A method and system for denoising power frequency harmonic noise based on semi-supervised learning

Abstract

The present application provides a method and system for denoising power frequency harmonic noise based on semi-supervised learning, which belongs to the technical field of electromagnetic exploration signal denoising. The method includes: using the original noise-free data set as a training set and the original noisy data set as a label set, inputting the data set into the power frequency harmonic noise extraction network to generate a simulated noise-free data set; inputting the simulated noise-free data set into the power frequency harmonic noise extraction network to obtain a simulated noisy data set; inputting the simulated noisy data set and the original noisy data set into the power frequency harmonic denoising network, comparing the output noise-free data set with the simulated noise-free data, judging whether the generation mode of the power frequency harmonic noise extracted by the power frequency harmonic noise extraction network reaches the preset target, and when the preset target is reached, using the noise-free data set and the noisy data set to train the power frequency harmonic denoising network. The power frequency harmonic denoising network model obtained by the present application can quickly remove power frequency harmonic noise.

Description

Technical Field

[0001] The present application belongs to the technical field of electromagnetic exploration signal denoising, and more specifically, to a method and system for denoising power frequency harmonic noise based on semi-supervised learning. Background Art

[0002] With the rapid advancement of informatization, human electromagnetic noise is becoming increasingly serious. In electromagnetic exploration, power frequency and its harmonic interference are particularly prominent. Transient electromagnetic measurement uses an ungrounded coil to transmit a transient magnetic field underground. Usually, a current square wave is supplied to the transmitting coil, which can generate a stable magnetic field distribution underground. When the current square wave is turned off, eddy currents will be generated underground. The eddy current will not disappear immediately, but will generate a secondary field to propagate to the surface in the process of gradual disappearance. The surface receiving coil is used to convert the change of the magnetic field into the change of the induced voltage. When there is power frequency interference, the change of the induced voltage in the receiving coil also contains the power frequency harmonic interference. The effective signal and interference are received by the system through a unified path, and the coil reception cannot separate the two. The industrial power frequency is 50Hz, but in practice the frequency is not fixed and will drift around 50Hz. The distortion of its voltage and current will also produce 150Hz, 250Hz or even higher harmonics. These powerful interference fields are often dozens of times the target signal, which brings a lot of trouble to the conventional magnetotelluric data processing. Therefore, an efficient method for removing power frequency harmonics is needed. Summary of the invention

[0003] In view of the defects of the prior art, the purpose of this application is to provide a method and system for denoising power frequency harmonic noise based on semi-supervised learning, aiming to solve the problem that in the existing transient electromagnetic method measurement process, when there is power frequency interference, the change of the induced voltage in the receiving coil also includes the power frequency harmonic interference, and the effective signal and interference are received by the system through a unified path, resulting in the inability to separate the effective signal and interference by using coil reception.

[0004] To achieve the above objectives, in a first aspect, the present application provides a method for denoising power frequency harmonic noise based on semi-supervised learning, comprising the following steps:

[0005] The original noise-free data set is used as the training set, and the original noisy data set is used as the label set, which is input into the power frequency harmonic noise extraction network for preliminary training to generate a simulated noise-free data set;

[0006] Inputting the simulated noise-free data set into the preliminarily trained power frequency harmonic noise extraction network for further training to obtain a simulated noisy data set;

[0007] The simulated noisy data set and the original noisy data set are input into the power frequency harmonic denoising network, and the noise-free data set output by the power frequency harmonic denoising network is compared with the simulated noise-free data to determine whether the generation mode of the power frequency harmonic noise extracted by the power frequency harmonic noise extraction network reaches the preset target. When the preset target is reached, the simulated noise-free data set and the original noise-free data set are used as noise-free data sets, and the simulated noisy data set and the original noisy data set are used as noisy data sets. The noise-free data set and the noisy data set are used to train the power frequency harmonic denoising network to remove the power frequency harmonic noise.

[0008] Further preferably, the original noise-free data set is obtained by:

[0009] Arrange the original noisy data set into a format that can train the power frequency harmonic noise extraction network;

[0010] The original noisy data set is denoised by the least squares method to obtain the original noise-free data set.

[0011] Further preferably, the preset target is that the degree of similarity between the noise-free data set output by the power frequency harmonic denoising network and the simulated noise-free data set is not less than a preset threshold.

[0012] Further preferably, when the generation mode of the power frequency harmonic noise extracted by the power frequency harmonic noise extraction network fails to reach a preset target, the parameters of the power frequency harmonic noise extraction network are readjusted.

