An EIT imaging algorithm optimization method based on improved particle swarm algorithm

Abstract

The application discloses an EIT imaging algorithm optimization method based on an improved particle swarm algorithm, and the method comprises the following steps: establishing an FEM model, and building a generalized regression neural network as an EIT imaging algorithm; a particle adaptive decomposition mechanism is proposed, which is combined with a particle swarm algorithm to adaptively decompose low-quality particles and retain and split high-quality particles; different particles in the adaptive decomposition mechanism particle swarm algorithm are subjected to dynamic Chebyshev chaotic mapping in different ways to obtain an adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm; the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm is used to perform optimization calculation on a smoothing factor in the generalized regression neural network of the EIT imaging algorithm, so that the optimal solution of the smoothing factor is obtained, and better EIT image reconstruction effects are obtained. The method is based on a particle swarm optimization algorithm, learns from the decomposition, retention and splitting mechanism of biological cells, proposes a particle adaptive decomposition mechanism, guarantees the optimization precision of the algorithm, improves the optimization efficiency, and further combines the dynamic Chebyshev chaotic mapping to solve the problem that the particle swarm optimization algorithm is prone to falling into a local optimum. The adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm is used to optimize the EIT imaging algorithm, so that higher algorithm convergence precision is obtained, and the EIT imaging algorithm is better optimized.

Description

Technical Field

[0001] This invention relates to an optimization method for electrical impedance tomography (EIT) algorithms, and more particularly to an optimization method for EIT imaging algorithms based on an improved particle swarm optimization algorithm. Background Technology

[0002] Electrical impedance tomography (EIT) is a novel, non-invasive biomedical detection and imaging technique that targets the distribution or changes in electrical impedance within a living organism. It reconstructs the organism's impedance distribution by applying a safe excitation current to the organism and measuring the voltage signal on the organism's surface. It offers advantages such as simplicity, non-invasiveness, and low cost, and is of great significance for the early prevention, diagnosis, treatment, and screening of diseases.

[0003] The challenge of EIT imaging lies in its inverse problem; the inverse problem of EIT is to reconstruct the impedance distribution inside the target from the measured voltage at the target boundary. The image reconstruction algorithm used to solve the EIT inverse problem is crucial to the final imaging quality of electrical impedance imaging technology. From an applied mathematics perspective, the EIT inverse problem is a problem of solving nonlinear integral equations, characterized by severe nonlinearity, underdeterminism, and ill-conditioned phenomena.

[0004] Currently, intelligent algorithms such as neural networks are commonly used as EIT imaging algorithms to solve inverse problems because they can effectively address the nonlinearity, underdeterminism, and ill-conditioned nature of inverse problems. However, these algorithms require complex pre-preparation and pre-setting of hyperparameters to achieve better convergence accuracy. Cross-validation has been a common method to obtain better hyperparameters, but due to the large range of hyperparameters, this method typically requires numerous repeated experiments to achieve better results, making it tedious and complex, and prone to getting trapped in local optima. Researchers often combine swarm optimization algorithms with neural networks to obtain better hyperparameter values. However, traditional swarm optimization algorithms also suffer from low optimization quality, poor efficiency, and an inability to escape local optima, resulting in low-quality EIT imaging results. Therefore, a highly efficient optimization method for EIT imaging algorithms is still lacking. Summary of the Invention

[0005] Purpose of the invention: The technical problem to be solved by the present invention is to provide an optimization method for EIT imaging algorithm based on an improved particle swarm optimization algorithm, which addresses the shortcomings of the existing technology.

[0006] To address the aforementioned technical problems, this invention discloses an optimization method for EIT imaging algorithms based on an improved particle swarm optimization algorithm, comprising the following steps:

[0007] Step 1: Establish the FEM model and build a generalized regression neural network as the EIT imaging algorithm;

[0008] Step 2 proposes a particle adaptive decomposition mechanism, which is combined with the particle swarm algorithm to adaptively decompose low-mass particles while retaining and splitting high-mass particles.

[0009] Step 3: Apply dynamic Chebyshev chaotic mapping to different particles in the adaptive decomposition mechanism particle swarm optimization algorithm in different ways to obtain the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm.

[0010] Step 4: The smoothing factor in the generalized regression neural network of the EIT imaging algorithm is optimized using the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm to obtain the optimal solution of the smoothing factor and achieve better EIT image reconstruction results.

[0011] In step 1 of this invention, the finite element method is used to solve the forward EIT problem, and the generalized regression neural network is used to solve the inverse EIT problem.

