and the position model gray image; a third step, according to a plurality of apparent conductivity images and their corresponding real conductivity model images, and their corresponding position model images, the first initial neural network and the second initial neural network are trained respectively to obtain a first target neural network and a second target neural network; a fourth step, the first target neural network and the second target neural network are combined to obtain a combined neural network; a fifth step, through the combined neural network, a conductivity model imaging prediction result based on the apparent conductivity image is determined. The conductivity imaging method based on a neural network of the present application constructs two neural networks, one for learning the mapping from apparent conductivity to a real model, and the other for learning the mapping from apparent conductivity to a position model. The combination of these two neural networks can enhance the inversion capability of the model, thereby enabling more accurate prediction results to be obtained in conductivity inversion. Therefore, the conductivity imaging method of the present application can improve the accuracy of conductivity detection in conductivity inversion by combining neural networks.
[0013] According to some embodiments of the first aspect of the present application, the Poisson equation is:
[0014]
[0015] Among them, σ(S / m) is the conductivity, φ(V) represents the - and r +The potential field generated by the injection of current I0(A) at two power supply electrodes, δ is the Dirac function;
[0016] The step of solving the Poisson equation to obtain voltage data between the observation electrodes includes:
[0017] According to the preset conductivity spatial distribution and potential boundary conditions, the Poisson equation is solved to obtain the potential field;
[0018] The respective potential values are determined at the locations of the observation electrodes according to the potential field, and the voltage data between the observation electrodes are obtained.
[0019] According to some embodiments of the first aspect of the present application, generating a plurality of apparent conductivity color images corresponding to the conductivity model according to the voltage data includes:
[0020] The voltage data is converted into apparent conductivity by a conversion formula;
[0021] According to the apparent conductivity, a plurality of apparent conductivity color images corresponding to the conductivity model are obtained.
[0022] Wherein, the conversion formula is:
[0023]
[0024] σ a (S / m) is the apparent conductivity, k(m) is the geometric coefficient, which depends on the current and the spatial configuration of the electric dipoles in the potential field, and ΔV(V) is the potential difference between the two observation electrodes in the potential field.
[0025] According to some embodiments of the first aspect of the present application, building the first initial neural network and the second initial neural network includes:
[0026] Establishing a first mapping relationship from the apparent conductivity color image to the real conductivity color image;
[0027] According to the first mapping relationship, the first initial neural network is constructed;
[0028] Establishing a second mapping relationship from the apparent conductivity color image to the position model gray image;
[0029] According to the second mapping relationship, the second initial neural network is constructed.
[0030] According to some embodiments of the first aspect of the present application, the first initial neural network and the second initial neural network both adopt an encoder-decoder network architecture, wherein the encoder-decoder network architecture is a U-Net architecture.
[0031] According to some embodiments of the first aspect of the present application, there are multiple types of the real conductivity color images, and the multiple types of the real conductivity color images are obtained by the following steps:
[0032] A background conductivity is randomly selected from a preset conductivity set;
[0033] The abnormal body conductivity of a single rectangular body, the abnormal body conductivity of two rectangular bodies, the abnormal body conductivity of a single inclined body and the abnormal body conductivity of a complex abnormal body are obtained respectively;
[0034] Different types of real conductivity color images are constructed according to the background conductivity and the abnormal body conductivity.
[0035] According to some embodiments of the first aspect of the present application, combining the first target neural network and the second target neural network to obtain a combined neural network includes:
[0036] According to a preset combination formula, the output of the first target neural network and the output of the second target neural network are combined to obtain the combined neural network, wherein the combination formula is:
[0037]
[0038] The first target neural network output is The second target neural network output is The combined neural network output is Z0 and Z1 are The number of points with values 0 and 1 in , i∈N, j∈N, i and j represent the positions on the output image, and N represents the total number of pixels in the output image.
