EIT imaging system and imaging method based on semi-inverse problem solution, and storage medium

Abstract

The invention provides an EIT imaging system based on semi-inverse problem solving, an imaging method and a storage medium, and belongs to the field of medical imaging, and the technical key points are as follows: a storage module is used for storing a boundary voltage (difference) matrix; the first solving module establishes a mapping relation from a boundary voltage (difference) matrix to an internal potential (difference) matrix; the second solving module establishes a mapping relation from an internal potential (difference) matrix to a node conductivity (difference) matrix. According to the technical scheme, full inverse problem solving can be converted into semi-inverse problem solving, the node conductivity (difference) solving precision is improved (the fault-tolerant capability for abnormal electrode signals can be improved), the method is suitable for EIT imaging when the field domain to be measured is an irregular graph, and the EIT imaging precision of tissues such as the brain, the lung and the abdomen can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical imaging and deep learning, and specifically relates to an EIT imaging system, an imaging method, and a storage medium based on solving a semi-inverse problem. Background Art

[0002] EIT (Electrical Impedance Tomography) is a new type of electrical detection technology with the advantages of low cost, no radiation, non-invasiveness and real-time monitoring. It is very suitable for medical imaging and biological research.

[0003] For EIT imaging, the technical difficulty lies in how to obtain the boundary voltage difference data of the field to be measured into a relative conductivity matrix.

[0004] For the above problems, there are two main technical routes:

[0005] The first technical route: formula method. For example, CN115177234B: σ=(S norm T S norm +k t I+k n D) -1 S norm T △V norm T First, the normalized sensitivity matrix S is obtained by using prior knowledge norm , and then directly solve the relative conductivity matrix σ using the above formula. The advantage of this method is that the imaging speed is fast. The disadvantage is that this method depends on S norm It is suitable for detection imaging of regular patterns (such as breast) in the field to be tested. When the field to be tested is an irregular pattern, the imaging effect is not good.

[0006] The second technical route: neural network. For example, in CN109598768A, the common neural model design is: the input layer is the boundary voltage matrix (or boundary voltage difference matrix), and the output layer is the relative conductivity matrix (or relative conductivity difference matrix).

[0007] The application of neural networks to solve the image reconstruction problem of EIT has two main advantages: (1) it avoids the shape problem of the field to be measured. (2) it can image absolute images as well as differential images.

[0008] However, when applying neural networks, the existing technologies generally (e.g., CN117274413B, CN109859285B) adopt a solution mode from the boundary voltage (difference) matrix to the relative conductivity (difference) matrix, which essentially belongs to the solution of a full inverse problem. And the solution of the full inverse problem brings the following problems: the neural network has insufficient fault tolerance. For example, when there are abnormal signals in the input boundary voltage (difference) (such as poor electrode contact or detachment), the calculated conductivity error is relatively large. Summary of the Invention

[0009] The purpose of the present invention is to solve the above problems existing in the prior art, and provide an EIT imaging system, imaging method, and storage medium based on the solution of semi-inverse problems.

[0010] The technical solution of this application lies in:

[0011] An EIT imaging system based on the solution of semi-inverse problems, which includes: a storage module, a first solution module, and a second solution module;

[0012] The storage module is used to store the boundary voltage matrix;

[0013] The first solution module establishes a mapping relationship from the boundary voltage matrix to the internal electric potential matrix;

[0014] The second solution module establishes a mapping relationship from the internal electric potential matrix to the node conductivity matrix;

[0015] Or,

[0016] The storage module is used to store the boundary voltage difference matrix;

[0017] The first solution module establishes a mapping relationship from the boundary voltage difference matrix to the internal electric potential difference matrix;

[0018] The second solution module establishes a mapping relationship from the internal electric potential difference matrix to the node conductivity difference matrix.

[0019] Furthermore, both the first solution module and the second solution module are solution modules constructed by neural networks.

