Weak magnetic field three-dimensional measurement big data analysis system based on Faraday effect

Abstract

The invention discloses a weak magnetic field three-dimensional measurement big data analysis system based on a Faraday effect. The system comprises a magnetic field excitation module, a polarization modulation module, an optical signal acquisition module, a data processing module, a network training module and a result output module. The magnetic field excitation module outputs an excitation magnetic field and transmits parameters; the polarization modulation module modulates laser and magnetic field effects; the optical signal acquisition module captures and processes optical signals; the data processing module combines a physical model and a resolving algorithm to obtain original data; the network training module constructs generative adversarial network optimization data; and the result output module outputs three-dimensional magnetic field distribution data. The system improves the measurement accuracy, the data processing efficiency and the adaptive capacity through multi-module cooperation in combination with an algorithm and a model, effectively overcomes the problems of poor interference resistance, low efficiency and the like, and is suitable for a weak magnetic field three-dimensional measurement scene.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional measurement of weak magnetic fields, and in particular to a big data analysis system for three-dimensional measurement of weak magnetic fields based on the Faraday effect. Background Technology

[0002] In modern industrial testing, geological exploration, and biomedical fields, the demand for precise three-dimensional measurement of weak magnetic fields is increasingly urgent. Weak magnetic fields typically have extremely low amplitudes and are easily affected by environmental interference, making it difficult for traditional measurement methods to balance accuracy and efficiency. Measurement techniques based on the Faraday effect under an excitation magnetic field, utilizing the rotational effect of the magnetic field on polarized light to achieve magnetic field sensing, combined with three-dimensional solution algorithms and big data analysis capabilities, can overcome the limitations of traditional equipment and provide a new path for analyzing the distribution of weak magnetic fields in complex scenarios. This is particularly valuable in scenarios such as precision instrument fault diagnosis and underground resource exploration, providing crucial data support for subsequent analysis and decision-making.

[0003] Existing technologies have two significant drawbacks: First, the measurement system lacks anti-interference capabilities. Electromagnetic noise in the environment and the coupling effect between the excitation magnetic field and the weak magnetic field can cause polarization of the light signal, resulting in large deviations in the calculated magnetic field component data, making it difficult to meet the requirements of high-precision measurement. Second, the data processing efficiency and model generalization ability are limited. Traditional algorithms are slow to process massive amounts of three-dimensional magnetic field data and lack effective adaptive optimization mechanisms. When faced with weak magnetic fields of different intensities and distribution characteristics, the stability and consistency of the measurement system are poor, and it cannot flexibly adapt to diverse measurement scenarios. Summary of the Invention

[0004] To overcome the shortcomings and deficiencies of existing technologies, this invention provides a big data analysis system for three-dimensional measurement of weak magnetic fields based on the Faraday effect.

[0005] The technical solution adopted in this invention is a three-dimensional measurement and big data analysis system for weak magnetic fields based on the Faraday effect, comprising: The magnetic field excitation module outputs an excitation magnetic field of preset frequency and intensity through multiple coil arrays and transmits the excitation magnetic field parameters to the synchronization control module; the polarization modulation module receives the timing signal from the synchronization control module, modulates the polarization state of the incident laser beam, and causes the modulated polarized laser beam to enter the area to be measured and interact with the weak magnetic field and the excitation magnetic field in the area; the optical signal acquisition module captures the polarized optical signal after it has passed through the area to be measured, converts the optical signal into an electrical signal and performs pre-amplification processing, and then transmits the processed electrical signal to the data processing module; The data processing module receives the electrical signal output from the optical signal acquisition module, combines it with the excitation magnetic field parameters provided by the magnetic field excitation module, and uses the polarization state change calculation formula derived from the Faraday effect physical model to calculate the original three-dimensional component data of the weak magnetic field in the area to be measured. This original data is then transmitted to the network training module. The network training module receives the original three-dimensional component data of the weak magnetic field output from the data processing module, constructs a generative adversarial network (GAN) model, where the generator uses the original data as input to generate reconstructed magnetic field data, and the discriminator distinguishes between the original and reconstructed data to determine their authenticity. The network parameters are optimized through alternating training, and the optimized network model parameters are fed back to the data processing module. The result output module receives the three-dimensional measurement results of the weak magnetic field processed by the optimized network model from the data processing module, performs format conversion and coordinate calibration on the measurement results, and generates directly output three-dimensional magnetic field distribution data.

