Method for generating sound field distribution diagram of electromagnetic ultrasonic transducer and electronic device

Abstract

The present application proposes a method and electronic device for generating a sound field distribution map of an electromagnetic ultrasonic transducer, wherein the method comprises: combining a first sound field distribution map sequence of ultrasonic waves excited by a sample electromagnetic ultrasonic transducer on the surface of a sample test piece within a first time range and a sample sound field distribution map of ultrasonic waves excited on the surface of a sample test piece at a sample moment after the first time range, training a generator and a discriminator in a generative adversarial network model, and generating a sound field distribution map of ultrasonic waves excited by a target electromagnetic ultrasonic transducer on the surface of a target test piece at a target moment after a second time range based on the trained generator. Thus, the generative adversarial network model is trained, and the sound field distribution map of the electromagnetic ultrasonic transducer is generated by the generator in the generative adversarial network model, thereby achieving accurate sound field distribution generation.

Description

Technical Field

[0001] The present application relates to the technical field of electromagnetic ultrasonic transducers, and in particular to a method for generating a sound field distribution diagram of an electromagnetic ultrasonic transducer and an electronic device. Background Art

[0002] With the continuous development of industrial nondestructive testing and material characterization technology, electromagnetic ultrasonic transducer (EMAT), as an important nondestructive testing tool, has been widely used in the inspection and evaluation of engineering structures such as metals and composite materials.

[0003] In the process of nondestructive testing of the test piece (hereinafter referred to as the test piece) by EMAT, obtaining the sound field distribution diagram of the ultrasonic wave excited by EMAT on the surface of the test piece at a certain moment is helpful to optimize the design of EMAT and improve the application efficiency of electromagnetic ultrasonic transducers in the field of nondestructive testing. Therefore, it is very meaningful to study a new method for generating sound field distribution. Summary of the invention

[0004] The present application proposes a method for generating a sound field distribution diagram of an electromagnetic ultrasonic transducer and an electronic device.

[0005] On one hand, an embodiment of the present application proposes a method for generating an acoustic field distribution diagram of an electromagnetic ultrasonic transducer, the method comprising: obtaining a first acoustic field distribution diagram sequence of an ultrasonic wave excited on the surface of a sample test piece by a sample electromagnetic ultrasonic transducer within a first time range, and a sample acoustic field distribution diagram of the ultrasonic wave excited on the surface of the sample test piece at a sample moment after the first time range; based on a generator in a generative adversarial network model, generating a generated acoustic field distribution diagram of the ultrasonic wave excited on the surface of the sample test piece by the sample electromagnetic ultrasonic transducer at the sample moment according to the first acoustic field distribution diagram sequence; based on a discriminant in the generative adversarial network model, generating a generated acoustic field distribution diagram of the ultrasonic wave excited on the surface of the sample test piece by the sample electromagnetic ultrasonic transducer at the sample moment A generator is provided for determining a first discrimination result of the generated sound field distribution map and a second discrimination result of the sample sound field distribution map according to the generated sound field distribution map and the sample sound field distribution map; the generator and the discriminator are trained according to the generated sound field distribution map, the sample sound field distribution map, the first discrimination result and the second discrimination result; based on the trained generator, a sound field distribution map of the ultrasonic wave excited by the target electromagnetic ultrasonic transducer on the surface of the target test piece at a target moment after the second time range is generated according to a second sound field distribution map sequence of the ultrasonic wave excited by the target electromagnetic ultrasonic transducer on the surface of the target test piece within the second time range.

[0006] The technical solution provided by the embodiments of the present application may have the following beneficial effects:

[0007] Combined with the first sound field distribution map sequence of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece within the first time range and the sample sound field distribution map of the ultrasonic wave excited on the surface of the sample test piece at the sample moment after the first time range, the generator and the discriminator in the generative adversarial network model are trained, and based on the trained generator, according to the second sound field distribution map sequence of the ultrasonic wave excited by the target electromagnetic ultrasonic transducer on the surface of the target test piece within the second time range, the sound field distribution map of the ultrasonic wave excited by the target electromagnetic ultrasonic transducer on the surface of the target test piece at the target moment after the second time range is generated. In this way, the generative adversarial network model is trained, and the sound field distribution map of the electromagnetic ultrasonic transducer is generated by the generator in the generative adversarial network model, so that accurate sound field distribution generation is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present application.

[0009] Figure 1 is a schematic flow chart of a method for generating a sound field distribution diagram of an electromagnetic ultrasonic transducer according to an embodiment of the present application;

[0010] Figure 2 It is an example diagram of a two-dimensional finite element simulation model including a permanent magnet, a winding coil, and a sample test piece;

[0011] Figure 3 It is an example diagram of the simulated sound field distribution diagram on the finite element model of the sample test piece at a certain moment;

[0012] Figure 4 is a schematic flow chart of a method for generating a sound field distribution diagram of an electromagnetic ultrasonic transducer according to another embodiment of the present application;

[0013] Figure 5 This is a diagram showing an example of the structure of a generator;

[0014] Figure 6 is a schematic flow chart of a method for generating a sound field distribution diagram of an electromagnetic ultrasonic transducer according to another embodiment of the present application;

[0015] Figure 7 This is an example diagram of the structure of the discriminator;

[0016] Figure 8 This is an example diagram of the structure of the conditional generative adversarial network model;

[0017] Fig. 9 is a schematic structural diagram of a device for generating a sound field distribution diagram of an electromagnetic ultrasonic transducer according to an embodiment of the present application;

[0018] Fig.10It is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] 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 intended to be used to explain the present application, and should not be construed as limiting the present application.

[0020] The following describes a method, device, electronic device, and storage medium for generating a sound field distribution diagram of an electromagnetic ultrasonic transducer according to an embodiment of the present application with reference to the accompanying drawings.

[0021] Figure 1 It is a flow chart of a method for generating a sound field distribution map of an electromagnetic ultrasonic transducer according to an embodiment of the present application. It should be noted that the method for generating a sound field distribution map of an electromagnetic ultrasonic transducer provided in this embodiment is executed by a sound field distribution map generating device of an electromagnetic ultrasonic transducer, and the sound field distribution map generating device of the electromagnetic ultrasonic transducer in this embodiment can be implemented by software and / or hardware, and the sound field distribution map generating device of the electromagnetic ultrasonic transducer in this example can be an electronic device, or can be configured in an electronic device.