[0020] Furthermore, the input layer of the first solution module is the boundary voltage matrix of the field to be measured, and the output layer is the internal electric potential matrix corresponding to the network nodes of the finite element model of the field to be measured;

[0021] The input layer of the second solution module is the internal electric potential matrix corresponding to the network nodes of the finite element model of the field to be measured, and the output layer is the node conductivity matrix;

[0022] Or,

[0023] The input layer of the first solution module is the boundary voltage difference matrix of the field to be measured, and the output layer is the internal potential difference matrix corresponding to the network nodes of the finite element model of the field to be measured;

[0024] The input layer of the second solution module is the internal potential difference matrix corresponding to the network nodes of the finite element model of the field to be measured, and the output layer is the node conductivity difference matrix.

[0025] An EIT imaging method based on the solution of semi-inverse problems, when used for absolute imaging, includes the following steps:

[0026] S100. Construct a training set; each training set includes: a boundary voltage matrix V, a node potential matrix u, and a node conductivity matrix σ;

[0027] S200. Construct a neural network model of the first solution module, which includes:

[0028] Input layer: boundary voltage matrix V;

[0029] Output layer: node potential matrix u;

[0030] Intermediate layer;

[0031] S300. Construct a neural network model of the second solution module, which includes:

[0032] Input layer: node potential matrix u;

[0033] Output layer: node conductivity matrix σ;

[0034] Intermediate layer;

[0035] S400. Determine the loss functions of the first solution module and the second solution module, and then train the first solution module and the second solution module;

[0036] S500. Read the boundary voltage matrix of the field to be measured, then sequentially obtain the corresponding node potential matrix and node conductivity matrix, and perform imaging according to the node conductivity matrix.

[0037] Furthermore, the loss function of the second solution module adopts:

[0038]

[0039] σ d0 represents the true value of the conductivity of the d-th node, and σ d represents the predicted value of the conductivity of the d-th node; represents the relevant parameters; N represents the number of nodes of the finite element model of the field to be measured; α represents the weight coefficient; represents the L2 regularization factor.

[0040] An EIT imaging method based on solving semi-inverse problems, when used for differential imaging, includes the following steps:

[0041] S100, construct a training set; each training set includes: boundary voltage difference matrix △V, node potential difference matrix △u, node conductivity difference matrix △σ;

[0042] S200, construct a neural network model for the first solution module, which includes:

[0043] Input layer: boundary voltage difference matrix △V;

[0044] Output layer: node potential difference matrix △u;

[0045] Intermediate layer;

[0046] S300, construct a neural network model for the second solution module, which includes:

[0047] Input layer: node potential difference matrix △u;

[0048] Output layer: node conductivity difference matrix △σ;

[0049] Intermediate layer;

[0050] S400, determine the loss functions of the first solution module and the second solution module, and then train the first solution module and the second solution module;

[0051] S500, read the boundary voltage difference matrix of the field to be measured, then sequentially obtain the corresponding node potential difference matrix and node conductivity difference matrix, and perform imaging according to the node conductivity difference matrix.

[0052] An EIT imaging method based on solving semi-inverse problems, when used for absolute imaging, includes the following steps:

[0053] S100, construct a training set; each training set includes: boundary voltage matrix V, node potential matrix u, node conductivity matrix σ;

[0054] S200, construct a neural network model for the first solution module, which includes:

[0055] Input layer: boundary voltage matrix V;

[0056] Output layer: node potential matrix u;

[0057] Intermediate layer;

[0058] S300, construct a neural network model for the second solution module, which includes:

[0059] Input layer: node potential matrix u;

[0060] Output layer: node conductivity matrix σ;

[0061] Intermediate layer;

[0062] S400, determine the loss functions of the first solution module and the second solution module, and then train the first solution module and the second solution module;

[0063] Among them, the loss function of the second solution module adopts:

[0064]

[0065] σ d0 represents the true value of the conductivity of the d-th node, and σ d represents the predicted value of the conductivity of the d-th node; u d represents the true value of the electric potential of the d-th node;

[0066] σ b-边界 、u b-边界 、n b-边界 represent the predicted value of the conductivity, the true value of the electric potential, and the normal vector of the b-th node in the boundary nodes of the finite element model of the field to be measured; ▽ represents the solution of divergence, and ▽ represents the solution of gradient;