[0006] Furthermore, when the data processing module calculates the original data of the three-dimensional components of the weak magnetic field, it employs an improved three-dimensional magnetic field calculation algorithm. This algorithm, combined with the Faraday effect physical model, constructs the following formula: ,in, For a point within the area to be measured The total polarization rotation angle at that location; The Field constant of the magneto-optical medium The propagation path length of the laser beam within the area to be measured; To excite the magnetic field at point Time The changing magnetic field strength; For a weak magnetic field at point The magnetic field strength at that location; The length of the infinitesimal element along the laser propagation path; The polarization rotation angle deviation is due to noise introduced during the measurement process; simultaneously, during the generative adversarial network training process, the generator's loss function adopts an improved formula based on the three-dimensional measurement parameters of the weak magnetic field. ,in, This represents the loss value of the generator; and This is the weighting coefficient, with a value ranging from 0 to... The reconstructed magnetic field data output by the generator; This is the original three-dimensional component data of the weak magnetic field; It is an L2 norm; This represents the discrimination result of the discriminator on the reconstructed magnetic field data.

[0007] Furthermore, when constructing the generative adversarial network model, the network training module introduces constraints based on the Faraday effect physical model, and the generator's output layer adopts the following activation function formula: ,in, For a point of output of the generator The reconstructed magnetic field strength at the location; and These are the training parameters for the generator output layer; The duration of one cycle of the excitation magnetic field; The polarization rotation angle as a function of time is obtained from simulations based on the Faraday effect physical model. The change in ; tanh is the hyperbolic tangent activation function; in addition, the discriminator uses the following discriminant function formula when distinguishing the authenticity of the original data and the reconstructed data: ,in, The value range for the judgment result is 0 to 1; The magnetic field data (raw data or reconstructed data) is input to the discriminator. The average value of the weak magnetic field data in the training set; and Adjustment parameters for the discriminator It is a natural constant.

[0008] Furthermore, when modulating the polarization state of the incident laser beam, the polarization modulation module uses the following modulation depth calculation formula, combined with the dynamic range parameters of the three-dimensional measurement of the weak magnetic field: ,in, for The polarization modulation depth at any given time; The maximum modulation depth, with a value ranging from 0 to... For modulation frequency, This is the initial phase; This represents the maximum possible value of the weak magnetic field within the area to be measured. To modulate the threshold magnetic field strength; simultaneously, during the calculation process, the data processing module employs the following component separation formula to address the coupling problem of the three-dimensional magnetic field components: ,in, For weak magnetic fields in Corresponding to Components along the axis; For the first Polarization rotation angles measured in each polarization direction; For the first Each polarization direction and The angle between axes.

[0009] Furthermore, when capturing polarized light signals, the optical signal acquisition module combines the light intensity changes caused by a weak magnetic field and adopts the following light intensity response formula: ,in, for At all times The light intensity collected at that location; The initial intensity of the incident laser; This represents the initial polarization angle of the polarizer; For the optical signal acquisition module at point The photoelectric conversion efficiency at the point ranges from 0 to 1; the network training module uses the following learning rate adjustment formula during the alternating training process of the generative adversarial network: ,in, For the first The learning rate of each training round; For the first The learning rate of each training round; This is the learning rate adjustment factor; This represents the maximum value in the original weak magnetic field data.