[0022] The electronic devices in this example embodiment may include terminal devices, servers, etc., wherein the terminal devices may be PCs (Personal Computers), mobile devices, tablet computers, etc., and this embodiment does not specifically limit this.

[0023] like Figure 1 As shown, the method for generating the sound field distribution diagram of the electromagnetic ultrasonic transducer may include:

[0024] Step 101, obtaining a first sound field distribution diagram sequence of ultrasonic waves excited on the surface of a sample test piece by a sample electromagnetic ultrasonic transducer within a first time range and a sample sound field distribution diagram of ultrasonic waves excited on the surface of the sample test piece at a sample time after the first time range.

[0025] The first time range refers to from the first time to the second time, wherein the second time is greater than the first time. For example, the first time range can be from 0 to 12.5 microseconds, or 0 to 10 milliseconds, or 2 to 8 milliseconds, etc. This embodiment does not specifically limit the first time range.

[0026] The first sound field distribution diagram sequence may include: sound field distribution diagrams of ultrasonic waves excited on the surface of the sample test piece by the sample electromagnetic ultrasonic transducer model at multiple first moments in time sequence, wherein the first moments are all within the first time range. In some embodiments, the first time range may be divided into multiple first moments, and the sound field distribution diagrams of ultrasonic waves excited on the surface of the sample test piece by the sample electromagnetic ultrasonic transducer model at each first moment are determined, and the multiple sound field distribution diagrams are sorted in order from the first moment to the back to obtain the first sound field distribution diagram sequence.

[0027] Among them, the multiple first times are separated by a preset time interval, and the preset time interval can be set according to actual needs. This implementation does not specifically limit the preset time interval.

[0028] The sample time refers to any time after the first time range. For example, the sample time may be the first time after the first time range, or the second time, or the third time, the fourth time, etc. This embodiment does not specifically limit this.

[0029] In some embodiments, in order to conveniently obtain a first sound field distribution diagram sequence, a possible implementation method for obtaining the first sound field distribution diagram sequence of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece within the first time range is: constructing an electromagnetic ultrasonic transducer model of the sample electromagnetic ultrasonic transducer and a test piece model of the sample test piece; performing simulation according to the electromagnetic ultrasonic transducer model and the test piece model to obtain a simulated sound field distribution diagram sequence of the ultrasonic wave excited by the electromagnetic ultrasonic transducer model on the surface of the test piece model within the first time range; and determining the first sound field distribution diagram sequence of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece within the first time range according to the simulated sound field distribution diagram sequence.

[0030] In some embodiments, the simulated sound field distribution map sequence can be used as the first sound field distribution map sequence. Correspondingly, the sample sound field distribution map is obtained by: simulating the electromagnetic ultrasonic transducer model and the test piece model to obtain a simulated sound field distribution map of the ultrasonic wave excited by the electromagnetic ultrasonic transducer model on the surface of the test piece model at the sample time, and determining the sample sound field distribution map according to the simulated sound field distribution map.

[0031] In some embodiments, the simulated sound field distribution map may be used as the sample sound field distribution map.

[0032] In this embodiment, an aluminum plate specimen is taken as an example for exemplary description.

[0033] It should be noted that the electromagnetic ultrasonic transducer model in this embodiment is highly consistent with the electromagnetic ultrasonic transducer in the real scene. Correspondingly, the sample test piece model is highly consistent with the test piece in the real scene.

[0034] In this embodiment, simulation is performed based on the electromagnetic ultrasonic transducer model and the test piece model to obtain a simulated sound field distribution diagram of the ultrasonic wave excited by the electromagnetic ultrasonic transducer model on the surface of the test piece model at the sample moment, which means that a scenario of non-destructive testing of the sample test piece by the sample electromagnetic ultrasonic transducer is simulated through the electromagnetic ultrasonic transducer model and the test piece model to obtain a simulated sound field distribution diagram of the ultrasonic wave excited by the electromagnetic ultrasonic transducer model on the surface of the test piece model at the sample moment.

[0035] In some embodiments, the electromagnetic ultrasonic transducer model and the test piece model may be finite element models.

[0036] As an example, a first finite element model of the electromagnetic ultrasonic transducer may be established according to the structural parameters of the electromagnetic ultrasonic transducer, and a second finite element model of the sample test piece may be established in combination with the structural parameters of the sample test piece.

[0037] The electromagnetic ultrasonic transducer may include a permanent magnet and a winding coil.

[0038] Among them, there are example diagrams of two-dimensional finite element simulation models of permanent magnets, winding coils, and sample test pieces, such as Figure 2 As shown. Among them, Figure 2 The S in it represents the S pole of the permanent magnet, and N represents the N pole of the permanent magnet. The S pole is the south pole, and the N pole is the north pole.

[0039] In order to clearly understand the process of obtaining a sequence of simulated sound field distribution diagrams, the process is described exemplarily below.

[0040] First, COMSOL is used to establish the geometric model of EMAT. The structural parts of EMAT include coils, permanent magnets and sample test pieces. The geometric dimensions of each component are defined in the model.

[0041] In this example, the sample test piece can be made of 7075 aluminum alloy, and the physical properties of the material are accurately assigned to ensure the accuracy of the simulation results.

[0042] Then, the electromagnetic field model is established based on the Maxwell equations to simulate the propagation of electromagnetic waves and the electromagnetic response of materials. The electromagnetic excitation source in EMAT is a time-varying current signal, which excites a changing magnetic field through the electromagnetic coil and generates sound waves in the metal material. The frequency and amplitude of the excitation source have an important influence on the distribution of the electromagnetic field and the acoustic field. Therefore, in the electromagnetic field modeling process, it is necessary to set an appropriate excitation signal for the coil and define the boundary conditions.

[0043] Then, the acoustic field is modeled based on the structural dynamics equation and the magnetostrictive constitutive equation. The magnetostrictive effect caused by the electromagnetic excitation signal causes the material to deform and generate sound waves. The propagation process of the sound wave is simulated by the structural dynamics equation, which involves the temporal and spatial changes of the sound field. In order to accurately capture the coupling between electromagnetic waves and sound waves, a multi-physics field coupling solver can be used to obtain the sound field distribution of the EMAT in the working state by solving the coupling equations of the electromagnetic field and the acoustic field.