[0039] According to the second aspect of the present application, the conductivity imaging system based on a neural network includes:
[0040] A generation module, used to establish a Poisson equation satisfied by the potential according to various types of conductivity models of two-dimensional underground distribution, and solve the Poisson equation to obtain voltage data between observation electrodes, generate multiple apparent conductivity color images corresponding to the conductivity model according to the voltage data, and obtain real conductivity model color images and position model gray images corresponding to the apparent conductivity color images one by one;
[0041] A building module, used to build a first initial neural network and a second initial neural network, wherein the first initial neural network is built based on a mapping relationship between the apparent conductivity color image and the real conductivity color image, and the second initial neural network is built based on a mapping relationship between the apparent conductivity color image and a position model gray image;
[0042] A training module, for training the first initial neural network and the second initial neural network respectively according to a plurality of apparent conductivity color images, corresponding real conductivity model color images, and corresponding position model gray images, to obtain a first target neural network and a second target neural network;
[0043] A combining module, used for combining the first target neural network and the second target neural network to obtain a combined neural network;
[0044] The prediction module is used to determine the conductivity model imaging prediction result based on the apparent conductivity image through the combined neural network.
[0045] According to the third aspect of the present application, the conductivity imaging system based on a neural network comprises:
[0046] at least one memory;
[0047] at least one processor;
[0048] at least one program;
[0049] The programs are stored in the memory, and the processor executes at least one of the programs to implement the neural network-based conductivity imaging method as described in the embodiment of the first aspect.
[0050] According to the computer-readable storage medium of the fourth aspect embodiment of the present application, the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the neural network-based conductivity imaging method as described in the first aspect embodiment.
[0051] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0052] The present application is further described below with reference to the accompanying drawings and embodiments, wherein:
[0053] Figure 1 A schematic diagram of a flow chart of a conductivity imaging method based on a neural network provided in an embodiment of the present application;
[0054] Figure 2A schematic diagram of apparent conductivity, a real model and a position model provided in an embodiment of the present application;
[0055] Figure 3 A schematic diagram of a combination of two neural networks provided in an embodiment of the present application;
[0056] Figure 4 Schematic diagram of the real conductivity color images of four types of anomalies provided in the embodiments of the present application;
[0057] Figure 5 A schematic diagram of the prediction result images corresponding to the four abnormal body types provided in the embodiment of the present application;
[0058] Figure 6 A schematic diagram of a prediction result image after adding noise provided in an embodiment of the present application;
[0059] Figure 7 A connection diagram of a neural network-based conductivity imaging system provided in one embodiment of the present application;
[0060] Figure 8 A schematic structural diagram of a neural network-based conductivity imaging system provided in another embodiment of the present application.
[0061] Reference numerals:
[0062] Memory 200 and processor 300 . DETAILED DESCRIPTION
[0063] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as limiting the present application.
[0064] It should be noted that although the functional modules are divided in the system diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described can be performed in a different order than the module division in the system or the order in the flowchart. The terms in the specification, claims and the above drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.
[0065] In the description of this application, "several" means more than one, "more" means more than two, "greater than", "less than", "exceed", etc. are understood to exclude the number itself, and "above", "below", "within", etc. are understood to include the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.
[0066] In the description of this application, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in this application based on the specific content of the technical solution.
[0067] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0068] In the field of conductivity detection inversion, a series of nonlinear optimization methods, such as genetic algorithms, simulated annealing algorithms, and particle swarm optimization, have become a way to solve the problem of revealing the characteristics of complex underground media due to their outstanding performance in the field of geophysical inversion [1-4]. However, in recent years, with the introduction of artificial neural networks (ANNs), a new wave of change is taking shape in the field of conductivity detection. The reason why ANNs have attracted much attention is that they can directly establish a mapping relationship between apparent conductivity data and the true conductivity model by repeatedly adjusting network parameters, thereby achieving more accurate and efficient inference of underground characteristics. However, although ANNs have shown exciting results in conductivity inversion, their training process still faces technical difficulties such as slow convergence, insufficient prediction accuracy, and easy overfitting, which limits their wide application in the study of complex underground media.
[0069] Based on the above problems, the present application proposes a combined neural network based on a deep learning inversion method, which realizes the mapping from input to output. First, a neural network is used to train the mapping of apparent conductivity to the real conductivity model. By back propagating and optimizing on the data set, the neural network gradually learns how to accurately map the apparent conductivity to the corresponding real conductivity model. Secondly, another neural network is designed to learn the position mapping of the anomaly. By training on the data set, this network can accurately predict the location of the anomaly from the apparent conductivity. Finally, after the training is completed, the prediction results of the two neural networks are combined to obtain a combined neural network. The combined neural network of the present application can make up for the limitations that may exist in a single network, not only to improve the overall inversion accuracy, but also to obtain more accurate and reliable conductivity inversion results.