[0067] are all related parameters; N represents the number of nodes in the finite element model of the field to be measured, and n represents the number of boundary nodes in the finite element model of the field to be measured; α, γ, and δ all represent weight coefficients; represents the L2 regularization factor;

[0068] S500, read the boundary voltage matrix of the field to be measured, and then sequentially obtain the corresponding node electric potential matrix and node conductivity matrix, and perform imaging according to the node conductivity matrix;

[0069] The so-called true value comes from the training set, and the so-called predicted value comes from the solution result of the neural network of the second solution module.

[0070] Furthermore, step S100 further includes: obtaining a node network matrix and calculating a boundary node loss parameter calculation factor matrix and an elliptic equation loss parameter calculation factor matrix according to the node network matrix;

[0071] The elliptic equation loss parameter calculation factor matrix includes the elliptic equation loss parameter calculation factors corresponding to the N nodes of the finite element model of the field to be measured. Among them, the elliptic equation loss parameter calculation factor corresponding to any d-th node is ▽u d ; u d represents the electric potential of the d-th node, and ▽ represents the solution of gradient (▽u d is essentially a vector).

[0072] The boundary node loss parameter calculation factor matrix includes boundary node loss parameter calculation factors corresponding to n boundary nodes; the calculation factor of the loss parameter of any b-th boundary node is:

[0073] where u b-边界 represents the electric potential of the b-th node in the boundary nodes, and n b-边界 represents the normal vector of the b-th node in the boundary nodes.

[0074] A storage medium stores a program capable of executing the EIT imaging method as described above.

[0075] The advantages of the technical solution of the present invention are mainly reflected in:

[0076] (1) This application converts "boundary voltage (difference) ~ node conductivity (difference)" into "boundary voltage (difference) ~ node electric potential (difference) ~ node conductivity (difference)", so that the solution of the full inverse problem can be converted into the solution of a semi-inverse problem, improving the solution accuracy of the node conductivity (difference) (able to improve the fault tolerance ability for abnormal electrode signals), applicable to EIT imaging when the field to be measured is an irregular graph, and can improve the EIT imaging accuracy of tissues such as the brain, lungs, and abdomen.

[0077] (2) This application uses the PINN method to reconstruct the conductivity. By constraining the solution process through the elliptic equation and the boundary constraint equation, the underdetermination of the EIT problem solution is reduced, the accuracy of the reconstructed conductivity is guaranteed, and the quality of the reconstructed image is effectively improved. At the same time, selecting the node electric potential as an intermediate parameter is also a key step in using the PINN method to solve the EIT problem.

[0078] The EIT solution combined with PINN is reflected in that when performing absolute imaging, the loss function of the second solution module adopts:

[0079]

[0080] (3) The technical solution of this application is applicable to two-dimensional imaging and also applicable to three-dimensional imaging. Description of the Drawings

[0081] The following further describes the present application in detail with reference to the embodiments in the drawings, but does not constitute any limitation to the present application.

[0082] Figure 1 It is the architecture design diagram (absolute imaging) of the solution model in the present invention.

[0083] Figure 2 It is the solution schematic diagram of ▽u5.

[0084] Figure 3 It is Schematic diagram for solution. Detailed implementation manners

[0085] The objectives, advantages and features of the present invention will be explained through non - restrictive descriptions of the following preferred embodiments. These embodiments are only typical examples of applying the technical solutions of the present invention, and any technical solutions formed by equivalent replacement or equivalent transformation fall within the scope of protection required by the present invention.

[0086] <Example 1>

[0087] In order to improve the prediction accuracy of the neural network, the full inverse problem is converted into a semi - full inverse problem. The basic solution of this application is as follows:

[0088] For absolute imaging:

[0089] An EIT imaging system, which includes: a storage module, an internal electric potential matrix solving module, and a node conductivity solving module;

[0090] The storage module is used to store the boundary voltage matrix; the internal electric potential matrix solving module establishes a mapping relationship from the boundary voltage matrix to the internal electric potential matrix; the node conductivity solving module establishes a mapping relationship from the internal electric potential matrix to the node conductivity matrix.