[0044] Then, the actual working environment of EMAT is simulated by setting appropriate boundary conditions. The boundary conditions of the electromagnetic field include radiation boundaries and conductive boundaries to ensure that the propagation and radiation process of electromagnetic waves inside the material are accurate. In the acoustic field analysis, free boundary conditions are used to simulate the propagation of sound waves, and the time evolution of the acoustic field is calculated by the transient solver.

[0045] Finally, based on the coupled modeling of the electromagnetic field and the acoustic field, a transient solution is performed. The transient solver is used to simulate the propagation process of electromagnetic waves under the action of the electromagnetic excitation source, and the propagation of the resulting acoustic waves in the sample test piece is calculated. The simulation process will output the simulated acoustic field distribution diagram at different times, and obtain the simulated acoustic field distribution diagram sequence based on the acoustic field distribution data.

[0046] Among them, an example diagram of the simulated sound field distribution diagram on the finite element model of the sample test piece at a certain moment, such as Figure 3 shown.

[0047] It should be noted that, in this embodiment, a plurality of first sound field distribution diagram sequences may be included.

[0048] Step 102: Based on the generator in the generative adversarial network model, a generated sound field distribution diagram of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece at the sample time is generated according to the first sound field distribution diagram sequence.

[0049] It should be noted that the generative adversarial network model in this embodiment may be a conditional generative adversarial network (CGAN) model.

[0050] In some embodiments, the first sound field distribution diagram sequence may be input into a generator so as to generate a generated sound field distribution diagram of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece at the sample time.

[0051] In some embodiments, the first sound field distribution diagram sequence and the sample time can be input into the generator to generate a generated sound field distribution diagram of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece at the sample time.

[0052] In some embodiments, in order to speed up the convergence of the model in network training and improve performance, the first sound field distribution map sequence and the sample sound field distribution map may be preprocessed before training the generator and the discriminator based on the first sound field distribution map sequence and the sample sound field distribution map.

[0053] The preprocessing may include but is not limited to at least one of the following: region division processing, normalization processing and data enhancement processing.

[0054] In some embodiments, an exemplary process of performing region division processing on the first sound field distribution map sequence and the sample sound field distribution map is: using matrix technology to process the geometric structures of the first sound field distribution map sequence and the sample sound field distribution map respectively to achieve region division. Thus, key information of the sound field distribution can be extracted to ensure that the network can focus on important physical features and avoid unnecessary noise interference.

[0055] In some embodiments, an exemplary process of normalizing the first sound field distribution map sequence and the sample sound field distribution map is: normalizing the first sound field distribution map sequence and the sample sound field distribution map. Thus, the pixel values ​​in the image are normalized, and the normalized image will help optimize the training process, speed up the convergence speed of gradient descent, and improve the performance of the model.

[0056] The standardization process refers to normalizing each pixel value in the first sound field distribution map sequence and the sample sound field distribution map so that the pixel value of each pixel is in the range of [0, 1].

[0057] In some embodiments, in the case where the first sound field distribution map sequence and the sample sound field distribution map are converted into multi-channel images, the first sound field distribution map sequence and the sample sound field distribution map can also be converted into single-channel images. For example, the first sound field distribution map sequence and the sample sound field distribution map are composed of three channels (RGB color images) of red, green, and blue, which can be converted into single-channel images. Among them, the RGB values ​​in the sound field image reflect different physical phenomena, but their important information may be concentrated in one or several channels. Therefore, converting the image into a single-channel image helps to simplify the problem and reduce unnecessary calculations.

[0058] In some implementations, the first sound field distribution map sequence may be converted to a single-channel image to obtain a converted first sound field distribution map sequence, and the sample sound field distribution map may be converted to a single-channel image to obtain a converted sample sound field distribution map, and the converted first sound field distribution map sequence and the sample sound field distribution image may be standardized (or normalized) respectively.

[0059] In some embodiments, in order to improve the generalization ability of the model, correspondingly, the first sound field distribution map sequence and the sample sound field distribution map can also be subjected to data enhancement processing, such as flipping and rotating. The angles and directions of these transformations are randomly selected, and images with diverse position distributions and tilt angles can be generated. This diversity helps to improve the adaptability of the model to different situations, thereby improving its robustness.

[0060] Among them, it can be understood that although these transformations increase the diversity of the data, the information differences introduced are not directly related to other features of the image. Therefore, in the feature extraction and training stages, these irrelevant information may interfere with the model's learning of effective features, thereby negatively affecting the model's performance. In order to eliminate this unnecessary interference and prevent the model from learning irrelevant deviations during training, a central symmetric cropping method can be used. Through central symmetric cropping, all input images are uniformly cropped to a size of 640×480 pixels before being sent to the network. This cropping method effectively concentrates the image content in a relatively small and fixed area, helping the model focus on the key features of the sound field image and reducing the uncertainty caused by the image position, angle, and tilt. This not only improves the consistency of the image input, but also enhances the stability of the model during the feature learning stage.

[0061] Step 103 , based on the discriminator in the generative adversarial network model, determine a first discrimination result of the generated sound field distribution map and a second discrimination result of the sample sound field distribution map according to the generated sound field distribution map and the sample sound field distribution map.

[0062] The first discrimination result is used to indicate the probability that the generated sound field distribution map is a real image.

[0063] The second discrimination result is used to indicate the probability that the sample sound field distribution diagram is a real image.

[0064] In some embodiments, the generated sound field distribution map and the sample sound field distribution map can be input into a discriminator so that the discriminator can respectively discriminate whether the generated sound field distribution map and the sample sound field distribution map are real images, so as to obtain a first discrimination result of the generated sound field distribution map and a second discrimination result of the sample sound field distribution map.

[0065] Step 104 , training the generator and the discriminator according to the generated sound field distribution map, the sample sound field distribution map, the first discrimination result and the second discrimination result.

[0066] In one embodiment of the present application, a possible implementation method of the above step 104 is: determine the total loss value of the generative adversarial network model based on the generated sound field distribution map, the sample sound field distribution map, the first discrimination result and the second discrimination result; according to the total loss value, alternately train the generator and the discriminator until the preset training end conditions are met to obtain the trained generator and discriminator.