[0070] The key point of this application is to establish a mapping between apparent conductivity and model position information, thereby greatly simplifying the complexity of the mapping and improving the accuracy of the inversion of the anomaly position.
[0071] Below, according to Figure 1 The present invention describes a conductivity imaging method based on a neural network according to an embodiment of the present application.
[0072] It is understandable that if Figure 1 As shown, a conductivity imaging method based on a neural network is provided, comprising:
[0073] Step S100, establishing a Poisson equation satisfied by the potential according to various types of conductivity models of underground two-dimensional distribution, solving the Poisson equation to obtain voltage data between observation electrodes, and generating multiple apparent conductivity color images corresponding to the conductivity model according to the voltage data;
[0074] Step S110, obtaining a real conductivity model color image and a position model gray image corresponding one-to-one to the apparent conductivity color image;
[0075] Step S120, building a first initial neural network and a second initial neural network, wherein the first initial neural network is built based on a mapping relationship between an apparent conductivity color image and a true conductivity color image, and the second initial neural network is built based on a mapping relationship between an apparent conductivity color image and a position model gray image;
[0076] Step S130, training the first initial neural network and the second initial neural network respectively according to the plurality of apparent conductivity color images, the corresponding real conductivity model color images, and the corresponding position model gray images, to obtain the first target neural network and the second target neural network;
[0077] Step S140, combining the first target neural network and the second target neural network to obtain a combined neural network;
[0078] Step S150, determining the conductivity model imaging prediction result based on the apparent conductivity image by combining the neural network.
[0079] Therefore, in the first step, according to various types of conductivity models of underground two-dimensional distribution, a Poisson equation satisfied by the potential is established, the Poisson equation is solved, and the voltage data between the observation electrodes is obtained, thereby generating a corresponding plurality of apparent conductivity color images, and obtaining a real conductivity model color image and a position model gray image corresponding to the apparent conductivity color image one by one; in the second step, a first initial neural network and a second initial neural network are constructed, wherein the first initial neural network is constructed based on the mapping relationship between the apparent conductivity color image and the real conductivity color image, and the second initial neural network is constructed based on the mapping relationship between the apparent conductivity color image and the position model gray image; in the third step, according to a plurality of apparent conductivity images and their corresponding real conductivity model images, and their corresponding position model images, the first initial neural network and the second initial neural network are trained respectively to obtain a first target neural network and a second target neural network; in the fourth step, the first target neural network and the second target neural network are combined to obtain a combined neural network; in the fifth step, the conductivity model imaging prediction result based on the apparent conductivity image is determined by the combined neural network. The conductivity imaging method based on a neural network of the present application constructs two neural networks, one for learning the mapping from apparent conductivity to a real model, and the other for learning the mapping from apparent conductivity to a position model. The combination of these two neural networks can enhance the inversion capability of the model, thereby enabling more accurate prediction results to be obtained in conductivity inversion. Therefore, the conductivity imaging method of the present application can improve the accuracy of conductivity detection in conductivity inversion by combining neural networks.
[0080] It can be understood that, based on step S100 , a theoretical apparent conductivity color image can be generated for use in training the combined neural network.
[0081] It should be noted that the observation electrodes are preset and can be selectively set according to actual needs. The voltage data is a collection of potential differences obtained by observing the electrodes at different positions.
[0082] A pixel of 1 in the gray image of the position model indicates an abnormal conductivity.
[0083] The position model gray image is a position mapping of the real conductivity model corresponding to the real conductivity model color image.
[0084] It should be noted that, during the pre-test, the apparent conductivity image obtained based on the observed electrode voltage data is input into the combined neural network to obtain the imaging prediction result corresponding to the actual conductivity.
[0085] It can be understood that Poisson's equation is:
[0086]
[0087] Among them, σ(S / m) is the conductivity, φ(V) represents the - and r + The potential field generated by injecting current I0(A) into the two power supply electrodes, δ is the Dirac function.
[0088] It should be noted that in the field of conductivity sounding, the input and output data are the apparent conductivity data (surface records) and the underground distributed conductivity, respectively. Since the actual distribution of conductivity is difficult to obtain directly through detection, it can usually only be predicted through theoretical inversion deduction. Based on this, by theoretically establishing the Poisson equation satisfied by the potential and using the forward method of numerically solving the Poisson equation, the underground two-dimensional conductivity distribution can be linked to the voltage data obtained by observing the electrode to obtain the apparent conductivity image, which can theoretically generate the input data for training the neural network.