[0091] For differential imaging:

[0092] An EIT imaging system, which includes: a storage module, an internal electric potential difference matrix solving module, and a node conductivity difference matrix solving module;

[0093] The storage module is used to store the boundary voltage difference matrix; the internal electric potential difference matrix solving module establishes a mapping relationship from the boundary voltage difference matrix to the internal electric potential difference matrix; the node conductivity difference matrix solving module establishes a mapping relationship from the internal electric potential difference matrix to the node conductivity difference matrix.

[0094] For the internal electric potential matrix solving module / internal electric potential difference matrix solving module, and the node conductivity / node conductivity difference matrix solving module, a convolutional neural network is used (it should be noted that it is feasible to use a suitable neural network structure).

[0095] Internal electric potential matrix solving module:

[0096] 1) Input layer: the boundary voltage matrix of the field to be measured;

[0097] 2) Output layer: the internal electric potential matrix corresponding to the nodes of the finite - element model of the field to be measured;

[0098] 3) Intermediate layer, which will not be elaborated here (the network structure of CN109598768B can be used).

[0099] Internal potential difference matrix solving module:

[0100] 1) Input layer: The boundary voltage difference matrix of the field to be measured;

[0101] 2) Output layer: The internal potential difference matrix corresponding to the nodes of the finite element model of the field to be measured;

[0102] 3) Intermediate layer, which will not be elaborated here (the network structure of CN109598768B can be adopted).

[0103] Node conductivity matrix solving module:

[0104] 1) Input layer: The internal potential matrix corresponding to the nodes of the finite element model of the field to be measured;

[0105] 2) Output layer: The node conductivity matrix corresponding to the nodes of the finite element model of the field to be measured;

[0106] 3) Intermediate layer, which will not be elaborated here (the network structure of CN109598768B can be adopted).

[0107] Node conductivity difference matrix solving module:

[0108] 1) Input layer: The internal potential difference matrix corresponding to the nodes of the finite element model of the field to be measured;

[0109] 2) Output layer: The node conductivity difference matrix corresponding to the nodes of the finite element model of the field to be measured;

[0110] 3) Intermediate layer, which will not be elaborated here (the network structure of CN109598768B can be adopted).

[0111] When constructing the training sets of the first solving module and the second solving module, they can be obtained by using Comsol software. That is, after importing the finite element model of the field to be measured into Comsol software, set the positions of E electrodes, and record the boundary voltage, node potential, node coordinate set, and node conductivity of this round of excitation (for example: adjacent excitation and adjacent measurement mode) in each round of excitation.

[0112] When performing differential imaging, the boundary voltage difference, node potential difference, and node conductivity difference are essentially obtained by performing differential solution on the corresponding boundary voltage, node potential, and node conductivity (the differential is generally frequency difference or time difference, such as: Yang Yuxiang, Bai Shizhan, Lin Haijun, etc. Design of a multi-frequency electrical impedance tomography system based on multisine excitation and full-cycle sampling [J]. Acta Physica Sinica, 2022(005): 073).

[0113] <Loss functions of the first solving module and the second solving module>

[0114] When using MSE, it is given by the following formula:

[0115] y i represents the predicted value of the sample (i.e., the output value of the model); represents the true value of the sample (i.e., the data in the training set), and N represents the number of nodes in the finite element model of the field to be measured.

[0116] If considering preventing overfitting of the network, the following can be adopted:

[0117] represents the L2 regularization factor (β represents the regularization coefficient, is the L2 norm).

[0118] When performing absolute imaging, the loss function of the second solution module adopts:

[0119]

[0120] σ d0 represents the true value of the conductivity of the d-th node (i.e., the data in the training set), and σ d represents the predicted value of the conductivity of the d-th node.

[0121] The EIT method is mainly used for measuring human tissues and biological tissues, and its conductivity is generally between 0.01 and 1 S / m. Therefore, relevant constraints need to be imposed in the loss function; are relevant parameters.