[0067] In some embodiments, a possible implementation method for determining the total loss value of the generative adversarial network model based on the generated sound field distribution map, the sample sound field distribution map, the first discrimination result and the second discrimination result is: determine the first loss value of the generator based on the generated sound field distribution map and the sample sound field distribution map; determine the second loss value of the discriminator based on the first discrimination result and the second discrimination result; determine the total loss value of the generative adversarial network model based on the first loss value and the second loss value.

[0068] In some embodiments, the first loss value of the generator may be determined based on the loss function of the generator and according to the generated sound field distribution map and the sample sound field distribution map.

[0069] In some embodiments, the second loss value of the discriminator may be determined according to the first discrimination result and the second discrimination result based on the loss function of the discriminator.

[0070] Among them, the preset training end condition is a pre-set condition for the end of training. For example, the preset training end condition can be a preset number of training times, or the total loss value is less than a preset value, or the change in the total loss value tends to be stable, that is, the difference in the total loss values ​​corresponding to two or more adjacent trainings is less than the set value, that is, the total loss value basically no longer changes.

[0071] In some other embodiments, another possible implementation of step 104 is: determine the first loss value of the generator based on the generated sound field distribution map and the sample sound field distribution map; determine the second loss value of the discriminator based on the first discrimination result and the second discrimination result; train the generator based on the first loss value; train the discriminator based on the second loss value; alternately perform the steps of training the generator and the discriminator until the training end condition is met to obtain the trained generator and discriminator.

[0072] Among them, the training end condition is pre-set according to actual needs. For example, the training end condition can be a preset number of training times, or the first loss value and the second loss value are both less than the preset value, or the changes in the first loss value and the second loss value are both close to being stable, that is, the difference in the first loss values ​​corresponding to two or more adjacent trainings is less than the set value, and the difference in the second loss value is also less than the set value, that is, the first loss value and the second loss value basically no longer change.

[0073] Step 105, based on the trained generator, according to the second sound field distribution diagram sequence of the ultrasonic wave excited by the target electromagnetic ultrasonic transducer on the surface of the target test piece within the second time range, generate a sound field distribution diagram of the ultrasonic wave excited by the target electromagnetic ultrasonic transducer on the surface of the target test piece at a target moment after the second time range.

[0074] The second time range refers to from the third time to the fourth time, wherein the fourth time is greater than the third time, for example, the first time may be 0 microseconds, the fourth time may be 12.5 microseconds, and correspondingly, the second time range may be 0 to 12.5 microseconds, or the first time may be 0 microseconds, the fourth time may be 10 microseconds, and correspondingly, the second time range may be 0 to 10 microseconds. It should be noted that, in this embodiment, the second time range may be 0 to 12.5 microseconds, or a time range between 0 and 12.5 microseconds, and this embodiment does not specifically limit this.

[0075] The target electromagnetic ultrasonic transducer and the sample electromagnetic ultrasonic transducer may be electromagnetic ultrasonic transducers of the same type.

[0076] The target time refers to any time after the second time range. For example, the target time may be the first time, the second time, or the third time after the second time range, and this embodiment does not specifically limit this.

[0077] In this embodiment, an exemplary description is given by taking the sample moment being the first moment after the first time range and the target moment being the first moment after the second time range as an example.

[0078] In this embodiment, an aluminum plate test piece is used as an example for the target test piece to be described. For example, the sample test piece and the target test piece can be different aluminum plate test pieces.

[0079] The method for generating the sound field distribution map of the electromagnetic ultrasonic transducer provided in the embodiment of the present application is combined with the first sound field distribution map sequence of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece within the first time range and the sample sound field distribution map of the ultrasonic wave excited on the surface of the sample test piece at the sample moment after the first time range, to train the generator and the discriminator in the generative adversarial network model, and based on the trained generator, according to the second sound field distribution map sequence of the ultrasonic wave excited by the target electromagnetic ultrasonic transducer on the surface of the target test piece within the second time range, generate the sound field distribution map of the ultrasonic wave excited by the target electromagnetic ultrasonic transducer on the surface of the target test piece at the target moment after the second time range. Thus, the generative adversarial network model is trained, and the sound field distribution map of the electromagnetic ultrasonic transducer is generated by the generator in the generative adversarial network model, so as to achieve accurate sound field distribution generation.

[0080] In some embodiments, when the generator includes an input layer, a feature layer, a long short-term memory network LSTM layer, a fully connected layer and an output layer connected in sequence, in order to clearly understand the generator based on the generative adversarial network model, the process of generating a sound field distribution map of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece at the sample time is generated according to the first sound field distribution map sequence. Figure 4 , an exemplary description of the process is given.

[0081] Figure 4 It is a flowchart of a method for generating a sound field distribution diagram of an electromagnetic ultrasonic transducer according to another embodiment of the present application.

[0082] like Figure 4 As shown, a possible implementation of step 102 may include:

[0083] Step 401: input the first sound field distribution map sequence received by the input layer into the feature layer.

[0084] Step 402: Determine a sound field distribution feature map sequence corresponding to the first sound field distribution map sequence through a feature layer.

[0085] In this embodiment, after the first sound field distribution map sequence is input into the feature layer, the feature layer performs feature extraction on each first sound field distribution map in the first sound field distribution map sequence to obtain a sound field distribution feature map corresponding to each first sound field distribution map in the first sound field distribution map sequence, and obtains a sound field distribution feature map corresponding to the first sound field distribution map sequence based on the obtained sound field distribution feature map.

[0086] In some embodiments, when the above-mentioned feature layer includes: a first convolutional layer and a residual block layer, a possible implementation method of determining the sound field distribution feature map sequence corresponding to the first sound field distribution map sequence through the feature layer is: inputting the first sound field distribution map sequence into the first convolutional layer to obtain an intermediate sound field distribution feature map sequence; inputting the intermediate sound field distribution feature map sequence into the residual block layer to obtain a sound field distribution feature map sequence.

[0087] In this embodiment, after the intermediate sound field distribution feature map sequence is input into the residual block layer, residual connection and feature fusion can be performed on the intermediate sound field distribution feature map sequence through the residual block layer to obtain the sound field distribution feature map sequence.

[0088] It should be noted that the first convolutional layer in this embodiment may be one layer or multiple layers.