[0089] It should be noted that by solving the Poisson equation, synthetic data with realistic characteristics can be generated, providing the necessary information for the training of machine learning neural networks. These training data are generated by underground conductivity models generated by geostatistical generators, and these models are integrated into the forward modeling to determine the apparent conductivity field under a given acquisition configuration. Although these data are not directly derived from field acquisition, relying on accurate physical models, the apparent conductivity scenarios in real scenes can be simulated, thereby providing representative data samples for training the network. The final training data covers the output underground real conductivity model and the corresponding input apparent conductivity field.
[0090] It should be noted that this method has played a huge role in solving the problem of data acquisition and provides an efficient solution for deep learning inversion methods in the field of conductivity sounding. By combining the solution of the Poisson equation and the geostatistical generator, a real underground conductivity model can be generated and integrated with the actual apparent conductivity measurement to simulate the real scene. This data generation strategy not only overcomes the problem of difficulty in obtaining actual data, but also provides enough real and diverse data samples for network training, thereby enhancing the accuracy and robustness of the deep learning inversion method.
[0091] It should be noted that in order to ensure the quality of the synthetic dataset, the synthetic dataset is divided into three parts: training set, validation set and test set. This division method can ensure that the neural network model is fully considered and evaluated in many aspects. The training set is used to train the neural network, and the validation set is used to adjust the network's hyperparameters to ensure that it will not overfit on the training data while performing well on new data. Finally, the test set is used to evaluate the model's predictive performance on new fields and unseen data.
[0092] For example, the detection area is set to [-50,50]m×[-50,0]m, and a total of 20,000 sets of samples are collected. 21 electrodes are used to collect voltage data through a dipole-dipole electrode arrangement, and then the apparent conductivity is obtained through the conversion formula. The distance between two adjacent electrodes is 5 meters, then it can be obtained Figure 2 The first row gives a schematic diagram of the apparent conductivity data level n=5; by inputting this schematic diagram into the combined neural network, the imaging prediction effect corresponding to the real conductivity model can be obtained.
[0093] It can be understood that the Poisson equation is solved to obtain the voltage data between the observation electrodes, including:
[0094] According to the preset conductivity spatial distribution and potential boundary conditions, the Poisson equation is solved to obtain the potential field;
[0095] The respective potential values are determined at the locations of the observation electrodes according to the potential field, and the voltage data between the observation electrodes are obtained.
[0096] Generate multiple apparent conductivity color images corresponding to the conductivity model based on the voltage data, including:
[0097] The voltage data is converted into apparent conductivity through the conversion formula;
[0098] According to the apparent conductivity, a plurality of apparent conductivity color images corresponding to the conductivity model are obtained.
[0099] The conversion formula is:
[0100]
[0101] σ a (S / m) is the apparent conductivity, k(m) is the geometric coefficient, which depends on the spatial configuration of electric dipoles in the current and potential field, and ΔV(V) is the voltage (potential difference) between the two observation electrodes in the potential field.
[0102] It is understandable that, in some embodiments, k(m) can be determined by looking up a preset mapping table according to the spatial configuration of electric dipoles in the current and potential field. In other embodiments, it can also be manually configured according to the spatial configuration of electric dipoles in the current and potential field.
[0103] It should be noted that for a given conductivity spatial distribution and potential boundary conditions, the potential field can be numerically calculated at any point in the domain, including the surface, using the finite element or finite difference method. Then the voltage data is obtained at the electrode position and converted to the apparent conductivity σ through the conversion formula a (S / m).
[0104] It can be understood that building the first initial neural network and the second initial neural network includes:
[0105] Establishing a mapping relationship from the apparent conductivity color image to the corresponding real conductivity color image;
[0106] According to the above mapping relationship, the U-Net network architecture of encoder-decoder is adopted, the initial network parameters are set, and the first initial neural network is constructed;
[0107] Establishing a mapping relationship from the apparent conductivity color image to the gray image of the corresponding position model;
[0108] According to the above mapping relationship, the encoder-decoder U-Net network architecture is also used to set the initial network parameters and build the second initial neural network.
[0109] It should be noted that the real conductivity color image is a graphical real conductivity model.
[0110] It should be noted that the position model gray image is a graphical position model.