[0122] <Example 2>

[0123] The model proposed in the technical solution of Example 1 mainly relies on data-driven. The Physics-Informed Neural Networks (PINN) is a new neural network framework that combines physical laws with deep learning.

[0124] Example 2 introduces the idea of PINN into the solution of Example 1.

[0125] <2.1. Physical Laws of EIT Theory>

[0126] For EIT, it satisfies the following physical laws (Reference: Tian Haiyan. Research on the Theory and Practice of Electrical Impedance Tomography Technology [D]. Chongqing University. 2025):

[0127] (1) The field region is denoted as Ω, and the distribution equation (elliptic equation) of the electric potential u of the internal nodes is:

[0128] -▽·(σ▽u) = 0 in Ω (takes effect in Ω)

[0129] Among them, ▽ represents divergence calculation, ▽ represents gradient calculation, σ represents conductivity, -σ▽u represents current density, and the minus sign indicates that current flows from high electric potential to low electric potential. Therefore, the overall formula -▽(σ▽u) = 0 means that within the entire field domain, under the action of electric potential u and conductivity σ, the current is in a conserved state, that is, there is no charge accumulation within the entire field domain.

[0130] (2) Neumann boundary condition:

[0131]

[0132] Among them, the Neumann boundary condition describes the flux and gradient on the boundary. n represents the normal vector on the boundary of the measurement region Ω, and g represents the magnitude of the current density on the boundary.

[0133] (3) Dirichlet boundary condition:

[0134]

[0135] The Dirichlet boundary condition describes the fixed state on the boundary. f represents the known value on the boundary, and u represents the value to be solved.

[0136] (4) Operator from Neumann boundary condition to Dirichlet boundary condition:

[0137]

[0138] Among them, ∧ σ represents the mapping relationship of applying current g to boundary voltage f, providing a feasible mathematical basis for the reconstruction of conductivity σ.

[0139] <2.2. Embodiment of the Physical Laws of EIT Theory in Neural Networks>

[0140] The embodiment of the physical laws of EIT in neural networks in 2.1 is mainly reflected in the loss function of the node second solution module.

[0141] When performing absolute imaging, the loss function of the second solution module is adopted as:

[0142]

[0143] The first term is the MSE calculation, and the calculation range is the conductivity of all nodes within the measurement region and on the boundary. σ d represents the predicted value of the conductivity of the i-th node, and σ d0 represents the true value of the conductivity of the i-th node (i.e., the data in the training set).

[0144] The second item is the non - negativity constraint. The EIT method is mainly used for measuring human tissues and biological tissues, and their conductivity is generally between 0.01 and 1 S / m. Therefore, relevant constraints should be imposed in the loss function; are relevant parameters.

[0145] The third item is the regularization term, denotes the L2 regularization factor (β represents the regularization coefficient, is the L2 norm).

[0146] The fourth item is the elliptical constraint loss, which is related to the internal electric potential u. Among them, Regarding the computer solution of gradients and divergences (the finite - difference method approximates partial derivatives. At boundary points, central differences cannot be used, and forward or backward differences can be used), refer to: https: / / blog.51cto.com / u_16213425 / 13171979;

[0147] As Figure 2 shown, for two - dimensional imaging:

[0148]

[0149] x i and y i and u i are the coordinates and electric potential of the i - th point grid.

[0150] Correspondingly, it can be solved to obtain:

[0151]

[0152] The fifth item represents the boundary - condition constraint. Among them, (only calculate boundary nodes), which is in the same form as the Neumann boundary condition. Since the total current flux on the boundary of the measurement area is zero, so g = 0. n b represents the vector formed by the current boundary point and its nearest normal node.

[0153] The loss function is used to train the second solution module. Therefore, u d uses the data of the node - electric - potential matrix u in the training set.

[0154] As Figure 3 shown, for two - dimensional imaging:

[0155]

[0156] d 78 represents the distance between node 7 and node 8.

[0157] Given the boundary nodes, The mathematical calculation method is as follows:

[0158] (1) Calculate the normal vector of each boundary node

[0159] For each boundary point, calculate its normal vector n 边界 .