[0089] In some embodiments, the size of the convolution kernel in the first convolution layer can be 3×4×4, which is used to extract low-level features from the input sound field image and capture spatial information through the convolution operation. The corresponding convolution layer bias is (64), and a bias term is assigned to each convolution kernel to adjust the output after convolution.

[0090] In some embodiments, when the above-mentioned residual block layer includes a plurality of residual blocks connected in sequence, the intermediate sound field distribution feature map sequence is input into the residual block layer, and a possible implementation method for obtaining the sound field distribution feature map sequence is as follows: for the first residual block in the residual block layer, the intermediate sound field distribution feature map sequence is input into the first residual block, and the output result of the first residual block and the intermediate sound field distribution feature map sequence are added and input into the second residual block in the residual block layer; for the i-th residual block in the residual block layer, the output result of the i-1-th residual block is added. The result is input into the i-th residual block, and the output result of the i-1-th residual block and the output result of the i-th residual block are added and input into the i+1-th residual block, wherein i is an integer greater than 1 and less than N, and N represents the number of residual blocks in the residual block layer; for the N-th residual block in the residual block layer, the output result of the N-1-th residual block is input into the N-th residual block, and the intermediate sound field distribution feature map sequence, the output result of the N-1-th residual block and the output result of the N-th residual block are merged to obtain a sound field distribution feature map sequence.

[0091] It should be noted that, in this embodiment, there may be four residual blocks.

[0092] The residual block in this embodiment includes a plurality of second convolutional layers connected in sequence. For example, the number of the second convolutional layers may be two, that is, the residual block may include two second convolutional layers connected in sequence.

[0093] Step 403: Analyze the sound field distribution feature map sequence through the LSTM layer to obtain time series features.

[0094] In some embodiments, there may be multiple LSTM layers, for example, there may be four LSTM layers.

[0095] In some examples, in the LSTM layer, the weight matrix dimension of the first layer is (2048, 65536), with 2048 hidden units and an input size of 65536. This layer captures the long-term dependencies in the sound field signal by processing time series data. The weight matrix dimension of the second layer LSTM is (2048, 512), still contains 2048 hidden units, and the input size is 512, which further processes the features passed from the first layer. The bias of each layer is (2048) and is used to adjust the output of each LSTM unit.

[0096] Step 404 , input the time series features into the fully connected layer to obtain the acoustic field distribution features of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece at the sample time.

[0097] The weights and biases of the fully connected layer are used to map the output of the LSTM layer to a higher dimensional feature space.

[0098] In some embodiments, the fully connected layer includes a first fully connected layer and a second fully connected layer connected in sequence, wherein the weight dimension of the first fully connected layer is (32768, 512), the bias is (32768), and the first fully connected layer maps the output of the LSTM layer to a 512-dimensional feature space. The weight of the second fully connected layer is (768, 32768), and the bias is (768), which is used to map the 512-dimensional feature space to the 768-dimensional feature space.

[0099] Step 405: input the sound field distribution characteristics into the output layer to generate a sound field distribution map.

[0100] For example, the structure diagram of the generator is as follows: Figure 5 As shown, Figure 5 In this paper, four LSTM layers and four residual blocks are used as examples. Figure 5 The process of obtaining and generating a sound field distribution map is described exemplarily.

[0101] First, the input layer of the generator receives the first sequence of sound field distribution maps, which is used to represent the initial state of the EMAT sound field. The input image is then processed by a convolutional layer, which extracts low-level spatial features in the image through filters, then introduces nonlinear transformations using the ReLU activation function, and reduces the size of the feature map through the maximum pooling layer, thereby reducing the amount of computation and enhancing the robustness of the features. Then, the feature map processed by the convolutional layer enters four residual blocks, each of which contains two convolutional layers, and the input and output are added through residual connections. Residual connections help alleviate the gradient vanishing problem in deep networks and improve the learning efficiency and stability of the network. Each convolutional layer is followed by a ReLU activation function, which enables the network to capture more complex image features.

[0102] In order to process time series data and capture the evolution of EMAT sound field distribution over time, the model further introduces four LSTM layers. LSTM layers can capture long-term dependencies in time series and effectively model the temporal dynamic characteristics of the sound field. Through these LSTM layers, the model can learn the complex relationship between time steps from the spatiotemporal features extracted by convolution, thereby generating sound field distribution maps at future moments.

[0103] Finally, the time series features processed by the LSTM layer enter the fully connected layer to obtain the acoustic field distribution characteristics of the ultrasonic waves excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece at the sample time. A vector representing the acoustic field distribution characteristics is generated through the output layer, and finally reconstructed into the corresponding generated acoustic field distribution map. Through this structure, the Residual Connections Res-LSTM generator can effectively integrate spatial features and time series information, generate accurate acoustic field evolution, and significantly improve the accuracy of acoustic field distribution generation at different times.

[0104] In summary, the feature extraction capability of convolutional neural network (CNN) and residual connection mechanism are combined to have the following advantages: residual connection alleviates the gradient vanishing problem in deep networks by adding input and output, so that the network can propagate gradients more efficiently during training, thereby improving the stability of training. Res-LSTM can extract richer information from spatiotemporal features. LSTM captures long-term dependencies in time series, while residual blocks improve the network's ability to learn spatial features, thereby enhancing the overall expression ability of the model.

[0105] In some embodiments, when the discriminator includes: a third convolutional layer, a fully connected layer, and an output layer connected in sequence, in order to clearly understand the discriminator based on the generative adversarial network model, the process of determining the first discrimination result of the generated sound field distribution map and the second discrimination result of the sample sound field distribution map according to the generated sound field distribution map and the sample sound field distribution map is described below in combination with Figure 6This process is described exemplarily.

[0106] Figure 6 It is a flowchart of a method for generating a sound field distribution diagram of an electromagnetic ultrasonic transducer according to another embodiment of the present application.

[0107] like Figure 6 As shown, a possible implementation of step 103 may include:

[0108] Step 601 : extracting features of the generated sound field distribution map and the sample sound field distribution map through the third convolution layer to obtain feature information corresponding to the generated sound field distribution map and the sample sound field distribution map.

[0109] It should be noted that, in this embodiment, a plurality of third convolutional layers may be included. For example, the number of the third convolutional layers is four.