[0012] The particle adaptive decomposition mechanism described in step 2 of this invention draws on the decomposition, retention, and division mechanisms of biological cells. Low-quality particles are regarded as aging or diseased useless cells in the organism and are decomposed. High-quality particles are regarded as normal cells in the organism that maintain the body's functions and are retained. At the same time, in order to maintain the total number of particles and ensure the search capability of the algorithm, the retained particles are further divided.

[0013] The particle adaptive decomposition mechanism described in step 2 determines the number of particles to be decomposed based on the average fitness value after each optimization iteration. Particles with fitness values ​​worse than the average fitness value are considered low-quality particles (decomposed particles), while particles with fitness values ​​better than the average fitness value are considered high-quality particles (retained particles). Since the position and velocity information of the particles participating in each optimization iteration changes, the average fitness value obtained each time also changes dynamically. Therefore, the determination of the number of particles to be decomposed is dynamically adaptive.

[0014] In step 3 of this invention, dynamic Chebyshev chaotic mapping is applied to different particles in different ways. The dynamic Chebyshev chaotic mapping is set for the inertia weight ω in the update formula of particles that adopt the retention operation. The specific settings are as follows:

[0015]

[0016] In the formula: i represents the number of particles; k represents the number of iterations; ζ i This represents the fitness value of the retained particle; ζ min ζ represents the minimum value for retaining the particle's fitness. max This represents the maximum value of the retained particle fitness.

[0017] Step 3 describes applying dynamic Chebyshev chaotic mapping to different particles in different ways, and then applying the position information of the particles undergoing the splitting operation. Configure dynamic Chebyshev chaos mapping as follows:

[0018]

[0019] In the formula: i represents the number of particles; k represents the number of iterations; ξ i ξ represents the fitness value of the splitting particle; min ξ represents the minimum fitness value of a splitting particle; max This represents the maximum fitness value of the splitting particle.

[0020] The adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm described in step 3 specifically includes the following steps:

[0021] (1) Initialize the population;

[0022] (2) Evaluate the fitness of each particle;

[0023] (3) Compare the fitness value of each particle with the best position pbest it has passed through. If it is better, then use it as the current best position pbest; otherwise, keep the original pbest.

[0024] (4) Compare the pbest of each particle and take the best pbest as the current best gbest;

[0025] (5) Calculate the average particle fitness.

[0026] (6) Decomposition fitness is inferior to particles;

[0027] (7) Retention fitness is better than particles;

[0028] (8) Calculate the number of particles m in the decomposition and arrange the remaining particles in order of fitness from best to worst;

[0029] (9) m retained particles before splitting;

[0030] (10) Set a dynamic chaotic mapping for the inertial weight ω of the information update formula of the particles;

[0031] (11) The position X of the information update formula for splitting particles i k Set up a dynamic chaos map;

[0032] (12) Update the particle's velocity and position;

[0033] (13) Determine whether the optimization work is completed: If it is completed, then end and output the optimization result; if it is not completed, then jump to step (2) to continue.

[0034] Compared with the prior art, the beneficial effects of the present invention are:

[0035] This invention proposes an adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm for optimizing the EIT imaging algorithm. Based on particle swarm optimization, this algorithm draws inspiration from the decomposition, preservation, and division mechanisms of biological cells, proposing an adaptive particle decomposition mechanism. This mechanism adaptively decomposes low-mass particles while preserving and dividing high-mass particles, ensuring the accuracy and efficiency of the optimization. Furthermore, it applies dynamic Chebyshev chaotic mapping to different particles in different ways, addressing the problem of particle swarm optimization easily getting trapped in local optima. By using the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm instead of particle swarm optimization to optimize the hyperparameters of the EIT imaging algorithm's GRNN, the EIT imaging algorithm achieves higher convergence accuracy and yields better image reconstruction results. Attached Figure Description

[0036] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following drawings and embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any way.

[0037] Figure 1 This is a flowchart of the present invention.

[0038] Figure 2 This is a flowchart illustrating the combination of the adaptive decomposition mechanism of this invention with the particle swarm optimization algorithm.

[0039] Figure 3 This is a flowchart of the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm in this invention.

[0040] Figure 4 This is a flowchart illustrating the combination of the adaptive decomposition mechanism, the dynamic chaotic particle swarm optimization algorithm, and the EIT imaging algorithm with a generalized regression neural network in this invention.

[0041] Figure 5 The graph shows the results of optimizing the generalized regression neural network of the EIT imaging algorithm using the particle swarm optimization algorithm and the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm in this invention.

[0042] Figure 6 This is a simulation diagram of the conductivity of some of the target objects to be tested in this invention.