[0160] In the two-dimensional case, the normal vector can be obtained by rotating the tangent vector of the boundary curve by 90 degrees.

[0161] In the three-dimensional case, the normal vector can be obtained by local surface fitting (such as plane fitting).

[0162] (2) Numerically calculate the derivative

[0163] Use the finite difference method or interpolation method to calculate the derivative of the electric potential with respect to the normal vector.

[0164] For any boundary point p, take a small step size △h along the normal vector direction to obtain the adjacent point p + △h.

[0165] Calculate the difference in the electric potential at points p and p + m△h, and approximate the derivative:

[0166]

[0167] (3) Interpolate the electric potential of the adjacent point

[0168] If the adjacent point p + △h is not a known point, an interpolation method (such as linear interpolation, bilinear interpolation, or cubic spline interpolation) needs to be used to calculate the electric potential at this point.

[0169] The physical quantities represented by the symbols in this application are explained as follows:

[0170] σ d0 Represents the true value of the conductivity of the d-th node.

[0171] σ d Represents the predicted value of the conductivity of the d-th node.

[0172] u d Represents the electric potential of the d-th node.

[0173] N represents the number of nodes in the finite element model of the field to be measured.

[0174] n represents the number of boundary nodes in the finite element model of the field to be measured.

[0175] σ b-边界 Represents the conductivity of the b-th node among the boundary nodes.

[0176] u b-边界 Represents the electric potential of the b-th node among the boundary nodes.

[0177] n b-边界 represents the normal vector of the b-th node among the boundary nodes.

[0178] α, γ, and δ all represent weight coefficients.

[0179] are all related parameters.

[0180] represents the L2 regularization factor.

[0181] The above-mentioned embodiments are the preferred embodiments of the present invention, which are only used to conveniently illustrate the present invention and do not impose any form of limitation on the present invention. Any person with ordinary knowledge in the technical field to which the present invention pertains, if without departing from the technical features of the present invention, makes local modifications or equivalent embodiments by using the technical content disclosed in the present invention, and without departing from the technical feature content of the present invention, still falls within the scope of the technical features of the present invention.

Claims

1. An EIT imaging system based on semi-inverse problem solving, characterized in that: include: A storage module, a first solution module, and a second solution module; The storage module is used to store the boundary voltage matrix; The first solution module establishes a mapping relationship between a boundary voltage matrix and an internal potential matrix; The second solution module establishes a mapping relationship from an internal potential matrix to a node conductivity matrix; or, The storage module is used to store the boundary voltage difference matrix; The first solution module establishes a mapping relationship between a boundary voltage difference matrix and an internal potential difference matrix; The second solving module establishes a mapping relationship from the internal potential difference matrix to the node conductivity difference matrix.

2. The EIT imaging system according to claim 1, characterized in that: The first solution module and the second solution module are both solution modules constructed by neural networks.

3. The EIT imaging system according to claim 2, characterized in that: The input layer of the first solution module is the boundary voltage matrix of the field to be measured, and the output layer is the internal potential matrix corresponding to the network nodes of the finite element model of the field to be measured; The input layer of the second solution module is the internal potential matrix corresponding to the network nodes of the finite element model of the field to be measured, and the output layer is the node conductivity matrix; or, The input layer of the first solution module is the boundary voltage difference matrix of the field to be measured, and the output layer is the internal potential difference matrix corresponding to the network nodes of the finite element model of the field to be measured; The input layer of the second solution module is the internal potential difference matrix corresponding to the network nodes of the finite element model of the field to be measured, and the output layer is the node conductivity difference matrix.

4. An EIT imaging method based on solving a semi-inverse problem, used for absolute imaging, characterized in that: The steps include: S100, constructing a training set; each training set includes: a boundary voltage matrix V, a node potential matrix u, and a node conductivity matrix σ; S200, constructing a neural network model of a first solution module, which includes: Input layer: boundary voltage matrix V; Output layer: node potential matrix u Middle layer; S300, constructing a neural network model of a second solution module, which includes: Input layer: node potential matrix u; Output layer: node conductivity matrix σ; Middle layer; S400, determining the loss functions of the first solution module and the second solution module, and then training the first solution module and the second solution module; S500, reading the boundary voltage matrix of the field to be measured, and then sequentially obtaining the corresponding node potential matrix and node conductivity matrix, and performing imaging according to the node conductivity matrix.