[0013] Further preferably, the power frequency harmonic noise extraction network is a deep learning neural network, which includes, in the order of signal processing, an input layer, a first convolutional layer, a pooling layer, a second convolutional layer, a batch normalization layer, a pooling layer, a third convolutional layer, a batch normalization layer, a pooling layer, a fourth convolutional layer, a fifth convolutional layer, a sixth convolutional layer, a pooling layer, a seventh convolutional layer, an eighth convolutional layer, a ninth convolutional layer, a pooling layer, a tenth convolutional layer, a pooling layer, a flattening layer, a first fully connected layer, a second fully connected layer, a third fully connected layer and an output layer.

[0014] Further preferably, the trained power frequency harmonic denoising network is used to remove power frequency harmonic interference during transient electromagnetic method measurement.

[0015] In a second aspect, the present application provides a power frequency harmonic noise denoising system based on semi-supervised learning, comprising:

[0016] The first data set generation module is used to use the original noise-free data set as a training set and the original noisy data set as a label set, input them into the power frequency harmonic noise extraction network for preliminary training, and generate a simulated noise-free data set;

[0017] A second data set acquisition module is used to input the simulated noise-free data set into the preliminarily trained power frequency harmonic noise extraction network for further training to obtain a simulated noisy data set;

[0018] A first network training module, used for inputting the simulated noisy data set and the original noisy data set into the power frequency harmonic denoising network;

[0019] A model determination module is used to compare the noise-free data set output by the power frequency harmonic denoising network with the simulated noise-free data to determine whether the generation mode of the power frequency harmonic noise extracted by the power frequency harmonic noise extraction network reaches the preset target;

[0020] A third data set acquisition module is used to, when a preset target is reached, use the simulated noise-free data set and the original noise-free data set as the noise-free data set, and use the simulated noisy data set and the original noisy data set as the noisy data set;

[0021] The second network training module is used to train the power frequency harmonic denoising network using the noise-free data set and the noise-containing data set to remove the power frequency harmonic noise.

[0022] Further preferably, the power frequency harmonic noise denoising system also includes a fourth data set acquisition module, which is used to arrange the original noisy data set into a format capable of training the power frequency harmonic noise extraction network; denoise the original noisy data set by the least squares method to obtain the original noise-free data set.

[0023] Further preferably, the preset target is that the degree of similarity between the noise-free data set output by the power frequency harmonic denoising network and the simulated noise-free data set is not less than a preset threshold.

[0024] Further preferably, the power frequency harmonic noise denoising system further includes: a parameter adjustment module, which is used to readjust the parameters of the power frequency harmonic noise extraction network when the generation mode of the power frequency harmonic noise extracted by the power frequency harmonic noise extraction network does not reach a preset target.

[0025] In a third aspect, the present application provides an electronic device comprising: at least one memory for storing programs; and at least one processor for executing the programs stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method described in the first aspect or any possible implementation of the first aspect.

[0026] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.

[0027] In a fifth aspect, the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.

[0028] It can be understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0029] In general, the above technical solutions conceived by this application have the following beneficial effects compared with the prior art:

[0030] The present application provides a method for denoising power frequency harmonic noise based on semi-supervised learning, which can construct a corresponding power frequency harmonic noise extraction network model according to the noise generation pattern in the signal, and is targeted at fixed power frequency harmonic noise. Based on semi-supervised learning (with the original noise-free data set as the training set and the original noisy data set as the label set), a large number of simulated noise-free data sets that conform to the actual noise generation pattern can be generated through a small amount of original noise-free data sets, thereby enhancing the generalization of the power frequency harmonic noise extraction network model. The simulated noise-free data set is then input into the power frequency harmonic noise extraction network after preliminary training for further training, and a simulated noisy data set is obtained, which provides a large number of noise-free data sets and noisy data sets for the power frequency harmonic denoising network for training, and the power frequency harmonic noise can be quickly removed after the trained power frequency harmonic denoising network model is obtained.

[0031] The present application provides a method for denoising power frequency harmonic noise based on semi-supervised learning, which compares a noise-free data set output by a power frequency harmonic denoising network with simulated noise-free data, and determines whether a generation mode of power frequency harmonic noise extracted by a power frequency harmonic noise extraction network reaches a preset target. When the generation mode of power frequency harmonic noise extracted by the power frequency harmonic noise extraction network does not reach the preset target, the parameters of the power frequency harmonic noise extraction network are readjusted; when the preset target is reached, the simulated noise-free data set is used together with the original noise-free data set as the noise-free data set, thereby making the training data set of the power frequency harmonic denoising network more accurate, and ensuring that the trained power frequency harmonic denoising network model can quickly remove power frequency harmonic noise. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flow chart of a method for denoising power frequency harmonic noise based on semi-supervised learning provided in an embodiment of the present application;