[0043] Figure 7 This is a schematic diagram of the image reconstruction simulation results of the present invention. Detailed Implementation

[0044] like Figure 1 The diagram shows a flowchart of the EIT imaging algorithm optimization method based on the improved particle swarm optimization algorithm of this invention. The specific steps are as follows:

[0045] Step 1: Establish the FEM model and build a generalized regression neural network as the EIT imaging algorithm;

[0046] Step 2 proposes a particle adaptive decomposition mechanism, which is combined with the particle swarm algorithm to adaptively decompose low-mass particles while retaining and splitting high-mass particles.

[0047] Step 3: Apply dynamic Chebyshev chaotic mapping to different particles in the adaptive decomposition mechanism particle swarm optimization algorithm in different ways to obtain the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm.

[0048] Step 4: The smoothing factor in the generalized regression neural network of the EIT imaging algorithm is optimized using the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm to obtain the optimal solution of the smoothing factor and achieve better EIT image reconstruction results.

[0049] like Figure 2 The diagram shows a flowchart of the integration of the particle adaptive decomposition mechanism of this invention with the particle swarm optimization algorithm. In this embodiment, a particle adaptive decomposition mechanism is proposed. This mechanism draws inspiration from the decomposition, retention, and division mechanisms of biological cells. Organisms check their own bodily functions, decompose aging or diseased harmful cells, retain normally functioning beneficial cells, and divide to fill the gaps created by decomposed cells, maintaining the normal operation of the organism. With the introduction of the particle adaptive decomposition mechanism, low-mass particles are considered aging or diseased harmful cells in the organism and are decomposed, while high-mass particles are considered normal cells maintaining bodily functions and are retained. Simultaneously, to maintain particle... To maintain the total number of particles and ensure the algorithm's search capability, the retained particles are then split. The number of split particles is consistent with the number of decomposed particles. The number of decomposed particles is determined by the average fitness value after each optimization iteration. In this paper, particles with fitness values ​​worse than the average fitness value are considered low-quality particles (decomposed particles), while particles with fitness values ​​better than the average fitness value are considered high-quality particles (retained particles). Since the position and velocity information of the particles participating in each optimization iteration changes, the average fitness value obtained each time also changes dynamically. Therefore, the determination of the number of decomposed particles is dynamically adaptive.

[0050] In this embodiment, the specific steps of combining the particle adaptive decomposition mechanism with the particle swarm optimization algorithm are as follows:

[0051] (1) Initialize the population;

[0052] (2) Evaluate the fitness of each particle;

[0053] (3) Compare the fitness value of each particle with the best position pbest it has passed through. If it is better, then use it as the current best position pbest; otherwise, keep the original pbest.

[0054] (4) Compare the pbest of each particle and take the best pbest as the current best gbest;

[0055] (5) Calculate the average fitness of the particles in this test.

[0056] (6) Decomposition fitness is inferior to The particles retain fitness better than particles;

[0057] (7) Calculate the number of particles m in the decomposition and arrange the remaining particles in order of fitness value from best to worst;

[0058] (8) m retained particles before splitting;

[0059] (9) Update the particle's velocity and position;

[0060] (10) Determine whether the optimization work is completed: If it is completed, then end and output the optimization result; if it is not completed, then jump to step (2) to continue.

[0061] like Figure 3 The diagram shows a flowchart of the dynamic chaotic particle swarm optimization algorithm with adaptive decomposition mechanism in this invention. In this embodiment, the specific way to combine the dynamic Chebyshev mapping with the PSO of the adaptive decomposition mechanism is to establish two sets of particle update formulas. For particles that take the hold operation, because their quality is higher, they have a greater probability of orbiting around the global optimum. Therefore, the Chebyshev mapping is used to enhance the local search ability of these particles and increase the possibility of finding the global optimum. So when they update their velocity and position information, the inertia weight ω in the update formula of these particles is set with Chebyshev mapping, and combined with the average fitness value of the particles, ω is set to be dynamically changing, as follows.

[0062]

[0063] In the formula: i represents the number of particles; k represents the number of iterations; ζ i This represents the fitness value of the retained particle; ζ min ζ represents the minimum value for retaining the particle's fitness. max This represents the maximum value of the retained particle fitness.

[0064] For particles undergoing splitting operations, to avoid premature algorithm convergence after introducing the adaptive decomposition mechanism and to ensure overall diversity, the position information of these particles is updated during the velocity and position information update process. By setting up a dynamic Chebyshev map to shuffle the positions of these particles, the algorithm avoids concentrating all particles around a single potential optimal solution. Instead, some particles surround potential optimal solutions while others search for new regions by randomly changing their positions. This effectively prevents the algorithm from getting trapped in local optima. The specific settings are as follows:

[0065]

[0066] In the formula: i represents the number of particles; k represents the number of iterations; ξ i ξ represents the fitness value of the splitting particle; min ξ represents the minimum fitness value of a splitting particle; max This represents the maximum fitness value of the splitting particle.