5. The EIT imaging method based on semi-inverse problem solving according to claim 4, characterized in that: The loss function of the second solution module is: σ d0 represents the true value of the conductivity of the dth node, σ d represents the predicted value of the conductivity of the dth node; Represents relevant parameters; N represents the number of nodes of the finite element model of the field to be tested; α represents the weight coefficient; represents the L2 regularization factor.

6. An EIT imaging method based on solving a semi-inverse problem, used for differential imaging, characterized in that: The steps include: S100, constructing a training set; each training set includes: a boundary voltage difference matrix △V, a node potential difference matrix △u, and a node conductivity difference matrix △σ; S200, constructing a neural network model of a first solution module, which includes: Input layer: boundary voltage difference matrix △V; Output layer: node potential difference matrix △u; Middle layer; S300, constructing a neural network model of a second solution module, which includes: Input layer: node potential difference matrix △u; Output layer: node conductivity difference matrix △σ; Middle layer; S400, determining the loss functions of the first solution module and the second solution module, and then training the first solution module and the second solution module; S500, reading the boundary voltage difference matrix of the field to be measured, and then sequentially obtaining the corresponding node potential difference matrix and node conductivity difference matrix, and performing imaging according to the node conductivity difference matrix.

7. An EIT imaging method based on solving a semi-inverse problem, used for absolute imaging, characterized in that: The steps include: S100, constructing a training set; each training set includes: a boundary voltage matrix V, a node potential matrix u, and a node conductivity matrix σ; S200, constructing a neural network model of a first solution module, which includes: Input layer: boundary voltage matrix V; Output layer: node potential matrix u; Middle layer; S300, constructing a neural network model of a second solution module, which includes: Input layer: node potential matrix u; Output layer: node conductivity matrix σ; Middle layer; S400, determining the loss functions of the first solution module and the second solution module, and then training the first solution module and the second solution module; Among them, the loss function of the second solution module adopts: σ d0 represents the true value of the conductivity of the dth node, σ d represents the predicted value of the conductivity of the dth node; u d represents the true value of the potential of the dth node; σ b-边界 、u b-边界 、n b-边界 Represents the predicted value of conductivity, the true value of potential, and the normal vector of the bth node in the boundary node of the finite element model of the field to be measured; represents the solution divergence, represents the solution gradient; are all related parameters; N represents the number of nodes of the finite element model of the field to be measured, n represents the number of boundary nodes of the finite element model of the field to be measured; α, γ, δ all represent weight coefficients; represents the L2 regularization factor; S500, reading the boundary voltage matrix of the field to be measured, and then sequentially obtaining the corresponding node potential matrix and node conductivity matrix, and performing imaging according to the node conductivity matrix.

8. The EIT imaging method based on semi-inverse problem solving according to claim 7, characterized in that: Step S100 also includes: obtaining a node network matrix and calculating a boundary node loss parameter calculation factor matrix and an elliptic equation loss parameter calculation factor matrix according to the node network matrix; The elliptic equation loss parameter calculation factor matrix includes the elliptic equation loss parameter calculation factors corresponding to N nodes of the finite element model of the field to be measured, wherein the elliptic equation loss parameter calculation factor corresponding to any d-th node is u d represents the potential of the dth node, represents the solution gradient; The boundary node loss parameter calculation factor matrix includes boundary node loss parameter calculation factors corresponding to n boundary nodes; the loss parameter calculation factor of any b-th boundary node is: Among them, u b-边界 represents the potential of the bth node among the boundary nodes, n b-边界 Represents the normal vector of the bth node among the boundary nodes.

9. A storage medium, characterized in that: The storage medium stores a program capable of executing the EIT imaging method according to any one of claims 4 to 8.

Images