[0033] Figure 2 It is a network structure diagram of power frequency harmonic noise extraction provided by an embodiment of the present application;

[0034] Figure 3 is a training flow chart of a power frequency harmonic denoising network provided in an embodiment of the present application;

[0035] Figure 4 is an unprocessed noise image provided by an embodiment of the present application;

[0036] Figure 5 This is a processing effect diagram provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0038] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0039] Example 1

[0040] like Figure 1 As shown, the present application provides a method for denoising power frequency harmonic noise based on semi-supervised learning, comprising the following steps:

[0041] Obtaining an original noisy data set, and rearranging the original noisy data set into a format that can be trained by a power frequency harmonic noise extraction network; wherein the data format required for training the power frequency harmonic noise extraction network and the power frequency harmonic denoising network is the same;

[0042] The original noisy data set is preliminarily denoised by the least square method to obtain a noise-free data set, and the original noisy data set and the corresponding noise-free data set are divided into multiple training subsets, and the number of data items in the multiple training subsets is equal, and the power frequency harmonic noise extraction network and the power frequency harmonic denoising network are pre-trained by the training data;

[0043] The power frequency harmonic noise extraction network is a deep learning neural network, more specifically, the deep learning network model is a convolutional neural network; the deep learning neural network includes, in order of signal processing, an input layer, a first convolutional layer, a pooling layer, a second convolutional layer, a batch normalization layer, a pooling layer, a third convolutional layer, a batch normalization layer, a pooling layer, a fourth convolutional layer, a fifth convolutional layer, a sixth convolutional layer, a pooling layer, a seventh convolutional layer, an eighth convolutional layer, a ninth convolutional layer, a pooling layer, a tenth convolutional layer, a pooling layer, a flattening layer, a first fully connected layer, a second fully connected layer, a third fully connected layer and an output layer;

[0044] The input layer size is 1000×1, and the signal is input to the first convolutional layer;

[0045] The convolution kernel size of the first, second, and third convolution layers is 2×1, the number of convolution kernels is 64, the stride is 1, and the ReLU activation function is used;

[0046] The pooling layer is the maximum pooling layer, the pooling window is 2, and the stride is 2;

[0047] The batch normalization layer is used to perform batch normalization on the output of the convolutional layer;

[0048] From the fourth to the sixth convolutional layers and from the seventh to the ninth convolutional layers, the fourth, fifth and sixth convolutional layers have 64, 128 and 256 convolutional kernels respectively, and the seventh, eighth and ninth convolutional layers have 64, 128 and 256 convolutional kernels respectively; the convolutional kernel size is 2×1, the stride is 1, and the ReLU activation function is used;

[0049] The convolution kernel size of the tenth convolution layer is 2×1, the number of convolution kernels is 512, the stride is 1, and the ReLU activation function is used;

[0050] The first fully connected layer contains 128 neurons, uses the ReLU activation function, and has an L2 regularization term with a regularization coefficient of 0.01;

[0051] The second fully connected layer contains 64 neurons, uses the ReLU activation function, and has an L2 regularization term with a regularization coefficient of 0.01;

[0052] The output layer size is 1000×1;

[0053] The power frequency harmonic noise extraction network uses the noise-free data set as the training set and the original noisy data set as the label set. After pre-training, the generation pattern of power frequency harmonic noise is initially obtained.

[0054] The power frequency harmonic denoising network uses the original noisy data set as the training set and the non-noisy data set as the label set. The trained power frequency harmonic denoising network model is used as the discriminator. The detailed network structure is as follows: Figure 2 As shown;

[0055] After the preliminary power frequency harmonic noise extraction network is trained, a large number of simulated noise-free data sets are generated according to the noise-free data set, and a large number of noise-free data sets are input into the pre-trained power frequency harmonic noise extraction network to obtain simulated noisy data sets, and then the simulated noisy data sets and the original noisy data sets are input into the pre-trained power frequency harmonic denoising network, and the obtained results are input into the discriminator, that is, the obtained results are compared with the corresponding simulated noise-free data sets to determine whether the power frequency harmonic noise extraction network accurately extracts the generation mode of the power frequency harmonic noise. If it cannot be accurately extracted, the parameters of the power frequency harmonic noise extraction network are adjusted and retrained. If it can be accurately extracted, the generated simulated noise-free data set and the simulated noisy data set are respectively classified into the noise-free data set and the original noisy data set;

[0056] After obtaining a large number of training sets, use a large amount of data to train the power frequency harmonic denoising network. The detailed steps are as follows: Figure 3 As shown;