[0110] In this embodiment, the multi-level and multi-scale features in the generated sound field distribution map and the sample sound field distribution map are extracted by the third convolution layer.

[0111] Each convolutional layer can extract local features through convolution operations and reduce spatial dimensions through pooling operations, so that the network can capture more abstract high-level features. These convolutional layers effectively extract hierarchical features of the input image through multiple convolution and pooling operations.

[0112] Step 602 : weighting the feature information corresponding to the generated sound field distribution map and the sample sound field distribution map respectively through the fully connected layer to obtain the global information corresponding to the generated sound field distribution map and the sample sound field distribution map respectively.

[0113] In this embodiment, the fully connected layer is responsible for weighted combination of the extracted features to further learn the global information of the image.

[0114] Step 603 , obtaining a first discrimination result of the generated sound field distribution map and a second discrimination result of the sample sound field distribution map by respectively analyzing the global information corresponding to the generated sound field distribution map and the sample sound field distribution map through the output layer.

[0115] In this embodiment, the output layer uses the Sigmoid activation function to convert the output of the fully connected layer into a probability value, which represents the probability that the input image is a real image. The output range of the Sigmoid function is [0, 1], where a value close to 1 indicates that the image is real, and a value close to 0 indicates that the image is generated.

[0116] Among them, the structure example diagram of the discriminator is as follows Figure 7 shown.

[0117] For example, the discriminator includes four convolutional layers connected in sequence, namely the first convolutional layer, the second convolutional layer, the third convolutional layer and the fourth convolutional layer. The size of the convolution kernel in the first convolutional layer is 126×128×32, the size of the convolution kernel in the second convolutional layer is 256×64×16, the size of the convolution kernel in the third convolutional layer is 512×32×8, and the size of the convolution kernel in the fourth convolutional layer is 1024×16×4. Correspondingly, the input of the discriminator can be image data of size 2×256×64, which contains multiple channel information. Through the processing of four convolutional layers, the network gradually reduces the spatial dimension and increases the number of feature channels to extract multi-level and multi-scale features in the image. Each convolutional layer extracts local features through convolution operations and reduces the spatial dimension through pooling operations, so that the network can capture more abstract high-level features. These convolutional layers effectively extract the hierarchical features of the input image through multiple convolution and pooling operations.

[0118] In this embodiment, the discriminator can effectively extract local features and global information of the image, and combine the convolution operation and the fully connected layer for final classification. The application of the Sigmoid activation function enables the network to output a probability value for judging the authenticity of the input data. Therefore, the discriminator can efficiently classify the generated data, thereby promoting the generator to further improve the quality of the generated samples.

[0119] In order to clearly understand this application, Figure 8 The structural example diagram of the conditional generative adversarial network model is used to exemplify the process of training the conditional generative adversarial network model.

[0120] First, the first sound field distribution map sequence is input into the generator as the conditional variable y, which provides dynamic information about the sound field changing over time. The generator uses these first sound field distribution map sequences to guide the generation of sound field distribution maps at the sample moment. The conditional variable y is the input of the model, ensuring that the generator can generate sound field data that conforms to the actual physical process. The generator part adopts the Res-LSTM (Residual Long Short-Term Memory Network) architecture. The LSTM network can effectively capture long-term dependencies in time series, while the residual connection avoids the gradient vanishing problem in deep networks. Through this structure, the generator can generate high-quality sound field distribution maps based on the conditional variables, reflecting the evolution characteristics of the sound field at different time series.

[0121] Correspondingly, the output result of the generator is the sound field data obtained after processing by the LSTM network, which represents the time-series sound field map generated by the model under given conditions. In order to verify the authenticity of the generated sound field distribution map, the sample sound field distribution map and the generated sound field distribution map are input into the discriminator. The task of the discriminator is to evaluate the similarity between the sample sound field distribution map and the generated sound field distribution map. The discriminator compares the output of the generator with the actual simulation results and outputs a probability value to determine whether the generated data is "real". During the training process, the feedback of the discriminator helps the generator to continuously optimize its generation process, thereby improving the generation ability and prediction accuracy of the model. Through the adversarial training of the generator and the discriminator, the CGAN-LSTM model can generate sound field distribution maps that are closer and closer to the real data, significantly improving the performance of the model in the EMAT sound field generation task. This adversarial training mechanism enables the model to effectively capture the complex characteristics of the sound field evolving over time, providing an accurate solution for non-destructive testing and material evaluation.

[0122] On the basis of any of the above embodiments, based on the trained generator, according to the second sound field distribution diagram sequence of the ultrasonic wave excited by the target electromagnetic ultrasonic transducer on the surface of the target test piece within the second time range, before generating the sound field distribution diagram of the ultrasonic wave excited on the surface of the target test piece by the target electromagnetic ultrasonic transducer at the target moment after the second time range, the trained generator and discriminator can also be verified by verification data to determine that the accuracy exceeds a preset accuracy threshold.

[0123] Fig. 9 It is a structural schematic diagram of a sound field distribution diagram generating device of an electromagnetic ultrasonic transducer according to an embodiment of the present application.

[0124] like Fig. 9 As shown, the sound field distribution diagram generating device 900 of the electromagnetic ultrasonic transducer includes:

[0125] The acquisition module 901 acquires a first sound field distribution diagram sequence of ultrasonic waves excited on the surface of the sample test piece by the sample electromagnetic ultrasonic transducer within a first time range and a sample sound field distribution diagram of ultrasonic waves excited on the surface of the sample test piece at a sample time after the first time range.

[0126] The first generating module 902 is used to generate a generated sound field distribution diagram of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece at the sample time according to the generator in the generative adversarial network model and the first sound field distribution diagram sequence.

[0127] The second generating module 903 is used to determine a first discrimination result of the generated sound field distribution map and a second discrimination result of the sample sound field distribution map based on the discriminator in the generative adversarial network model and the generated sound field distribution map and the sample sound field distribution map.

[0128] The training module 904 is used to train the generator and the discriminator according to the generated sound field distribution map, the sample sound field distribution map, the first discrimination result and the second discrimination result.

[0129] The third generating module 905 is used to generate, based on the trained generator and according to a second sound field distribution diagram sequence of the ultrasonic wave excited by the target electromagnetic ultrasonic transducer on the surface of the target test piece within the second time range, a sound field distribution diagram of the ultrasonic wave excited by the target electromagnetic ultrasonic transducer on the surface of the target test piece at a target moment after the second time range.