[0111] It should be noted that, first, the apparent conductivity σ is established from the apparent conductivity color image. a To the corresponding conductivity model σ, a first mapping relationship is obtained. At this time, the first mapping relationship reflects the relationship between the apparent conductivity color image and the real conductivity color image. At the same time, an intermediate model s is introduced to represent the position information of the real conductivity model σ, and the relationship between the apparent conductivity σ and the real conductivity model σ is established. a The mapping relationship between the apparent conductivity color image and the position model gray image is obtained by converting the initial parameters into the corresponding position model s. and The initial neural network and
[0112]
[0113] On the basis of synthesizing the real conductivity model and the corresponding apparent conductivity data set, a plurality of apparent conductivity images and their corresponding real conductivity model images, as well as their corresponding position model images are used as training sets to train the first initial neural network. and the second initial neural network Perform training and update network parameters and Get the first target neural network and the second target neural network. Then, combine the output of the first target neural network and the output of the second target neural network To improve the accuracy of conductivity inversion.
[0114] It should be noted that if Figure 2 As shown in the figure, the data-model correspondence of the neural network is shown. From top to bottom, the first row is the apparent conductivity color image, the second row is the real conductivity color image, and the third row is the position model gray image. The position model gray image is obtained by simplifying the real conductivity color image. The value of the abnormal body position (white part) is 1, and the value of other positions (black part) is 0. Neural Network As a basic model, it directly learns the mapping from the apparent conductivity data to the true conductivity model. It is used to learn the mapping from the apparent conductivity data to the position model. It is worth noting that the pixel values of the gray image of the position model only contain two values, namely 0 and 1. Therefore, by introducing the position model, we have successfully reduced the complexity of learning greatly, thereby more effectively capturing the mapping relationship between models.
[0115] It can be understood that both the first initial neural network and the second initial neural network adopt an encoder-decoder network architecture, wherein the encoder-decoder network architecture is a U-Net architecture.
[0116] It can be understood that there are various types of real conductivity color images, and various types of real conductivity color images are obtained by the following steps:
[0117] A background conductivity is randomly selected from a preset conductivity set;
[0118] The abnormal body conductivity of a single rectangular body, the abnormal body conductivity of two rectangular bodies, the abnormal body conductivity of a single inclined body and the abnormal body conductivity of a complex abnormal body are obtained respectively;
[0119] Different types of real conductivity color images are constructed based on the background conductivity and the conductivity of the abnormal body.
[0120] It should be noted that if Figure 3 As shown in , the real conductivity color image contains two parts, background conductivity and abnormal body conductivity. For all types of real conductivity color images, the background conductivity is randomly selected from the set [1 / 100, 1 / 150, 1 / 200, 1 / 250, 1 / 300, 1 / 400] S / m. Figure 3 As shown in the figure, the abnormal body conductivity in the collected 20,000 sets of samples includes the following four types:
[0121] A single rectangular volume (5000 sample pairs) is as follows Figure 3Upper left corner. The length and width of the rectangle vary in the range [4, 12] meters, and the conductivity varies in the range [1 / 10000, 1] S / m.
[0122] Two rectangular bodies (5000 sample pairs) such as Figure 3 Top right corner. The length and width of each rectangle vary in the range [4, 12] meters, and the conductivity varies in the range [1 / 10000, 1] S / m.
[0123] A single tilt (5000 sample pairs) Figure 3 Lower left corner. The width of each rectangle is two to three times its length, varying in the range of [4,7] meters. The vertex of the lower rectangle is located at the midpoint of the base of the higher rectangle, forming an anomaly composed of three rectangles, with conductivity varying in the range of [1 / 10000, 1] S / m.
[0124] Complex anomaly model (5000 sample pairs) Figure 3 Bottom right. It consists of a square and a square with one quarter removed. The width of the square varies in the range [4,12] meters, and the conductivity varies in the range [1 / 10000, 1] S / m.
[0125] At this time, an abnormal body model to be used to generate a real conductivity model is selected from the above four abnormal body models, and the background conductivity is combined with any abnormal body model to obtain a real conductivity model.
[0126] It can be understood that the first target neural network and the second target neural network are combined to obtain a combined neural network, including:
[0127] According to the preset combination formula, the output of the first target neural network and the output of the second target neural network Combine and get the output of the combined neural network The combination formula is:
[0128]
[0129] Z0 and Z1 are The number of points with values 0 and 1 in , i∈N, j∈N, i and j represent the positions on the output image, and N represents the total number of pixels in the output image.