[0057] After the power frequency harmonic denoising network training is completed, the trained power frequency harmonic denoising network model is saved; when using it, the original signal needs to be preprocessed in the same way as the training model, and the data format is adjusted to a format that the power frequency harmonic denoising network model can denoise; the power frequency harmonic denoising network model obtained after training only removes the power frequency noise of the same noise mode in the same area; after denoising the trained model, a clean signal without power frequency harmonic noise can be obtained;

[0058] like Figure 4 The figure shows a simulated outdoor noisy signal, which contains random noise and power frequency harmonic noise. The noise pattern consists of two forms: regular and irregular. This application can preliminarily remove most of the original noise through the least squares method to obtain a preliminary denoised signal. The signal data set can be converted into a training set for training. Figure 3 The training shown in the figure obtains the trained power frequency harmonic denoising network. Figure 4 The simulated outdoor noisy signal is input into the power frequency harmonic denoising network to obtain a noise-free signal with power frequency noise and random noise removed, such as Figure 5 As shown in the figure, the processing results show that the network can extract and remove the power frequency harmonic noise well, and the denoised signal is very close to the original signal.

[0059] Example 2

[0060] The embodiment of the present application provides a power frequency harmonic noise denoising system based on semi-supervised learning, comprising:

[0061] The first data set generation module is used to use the original noise-free data set as a training set and the original noisy data set as a label set, input them into the power frequency harmonic noise extraction network for preliminary training, and generate a simulated noise-free data set;

[0062] A second data set acquisition module is used to input the simulated noise-free data set into the preliminarily trained power frequency harmonic noise extraction network for further training to obtain a simulated noisy data set;

[0063] A first network training module, used for inputting the simulated noisy data set and the original noisy data set into the power frequency harmonic denoising network;

[0064] A model determination module is used to compare the noise-free data set output by the power frequency harmonic denoising network with the simulated noise-free data to determine whether the generation mode of the power frequency harmonic noise extracted by the power frequency harmonic noise extraction network reaches the preset target;

[0065] A third data set acquisition module is used to, when a preset target is reached, use the simulated noise-free data set and the original noise-free data set as the noise-free data set, and use the simulated noisy data set and the original noisy data set as the noisy data set;

[0066] The second network training module is used to train the power frequency harmonic denoising network using the noise-free data set and the noise-containing data set to remove the power frequency harmonic noise.

[0067] Further preferably, the power frequency harmonic noise denoising system also includes a fourth data set acquisition module, which is used to arrange the original noisy data set into a format capable of training the power frequency harmonic noise extraction network; denoise the original noisy data set by the least squares method to obtain the original noise-free data set.

[0068] Further preferably, the preset target is that the degree of similarity between the noise-free data set output by the power frequency harmonic denoising network and the simulated noise-free data set is not less than a preset threshold.

[0069] Further preferably, the power frequency harmonic noise denoising system further includes: a parameter adjustment module, which is used to readjust the parameters of the power frequency harmonic noise extraction network when the generation mode of the power frequency harmonic noise extracted by the power frequency harmonic noise extraction network does not reach a preset target.

[0070] Compared with the prior art, this application has the following advantages:

[0071] The present application can construct a corresponding power frequency harmonic noise extraction network model according to the noise generation mode in the signal, and is targeted at fixed power frequency harmonic noise.

[0072] This application is based on semi-supervised learning, which can generate a large amount of simulated data that conforms to the actual noise generation pattern through a small amount of preprocessed data, thereby enhancing the generalization of the model.

[0073] This application can quickly remove power frequency harmonic noise after obtaining a trained power frequency harmonic denoising network model.

[0074] It should be understood that the above-mentioned system is used to execute the methods in the above-mentioned embodiments. The implementation principles and technical effects of the corresponding program modules in the system are similar to those described in the above-mentioned methods. The working process of the system can refer to the corresponding process in the above-mentioned method and will not be repeated here.

[0075] Based on the method in the above embodiment, an embodiment of the present application provides an electronic device. The device may include: at least one memory for storing programs and at least one processor for executing the programs stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method described in the above embodiment.

[0076] Based on the method in the above embodiment, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method in the above embodiment.

[0077] Based on the method in the above embodiment, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method in the above embodiment.

[0078] It is understood that the processor in the embodiments of the present application may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0079] The method steps in the embodiments of the present application can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, and the software modules can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC.

[0080] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions may be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions may be transmitted from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state disk (SSD)), etc.

[0081] It should be understood that the various numerical numbers involved in the embodiments of the present application are only used for the convenience of description and are not used to limit the scope of the embodiments of the present application.