[0130] In some embodiments, obtaining a first sound field distribution diagram sequence of ultrasonic waves excited by a sample electromagnetic ultrasonic transducer on a surface of a sample test piece within a first time range includes: constructing an electromagnetic ultrasonic transducer model of the sample electromagnetic ultrasonic transducer and a test piece model of the sample test piece; performing simulation according to the electromagnetic ultrasonic transducer model and the test piece model to obtain a simulated sound field distribution diagram sequence of ultrasonic waves excited by the electromagnetic ultrasonic transducer model on the surface of the test piece model within the first time range; and determining a first sound field distribution diagram sequence of ultrasonic waves excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece within the first time range according to the simulated sound field distribution diagram sequence.

[0131] In some embodiments, the generator includes an input layer, a feature layer, a long short-term memory network LSTM layer, a fully connected layer, and an output layer connected in sequence.

[0132] The first generating module 902 is specifically used for:

[0133] Inputting the first sound field distribution map sequence received by the input layer into the feature layer;

[0134] Determine a sound field distribution feature map sequence corresponding to the first sound field distribution map sequence through the feature layer;

[0135] The sound field distribution feature map sequence is analyzed through the LSTM layer to obtain the time series features;

[0136] The time series features are input into the fully connected layer to obtain the acoustic field distribution features of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece at the sample time;

[0137] The sound field distribution features are input into the output layer to generate the sound field distribution map.

[0138] In some embodiments, the feature layer includes: a first convolutional layer and a residual block layer;

[0139] The step of determining the sound field distribution feature map sequence corresponding to the first sound field distribution map sequence through the feature layer includes:

[0140] Inputting the first sound field distribution map sequence into the first convolutional layer to obtain an intermediate sound field distribution feature map sequence;

[0141] The intermediate sound field distribution feature map sequence is input into the residual block layer to obtain the sound field distribution feature map sequence.

[0142] In some embodiments, the residual block layer includes a plurality of residual blocks connected in sequence;

[0143] The intermediate sound field distribution feature map sequence is input into the residual block layer to obtain the sound field distribution feature map sequence, including:

[0144] For the first residual block in the residual block layer, the intermediate sound field distribution feature map sequence is input into the first residual block, and the output result of the first residual block and the intermediate sound field distribution feature map sequence are added and input into the second residual block in the residual block layer;

[0145] For the i-th residual block in the residual block layer, the output result of the i-1-th residual block is input into the i-th residual block, and the output result of the i-1-th residual block is added to the output result of the i-th residual block and then input into the i+1-th residual block, where i is an integer greater than 1 and less than N, and N represents the number of residual blocks in the residual block layer;

[0146] For the Nth residual block in the residual block layer, the output result of the N-1th residual block is input into the Nth residual block, and the intermediate sound field distribution feature map sequence, the output result of the N-1th residual block and the output result of the Nth residual block are merged to obtain the sound field distribution feature map sequence.

[0147] In some embodiments, the residual block includes a plurality of second convolutional layers connected in sequence.

[0148] In some embodiments, the discriminator includes: a third convolutional layer, a fully connected layer, and an output layer connected in sequence,

[0149] The second generating module 903 is specifically used for:

[0150] The feature extraction of the generated sound field distribution map and the sample sound field distribution map is performed through the third convolution layer to obtain feature information corresponding to the generated sound field distribution map and the sample sound field distribution map;

[0151] Through the fully connected layer, the feature information corresponding to the generated sound field distribution map and the sample sound field distribution map is weighted to obtain the global information corresponding to the generated sound field distribution map and the sample sound field distribution map;

[0152] The global information corresponding to the generated sound field distribution map and the sample sound field distribution map is respectively processed through the output layer to obtain a first discrimination result of the generated sound field distribution map and a second discrimination result of the sample sound field distribution map.

[0153] In some embodiments, the training module 904 is specifically used to: determine the total loss value of the generative adversarial network model based on the generated sound field distribution map, the sample sound field distribution map, the first discrimination result and the second discrimination result; according to the total loss value, alternately train the generator and the discriminator until the preset training end conditions are met to obtain the trained generator and discriminator.

[0154] In some embodiments, the training module 904 is specifically used to: determine a first loss value of the generator based on a generated sound field distribution map and a sample sound field distribution map; determine a second loss value of the discriminator based on a first discrimination result and a second discrimination result; train the generator based on the first loss value; train the discriminator based on the second loss value; and alternately perform the steps of training the generator and the discriminator until a training end condition is met.

[0155] According to an embodiment of the present application, the present application also provides an electronic device.

[0156] Fig.10 It is a structural block diagram of an electronic device according to an embodiment of the present application.

[0157] like Fig.10 As shown, the electronic device 1000 includes: a memory 1010, a processor 1020, and computer instructions stored in the memory 1010 and executable on the processor 1020.

[0158] When the processor 1020 executes the instructions, the method for generating the sound field distribution diagram of the electromagnetic ultrasonic transducer provided in the above embodiment is implemented.

[0159] Furthermore, the electronic device 1000 further includes:

[0160] The communication interface 1030 is used for communication between the memory 1010 and the processor 1020 .

[0161] The memory 1010 is used to store computer instructions that can be executed on the processor 1020 .

[0162] The memory 1010 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0163] The processor 1020 is used to implement the method for generating a sound field distribution diagram of the electromagnetic ultrasonic transducer of the above embodiment when executing the program.

[0164] If the memory 1010, the processor 1020 and the communication interface 1030 are implemented independently, the communication interface 1030, the memory 1010 and the processor 1020 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.10 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0165] Optionally, in a specific implementation, if the memory 1010, the processor 1020 and the communication interface 1030 are integrated on a chip, the memory 1010, the processor 1020 and the communication interface 1030 can communicate with each other through an internal interface.

[0166] The processor 1020 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0167] On the other hand, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for generating a sound field distribution diagram of an electromagnetic ultrasonic transducer disclosed in an embodiment of the present application is implemented.

[0168] Another aspect of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the method for generating a sound field distribution diagram of the electromagnetic ultrasonic transducer of the embodiment of the present application.