[0130] It should be noted that if Figure 4 As shown, Figure 4 The upper left image is the output of the first target neural network, which is a color image and is recorded as Figure 4 The image in the lower left is the output of the second target neural network, which is a grayscale image and is recorded as calculate Zhongyu The mean of the corresponding position of the white area (the area with a value of 1), that is, The mean value inside the red dashed box is assigned to White area. Similarly, calculate Zhongyu The mean of the corresponding position in the black area (the area with a value of 0), that is, the mean outside the red dotted box, is assigned to Black area. The result is as follows Figure 4 As shown on the right, it is the output image obtained by combining the neural networks, which has a certain improvement in boundary accuracy.
[0131] It should be noted that the input of the first target neural network is the apparent conductivity color image, and the output is the conductivity model image, that is, The input of the second target neural network is also the apparent conductivity color image, and the output is the location information of the conductivity model, that is, Grayscale image of information.
[0132] It should be noted that for the first target neural network, the following loss function is used:
[0133]
[0134] Where σ is the real conductivity, is the network prediction result, the first term of L is the mean square error, N is the total number of pixels, which is used as a normalization term, and the second term of L is a smoothing term used to improve the overfitting problem and enhance the stability of the inversion, and α is the smoothing factor.
[0135] It should be noted that a sigmoid function is added to the output layer of the second target neural network U-Net architecture, and the loss function of the second target neural network is set to binary cross entropy, which is defined as follows:
[0136]
[0137] Where s is a real value (either 0 or 1), is the prediction result of the network (value is in (0,1)), L is the loss function, and N is the total number of pixels, which is used as a normalization term. Since the value of the position model is relatively simple, consisting of 0 and 1, the prediction result of the network can be Set the threshold to 0.5 to convert it to 0 or 1.
[0138] It should be noted that in the process of averaging, due to the possible differences in the selection of regions, some points with outliers may be included. Figure 4In the example above, the area inside the red dashed box may contain some blue parts that should be excluded. To address this problem, some maximum and minimum values can be removed before averaging to further improve the accuracy of the results.
[0139] It should be noted that this example uses PyTorch as the framework for building the neural network, uses the Adam optimizer, and takes the smoothing factor α as 0.2. The samples are divided into training sets, validation sets, and test sets according to the ratio of 8:1:1. When training the neural network, the initial learning rate is set to 0.001. After each epoch, the learning rate is gradually reduced by a factor of 0.8. The entire training process takes a total of 15 epochs.
[0140] It should be noted that if Figure 5 As shown in Figure 1, this figure compares the inversion performance of the first neural network (traditional U-Net) and the combined neural network (combined U-Net). The two networks were trained using the same dataset and the same number of iterations. We randomly selected four results from the test set and Figure 5 In the Figure 5 In the figure, the first column shows the input data color image, the second column shows the U-Net output data color image, the third column shows the combined U-Net output data color image, and the fourth column shows the true conductivity model color image. Figure 5 It can be seen that the apparent conductivity data image in the first column corresponds to the real conductivity model image in the last column. In addition, both U-Net and combined U-Net are able to predict the location of the anomaly. However, by comparison, it can be clearly seen that the inversion results obtained by the combined U-Net are more accurate. Table 1 lists the inversion accuracy of U-Net and the combined neural network. It can be seen that the combined neural network has an accuracy improvement of 6.5% over the traditional U-Net method.
[0141] Table 1. Accuracy of U-Net and combined neural network output results
[0142]
[0143] Therefore, it can be shown that the method proposed in this application is of great help in improving the inversion accuracy.
[0144] It should be noted that 5% and 10% Gaussian noise can also be added to illustrate the accuracy of the prediction results of this solution. Figure 6As shown, four apparent conductivity model images are shown, and the inversion results under 5% and 10% noise are shown when different types of complex anomalies are added. From left to right, the first column is the true model, the second column is the inversion result without noise, the third column is the inversion result with 5% noise, and the fourth column is the inversion result with 10% noise. It can be seen from the data that the combined neural network proposed in this application can still maintain good inversion results and improve the inversion accuracy of noisy data.