[0067] In this embodiment, the specific steps of the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm are as follows:

[0068] (1) Initialize the population;

[0069] (2) Evaluate the fitness of each particle;

[0070] (3) Compare the fitness value of each particle with the best position pbest it has passed through. If it is better, then use it as the current best position pbest; otherwise, keep the original pbest.

[0071] (4) Compare the pbest of each particle and take the best pbest as the current best gbest;

[0072] (5) Calculate the average particle fitness.

[0073] (6) Decomposition fitness is inferior to particles;

[0074] (7) Retention fitness is better than particles;

[0075] (8) Calculate the number of particles m in the decomposition and arrange the remaining particles in order of fitness from best to worst;

[0076] (9) m retained particles before splitting;

[0077] (10) Set a dynamic chaotic mapping for the inertial weight ω of the information update formula of the particles;

[0078] (11) The position X of the information update formula for splitting particles i k Set up a dynamic chaos map;

[0079] (12) Update the particle's velocity and position;

[0080] (13) Determine whether the optimization work is completed: If it is completed, then end and output the optimization result; if it is not completed, then jump to step (2) to continue.

[0081] like Figure 4 The diagram shows a flowchart of the combination of the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm and the generalized regression neural network in this invention. In this embodiment, the generalized regression neural network serves as the master solver for the EIT inverse problem, performing image reconstruction. The adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm further optimizes the hyperparameters of the generalized regression neural network, resulting in a higher quality reconstructed image. The two algorithms are combined by using the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm to optimize the hyperparameters of the generalized regression neural network, and using the evaluation index of the generalized regression neural network's performance as the particle fitness of the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm.

[0082] In this embodiment, the specific steps of combining the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm and the generalized regression neural network are as follows:

[0083] (1) Input the boundary voltage difference data of the target field to be measured;

[0084] (2) Initialize the smoothing factor ε of the generalized regression neural network;

[0085] (3) Initialize the population and map the ε to be optimized to particle position information;

[0086] (4) Evaluate the fitness of each particle;

[0087] (5) Compare the fitness value of each particle with the best position pbest it has passed through. If it is better, then use it as the current best position pbest; otherwise, keep the original pbest.

[0088] (6) Compare the pbest of each particle and take the best pbest as the current best gbest;

[0089] (7) Calculate the average particle fitness. Decomposition fitness is inferior to The particles retain fitness better than particles;

[0090] (8) Calculate the number of particles m to be decomposed, and arrange the retained particles in order of fitness from best to worst. The first m retained particles are the first to split.

[0091] (9) Set a dynamic chaotic mapping for the inertial weight ω of the information update formula of the particles;

[0092] (10) is the location of the information update formula for splitting particles. Set up a dynamic chaos map;

[0093] (11) Update the particle's velocity and position;

[0094] (12) Determine if the optimization work is complete: If it is complete, end the process, output the optimization result, and continue to step (13); if it is not complete, jump to step (4) and continue.

[0095] (13) Substitute the smoothing factor optimized by the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm into the generalized regression neural network.

[0096] (14) Assign the input boundary voltage data to the input layer neurons;

[0097] (15) Perform calculations on the training samples corresponding to the pattern layer neurons;

[0098] (16) The calculation results are passed to the summation layer for weighted summation and arithmetic summation;

[0099] (17) The summation result is passed to the output layer for output;

[0100] (18) Complete EIT image reconstruction.

[0101] like Figure 5 The figure shows the optimization results of the generalized regression neural network using the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm and the particle swarm optimization algorithm in this invention. In this embodiment, the generalized regression neural network was optimized using the particle swarm optimization algorithm and the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm, respectively. The RMSE of the optimized generalized regression neural network were 0.0233 and 0.0216, respectively. It can be seen that the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm is better than the particle swarm optimization algorithm in optimizing the generalized regression neural network. Therefore, the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm effectively solves the problems of poor optimization quality and easy getting trapped in local optima in the particle swarm optimization algorithm, and obtains higher algorithm convergence accuracy, thus better optimizing the EIT imaging algorithm.

[0102] like Figure 6The figure shown is a simulation diagram of the conductivity of some of the target objects to be tested according to the present invention. In this embodiment, the simulation uses a 16-electrode EIT system to reconstruct the conductivity distribution inside the field. The background conductivity value is equal to 0.142 S / m, the target object conductivity value is equal to 0.667 S / m, and the number of target objects is set to 1-3, with their positions and sizes randomly generated.