[0082] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for denoising power frequency harmonic noise based on semi-supervised learning, characterized in that: The following steps are involved: The original noise-free data set is used as a training set, and the original noisy data set is used as a label set, which are input into a power frequency harmonic noise extraction network for preliminary training. After the preliminary power frequency harmonic noise extraction network is obtained through training, a simulated noise-free data set is generated based on the noise-free data set. Inputting the simulated noise-free data set into the preliminarily trained power frequency harmonic noise extraction network for further training to obtain a simulated noisy data set; Input the simulated noisy data set and the original noisy data set into the pre-trained power frequency harmonic denoising network, compare the noise-free data set output by the power frequency harmonic denoising network with the simulated noise-free data set, judge whether the generation mode of the power frequency harmonic noise extracted by the power frequency harmonic noise extraction network reaches the preset target, when the preset target is reached, take the simulated noise-free data set and the original noise-free data set as the noise-free data set, take the simulated noisy data set and the original noisy data set as the noisy data set, use the noise-free data set and the noisy data set to train the power frequency harmonic denoising network, so as to remove the power frequency harmonic noise; Wherein, when the generation mode of the power frequency harmonic noise extracted by the power frequency harmonic noise extraction network does not reach the preset target, the parameters of the power frequency harmonic noise extraction network are readjusted; Among them, the trained power frequency harmonic denoising network is used to remove the power frequency harmonic interference in the transient electromagnetic method measurement process.

2. The method for denoising power frequency harmonic noise according to claim 1, characterized in that: The original noise-free dataset is obtained as follows: Arrange the original noisy data set into a format that can train the power frequency harmonic noise extraction network; The original noisy data set is denoised by the least squares method to obtain the original noise-free data set.

3. The method for denoising power frequency harmonic noise according to claim 1 or 2, characterized in that: The preset goal is that the similarity between the noise-free data set output by the power frequency harmonic denoising network and the simulated noise-free data set is not less than a preset threshold.

4. The method for denoising power frequency harmonic noise according to claim 1, characterized in that: The power frequency harmonic noise extraction network is a deep learning neural network, which includes, in the order of signal processing, an input layer, a first convolutional layer, a pooling layer, a second convolutional layer, a batch normalization layer, a pooling layer, a third convolutional layer, a batch normalization layer, a pooling layer, a fourth convolutional layer, a fifth convolutional layer, a sixth convolutional layer, a pooling layer, a seventh convolutional layer, an eighth convolutional layer, a ninth convolutional layer, a pooling layer, a tenth convolutional layer, a pooling layer, a flattening layer, a first fully connected layer, a second fully connected layer, a third fully connected layer and an output layer.

5. A power frequency harmonic noise denoising system based on semi-supervised learning, characterized in that: include: The first data set generation module is used to use the original noise-free data set as a training set and the original noisy data set as a label set, input them into the power frequency harmonic noise extraction network for preliminary training, and after the preliminary power frequency harmonic noise extraction network is obtained through training, generate a simulated noise-free data set according to the noise-free data set; A second data set acquisition module is used to input the simulated noise-free data set into the preliminarily trained power frequency harmonic noise extraction network for further training to obtain a simulated noisy data set; A first network training module, used for inputting the simulated noisy data set and the original noisy data set into a pre-trained power frequency harmonic denoising network; A model determination module is used to compare the noise-free data set output by the power frequency harmonic denoising network with the simulated noise-free data set to determine whether the generation mode of the power frequency harmonic noise extracted by the power frequency harmonic noise extraction network reaches a preset target; A third data set acquisition module is used to, when a preset target is reached, use the simulated noise-free data set and the original noise-free data set as the noise-free data set, and use the simulated noisy data set and the original noisy data set as the noisy data set; The second network training module is used to train the power frequency harmonic denoising network using a noise-free data set and a noise-containing data set to remove the power frequency harmonic noise; A parameter adjustment module, used for readjusting the parameters of the power frequency harmonic noise extraction network when the generation mode of the power frequency harmonic noise extracted by the power frequency harmonic noise extraction network does not reach a preset target; Among them, the trained power frequency harmonic denoising network is used to remove the power frequency harmonic interference in the transient electromagnetic method measurement process.

6. The power frequency harmonic noise denoising system according to claim 5, characterized in that: It also includes a fourth data set acquisition module, which is used to arrange the original noisy data set into a format that can train the power frequency harmonic noise extraction network; denoise the original noisy data set by the least squares method to obtain the original noise-free data set.

7. The power frequency harmonic noise denoising system according to claim 5, characterized in that: The preset goal is that the similarity between the noise-free data set output by the power frequency harmonic denoising network and the simulated noise-free data set is not less than a preset threshold.

Images