[0145] It is understandable that if Figure 7 As shown, the present application also proposes a conductivity imaging system based on a neural network, comprising:
[0146] A generation module is used to establish a Poisson equation satisfied by the potential according to various types of conductivity models of underground two-dimensional distribution, solve the Poisson equation, obtain voltage data between observation electrodes, and generate multiple apparent conductivity color images corresponding to the conductivity model according to the voltage data;
[0147] A building module, used to build a first initial neural network and a second initial neural network, wherein the first initial neural network is built based on a mapping relationship between an apparent conductivity color image and a real conductivity color image, and the second initial neural network is built based on a mapping relationship between an apparent conductivity color image and a position model gray image;
[0148] A training module, for training the first initial neural network and the second initial neural network respectively according to a plurality of apparent conductivity color images, corresponding real conductivity model color images, and corresponding position model gray images, to obtain a first target neural network and a second target neural network;
[0149] A combination module, used for combining the first target neural network and the second target neural network to obtain a combined neural network;
[0150] The prediction module is used to determine the conductivity model imaging prediction result based on the apparent conductivity image by combining neural networks.
[0151] Refer to the following Figure 8 A conductivity imaging system based on a neural network according to an embodiment of the present application is described.
[0152] It is understandable that if Figure 8 As shown, the conductivity imaging system based on neural network includes:
[0153] at least one memory 200;
[0154] at least one processor 300;
[0155] at least one program;
[0156] The programs are stored in the memory 200, and the processor 300 executes at least one program to implement the above-mentioned conductivity imaging method based on neural network. Figure 8 Take a processor 300 as an example.
[0157] The processor 300 and the memory 200 may be connected via a bus or other means. Figure 8 Take the example of connecting via a bus.
[0158] The memory 200, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer executable programs and signals, such as program instructions / signals corresponding to the neural network-based conductivity imaging system in the embodiment of the present application. The processor 300 executes various functional applications and data processing by running the non-transitory software programs, instructions and signals stored in the memory 200, that is, the neural network-based conductivity imaging method of the above method embodiment is implemented.
[0159] The memory 200 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store relevant data of the above-mentioned conductivity imaging method based on a neural network, etc. In addition, the memory 200 may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 200 may optionally include a memory remotely arranged relative to the processor 300, and these remote memories may be connected to the conductivity imaging system based on a neural network via a network. Examples of the above-mentioned network include, but are not limited to, the Internet of Things, a software-defined network, a sensor network, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0160] One or more signals are stored in the memory 200, and when executed by one or more processors 300, the conductivity imaging method based on a neural network in any of the above method embodiments is executed. Figure 1 The method in .
[0161] Refer to the following Figure 8 A computer-readable storage medium according to an embodiment of the present application is described.
[0162] like Figure 8 As shown, the computer-readable storage medium stores computer-executable instructions, which are executed by one or more processors 300, for example, Figure 8The one or more processors 300 may execute the neural network-based conductivity imaging method in the above method embodiment. Figure 1 The method in .
[0163] The system embodiments described above are merely illustrative, in which the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0164] Through the description of the above embodiments, it can be understood by those skilled in the art that all or some steps and systems in the above disclosed methods can be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, and the computer-readable medium can include computer storage media and communication media. As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information. Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk or other optical disk storage, magnetic cassette, magnetic tape, disk storage or other magnetic storage device, or any other medium that can be used to store desired information and can be accessed by a computer. In addition, it is known to those skilled in the art that communication media generally contain computer-readable signals, data structures, program modules or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium. The embodiments of the present application are described in detail above in conjunction with the accompanying drawings, but the present application is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary technicians in the relevant technical field without departing from the purpose of the present application. In addition, the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.
Claims
1. A conductivity imaging method based on a neural network, characterized in that: include: According to various types of electrical conductivity models of two-dimensional underground distribution, a Poisson equation satisfied by the potential is established, and the Poisson equation is solved to obtain voltage data between observation electrodes, and a plurality of apparent conductivity color images corresponding to the electrical conductivity model are generated according to the voltage data; Acquire a real conductivity model color image and a position model gray image that correspond one-to-one to the apparent conductivity color image; Building a first initial neural network and a second initial neural network, wherein the first initial neural network is built based on a mapping relationship between the apparent conductivity color image and the real conductivity color image, and the second initial neural network is built based on a mapping relationship between the apparent conductivity color image and a position model gray image; According to a plurality of apparent conductivity color images, corresponding real conductivity model color images, and corresponding position model gray images, respectively training the first initial neural network and the second initial neural network to obtain a first target neural network and a second target neural network; Combining the first target neural network and the second target neural network to obtain a combined neural network; The conductivity model imaging prediction result based on the apparent conductivity image is determined through the combined neural network.