[0103] like Figure 7 The diagram shown illustrates the image reconstruction simulation results of this invention. In this embodiment, based on the particle swarm optimization algorithm, and drawing inspiration from the decomposition, preservation, and division mechanisms of biological cells, an adaptive particle decomposition mechanism is proposed. This mechanism adaptively decomposes low-quality particles while preserving and dividing high-quality particles, ensuring the accuracy of the algorithm optimization and improving its efficiency. Furthermore, a dynamic Chebyshev chaotic mapping is applied to different particles in different ways, resulting in an adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm. This addresses the problem of particle swarm optimization algorithms easily getting trapped in local optima. Combined with the EIT imaging algorithm, the smoothing factor of the generalized regression neural network is optimized, addressing the problem of insufficient imaging accuracy caused by inappropriate network smoothing factor selection, improving the network's nonlinear approximation capability, and thus enhancing the quality of image reconstruction.

[0104] This invention provides an idea and method for optimizing the EIT imaging algorithm based on an improved particle swarm optimization algorithm. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. An optimization method for EIT imaging algorithm based on an improved particle swarm optimization algorithm, characterized in that, Includes the following steps: Step 1: Establish the FEM model and build a generalized regression neural network as the EIT imaging algorithm; Step 2 proposes an adaptive particle decomposition mechanism, which is combined with the particle swarm optimization algorithm to adaptively decompose low-quality particles and retain and split high-quality particles. The adaptive particle decomposition mechanism draws on the decomposition, retention and splitting mechanism of biological cells. Low-quality particles are regarded as aging or diseased useless cells in the organism and are decomposed. High-quality particles are regarded as normal cells in the organism that maintain the body's functions and are retained. At the same time, in order to maintain the total number of particles and ensure the search capability of the algorithm, the retained particles are split. Step 3: Apply dynamic Chebyshev chaotic mapping to different particles in the adaptive decomposition mechanism particle swarm optimization algorithm in different ways to obtain the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm; update the inertia weight in the particle update formula by adopting the retention operation. Configure dynamic Chebyshev chaos mapping as follows: （1） In the formula: Indicates the number of particles; Indicates the number of iterations; This represents the fitness value of the retained particle; This represents the minimum value for retaining the particle's fitness. This represents the maximum value of the retained particle fitness. Position information of particles that undergo splitting operation Configure dynamic Chebyshev chaos mapping as follows: （2） In the formula: Indicates the number of particles; Indicates the number of iterations; Represents the fitness value of the splitting particle; This represents the minimum fitness value of a splitting particle. This represents the maximum fitness value of the splitting particle; Step 4: The smoothing factor in the generalized regression neural network of the EIT imaging algorithm is optimized using the adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm to obtain the optimal solution of the smoothing factor and achieve better EIT image reconstruction results.

2. The method for optimizing the EIT imaging algorithm based on the improved particle swarm optimization algorithm according to claim 1, characterized in that, The number of decomposed particles is determined by the average fitness value after each optimization iteration. Particles with fitness values ​​worse than the average fitness value are considered low-quality particles, while particles with fitness values ​​better than the average fitness value are considered high-quality particles. Since the position and velocity information of the particles participating in each optimization iteration changes, the average fitness value obtained each time also changes dynamically. Therefore, the determination of the number of decomposed particles is dynamic and adaptive.

3. The method for optimizing the EIT imaging algorithm based on the improved particle swarm optimization algorithm according to claim 1, characterized in that, The adaptive decomposition mechanism dynamic chaotic particle swarm optimization algorithm described in step 3 specifically includes the following steps: (1) Initialize the population; (2) Evaluate the fitness of each particle; (3) Assign each particle's fitness value to the best position it has passed through. Compare the two positions, and if the former is better, then use it as the current best position. Otherwise, retain the original. ; (4) Each particle's Compare and select the optimal one. As the best at present ; (5) Calculate the average fitness of the particles. ; (6) Decomposition fitness is inferior to particles; (7) Retention fitness is better than particles; (8) Calculate the number of particles in the decomposition. The retained particles will be arranged from best to worst fitness. (9) Before splitting One retained particle; (10) Inertial weights for updating the formula to preserve particle information Set up a dynamic chaos map; (11) Position of the information update formula for splitting particles Set up a dynamic chaos map; (12) Update the particle's velocity and position; (13) Determine whether the optimization work is completed: If it is completed, then end and output the optimization result; if it is not completed, then jump to step (2) to continue.

Images