[0169] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. 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 representations of the above terms do 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. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0170] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0171] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for generating a sound field distribution diagram of an electromagnetic ultrasonic transducer, characterized in that: The method comprises: Acquire a sequence of first acoustic field distribution diagrams of ultrasonic waves excited on the surface of a sample test piece by a sample electromagnetic ultrasonic transducer within a first time range and a sample acoustic field distribution diagram of ultrasonic waves excited on the surface of the sample test piece at a sample time after the first time range; Based on the generator in the generative adversarial network model, generating a generated sound field distribution map of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece at the sample moment according to the first sound field distribution map sequence; Based on the discriminator in the generative adversarial network model, determining a first discrimination result of the generated sound field distribution map and a second discrimination result of the sample sound field distribution map according to the generated sound field distribution map and the sample sound field distribution map; Training the generator and the discriminator according to the generated sound field distribution map, the sample sound field distribution map, the first discrimination result, and the second discrimination result; Based on the trained generator, according to the second sound field distribution diagram sequence of the ultrasonic wave excited by the target electromagnetic ultrasonic transducer on the surface of the target test piece within the second time range, a sound field distribution diagram of the ultrasonic wave excited by the target electromagnetic ultrasonic transducer on the surface of the target test piece at the target moment after the second time range is generated.

2. The method according to claim 1, characterized in that The step of obtaining a first sound field distribution diagram sequence of ultrasonic waves excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece within a first time range includes: Constructing an electromagnetic ultrasonic transducer model of the sample electromagnetic ultrasonic transducer and a test piece model of the sample test piece; Performing simulation according to the electromagnetic ultrasonic transducer model and the test piece model to obtain a sequence of simulated sound field distribution diagrams of the ultrasonic wave excited by the electromagnetic ultrasonic transducer model on the surface of the test piece model within a first time range; According to the simulated sound field distribution diagram sequence, a first sound field distribution diagram sequence of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece within a first time range is determined.

3. The method according to claim 1, characterized in that The generator includes an input layer, a feature layer, a long short-term memory network LSTM layer, a fully connected layer and an output layer connected in sequence. Wherein, based on the generator in the generative adversarial network model, generating a generated sound field distribution map of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece at the sample moment according to the first sound field distribution map sequence, including: Inputting the first sound field distribution map sequence received by the input layer into the feature layer; Determine, by means of the feature layer, a sequence of sound field distribution feature maps corresponding to the first sequence of sound field distribution maps; The sound field distribution feature map sequence is analyzed by the LSTM layer to obtain a time series feature; Inputting the time series feature into the fully connected layer to obtain the acoustic field distribution feature of the ultrasonic wave excited by the sample electromagnetic ultrasonic transducer on the surface of the sample test piece at the sample time; The sound field distribution characteristics are input into the output layer to obtain the generated sound field distribution map.

4. The method according to claim 3, characterized in that The feature layer includes: a first convolutional layer and a residual block layer; Wherein, determining the sound field distribution feature map sequence corresponding to the first sound field distribution map sequence through the feature layer includes: Inputting the first sound field distribution map sequence into the first convolutional layer to obtain an intermediate sound field distribution feature map sequence; The intermediate sound field distribution feature map sequence is input into the residual block layer to obtain the sound field distribution feature map sequence.

5. The method according to claim 4, characterized in that The residual block layer includes a plurality of residual blocks connected in sequence; The step of inputting the intermediate sound field distribution feature map sequence into the residual block layer to obtain the sound field distribution feature map sequence comprises: For a first residual block in the residual block layer, input the intermediate sound field distribution feature map sequence into the first residual block, and add an output result of the first residual block and the intermediate sound field distribution feature map sequence and input the result to a second residual block in the residual block layer; For the i-th residual block in the residual block layer, input the output result of the i-1th residual block into the i-th residual block, and add the output result of the i-1th residual block and the output result of the i-th residual block and input them into the i+1th residual block, where i is an integer greater than 1 and less than N, and N represents the number of residual blocks in the residual block layer; For the Nth residual block in the residual block layer, the output result of the N-1th residual block is input into the Nth residual block, and the intermediate sound field distribution feature map sequence, the output result of the N-1th residual block and the output result of the Nth residual block are merged to obtain the sound field distribution feature map sequence.

6. The method according to claim 5, characterized in that The residual block includes a plurality of second convolutional layers connected in sequence.

7. The method according to claim 1, characterized in that The discriminator comprises: a third convolutional layer, a fully connected layer and an output layer connected in sequence, The method based on the discriminator in the generative adversarial network model, determining a first discrimination result of the generated sound field distribution map and a second discrimination result of the sample sound field distribution map according to the generated sound field distribution map and the sample sound field distribution map, comprises: Performing feature extraction on the generated sound field distribution map and the sample sound field distribution map through the third convolution layer to obtain feature information corresponding to the generated sound field distribution map and the sample sound field distribution map respectively; Through the fully connected layer, weighted processing is performed on the feature information corresponding to the generated sound field distribution map and the sample sound field distribution map, respectively, to obtain the global information corresponding to the generated sound field distribution map and the sample sound field distribution map; The output layer respectively processes the global information corresponding to the generated sound field distribution map and the sample sound field distribution map to obtain a first discrimination result of the generated sound field distribution map and a second discrimination result of the sample sound field distribution map.

8. The method according to any one of claims 1 to 7, characterized in that The training of the generator and the discriminator according to the generated sound field distribution map, the sample sound field distribution map, the first discrimination result and the second discrimination result comprises: Determining a total loss value of the generative adversarial network model according to the generated sound field distribution map, the sample sound field distribution map, the first discrimination result, and the second discrimination result; According to the total loss value, the generator and the discriminator are alternately trained until a preset training end condition is met, thereby obtaining a trained generator and discriminator.

9. The method according to any one of claims 1 to 7, characterized in that The training of the generator and the discriminator according to the generated sound field distribution map, the sample sound field distribution map, the first discrimination result and the second discrimination result comprises: Determine a first loss value of the generator according to the generated sound field distribution map and the sample sound field distribution map; Determine a second loss value of the discriminator according to the first discrimination result and the second discrimination result; Training the generator according to the first loss value; Training the discriminator according to the second loss value; The steps of training the generator and the discriminator are performed alternately until a training end condition is met, thereby obtaining a trained generator and discriminator.

10. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 9 when executing the computer program.

Images