2. The conductivity imaging method based on neural network according to claim 1, characterized in that: The Poisson equation is: Among them, σ(S / m) is the conductivity, φ(V) represents the - and r + The potential field generated by the injection of current I0(A) at two power supply electrodes, δ is the Dirac function; The step of solving the Poisson equation to obtain voltage data between the observation electrodes includes: According to the preset conductivity spatial distribution and potential boundary conditions, the Poisson equation is solved to obtain the potential field; The respective potential values are determined at the locations of the observation electrodes according to the potential field, and the voltage data between the observation electrodes are obtained.
3. The conductivity imaging method based on neural network according to claim 2, characterized in that: Generating a plurality of apparent conductivity color images corresponding to the conductivity model according to the voltage data, comprising: The voltage data is converted into apparent conductivity by a conversion formula; According to the apparent conductivity, a plurality of apparent conductivity color images corresponding to the conductivity model are obtained; Wherein, the conversion formula is: σ a (S / m) is the apparent conductivity, k(m) is the geometric coefficient, which depends on the current and the spatial configuration of the electric dipoles in the potential field, and ΔV(V) is the potential difference between the two observation electrodes in the potential field.
4. The conductivity imaging method based on neural network according to claim 1, characterized in that: The constructing of the first initial neural network and the second initial neural network comprises: Establishing a first mapping relationship from the apparent conductivity color image to the real conductivity color image; According to the first mapping relationship, the first initial neural network is constructed; Establishing a second mapping relationship from the apparent conductivity color image to the position model gray image; According to the second mapping relationship, the second initial neural network is constructed.
5. The conductivity imaging method based on neural network according to claim 1, characterized in that: The first initial neural network and the second initial neural network both adopt an encoder-decoder network architecture, wherein the encoder-decoder network architecture is a U-Net architecture.
6. The method for conductivity imaging based on a neural network according to claim 5, characterized in that: There are multiple types of the real conductivity color images, and the multiple types of the real conductivity color images are obtained by the following steps: A background conductivity is randomly selected from a preset conductivity set; The abnormal body conductivity of a single rectangular body, the abnormal body conductivity of two rectangular bodies, the abnormal body conductivity of a single inclined body and the abnormal body conductivity of a complex abnormal body are obtained respectively; Different types of real conductivity color images are constructed according to the background conductivity and the abnormal body conductivity.
7. The method for conductivity imaging based on a neural network according to claim 1, characterized in that: The step of combining the first target neural network and the second target neural network to obtain a combined neural network includes: According to a preset combination formula, the output of the first target neural network and the output of the second target neural network are combined to obtain the combined neural network, wherein the combination formula is: The first target neural network output is The second target neural network output is The combined neural network output is Z0 and Z1 are The number of points with values 0 and 1 in , i∈N, j∈N, i and j represent the positions on the output image, and N represents the total number of pixels in the output image.
8. A conductivity imaging system based on a neural network, characterized in that: include: A generation module, for establishing a Poisson equation satisfied by the potential according to various types of conductivity models of two-dimensional underground distribution, solving the Poisson equation to obtain voltage data between observation electrodes, and generating a plurality of apparent conductivity color images corresponding to the conductivity model according to the voltage data; A building module, used to build a first initial neural network and a second initial neural network, wherein the first initial neural network is built based on a mapping relationship between the apparent conductivity color image and the real conductivity color image, and the second initial neural network is built based on a mapping relationship between the apparent conductivity color image and a position model gray image; A training module, for training the first initial neural network and the second initial neural network respectively according to a plurality of apparent conductivity color images, corresponding real conductivity model color images, and corresponding position model gray images, to obtain a first target neural network and a second target neural network; A combining module, used for combining the first target neural network and the second target neural network to obtain a combined neural network; The prediction module is used to determine the conductivity model imaging prediction result based on the apparent conductivity image through the combined neural network.
9. A conductivity imaging system based on a neural network, characterized in that: include: at least one memory; at least one processor; at least one program; The programs are stored in the memory, and the processor executes at least one of the programs to implement the neural network-based conductivity imaging method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the neural network-based conductivity imaging method according to any one of claims 1 to 7.