[0010] Furthermore, the excitation magnetic field output by the magnetic field excitation module satisfies the following time-varying characteristic formula: ,in, To the amplitude of the excitation magnetic field; The frequency of the excitation magnetic field; This represents the initial phase of the excitation magnetic field; To excite the DC component in the magnetic field, the data processing module uses the following filtering formula on the original electrical signal when employing the three-dimensional magnetic field calculation algorithm: ,in, The filtered electrical signal spectrum; The original electrical signal spectrum; For frequency variables; This is the cutoff frequency for the low-pass filter; This is the filter order; is the standard deviation of the Gaussian filter.

[0011] Furthermore, the optical signal acquisition module includes: a polarization state detection unit, which incorporates multiple sets of orthogonal polarizers and photodetectors to receive polarized light signals after they have passed through the area to be measured. Light signals with different polarization directions pass through their corresponding polarizers, and the photodetectors convert these signals into corresponding current signals. The responsivity of each photodetector is individually calibrated according to the polarization direction of the incident light. A signal amplification unit receives the current signal output from the polarization state detection unit and uses a multi-stage differential amplifier circuit to convert and amplify the current signal. The gain of the amplifier circuit is dynamically adjusted according to the intensity of the input current signal to avoid signal saturation. Simultaneously, electromagnetic interference is reduced through shielding and grounding. The process involves several steps: A noise suppression unit receives the amplified electrical signal from the signal amplification unit and uses an adaptive noise cancellation algorithm to suppress environmental and circuit noise in the signal. This unit uses real-time background noise signals as a reference to generate a compensation signal that is the inverse of the noise signal. The compensation signal is then superimposed on the amplified electrical signal to cancel out the noise components. A data buffer unit receives the processed electrical signal from the noise suppression unit and uses a high-speed buffer to temporarily store the signal. During storage, the signal is categorized and marked according to timestamps and spatial coordinates. When the buffered data reaches a preset threshold, the data is automatically transmitted in batches to the data processing module, while the buffer space is cleared to prepare for receiving new electrical signals.

[0012] Furthermore, the data processing module includes: a signal analysis unit, which receives the electrical signal transmitted by the optical signal acquisition module, extracts the waveform features of the electrical signal, including peak value, frequency, and phase parameters, and converts the feature parameters of the electrical signal into corresponding polarization rotation angle changes based on the correspondence between polarization state changes and magnetic field strength in the Faraday effect physical model, introducing the propagation path length and magneto-optical medium parameters for correction during the conversion process; and a magnetic field component calculation unit, which receives the polarization rotation angle change output by the signal analysis unit, combines it with the spatial distribution data of the excitation magnetic field provided by the magnetic field excitation module, and uses a three-dimensional magnetic field calculation algorithm to decompose the polarization rotation angle change to obtain the component data of the weak magnetic field in the x, y, and z coordinate axes, combining the laser beam propagation direction and... The influence of the included angle of the magnetic field direction on the measurement results; the data association unit receives the weak magnetic field three-dimensional component data output by the magnetic field component calculation unit, and integrates this data with the corresponding spatial coordinate information, excitation magnetic field parameters, and measurement time information to form a dataset containing multi-dimensional information. During the association process, timestamps and coordinate markers are used to ensure the spatiotemporal consistency of data from different sources; the outlier identification unit receives the integrated dataset output by the data association unit, and identifies outlier data points that exceed the normal range by analyzing the distribution range and changing trend of different parameters in the dataset. The identification is based on a normal data model constructed based on historical measurement data. For the identified outlier data points, their location and corresponding measurement parameters are marked, and they are not included in the data subsequently transmitted to the network training module.

[0013] Furthermore, the network training module includes: a dataset partitioning unit, which receives the original data of the three-dimensional components of the weak magnetic field output by the data processing module, and randomly partitions the original data into a training set, a validation set, and a test set according to a preset ratio. During the partitioning process, it ensures that the spatial distribution of the data in different sets is consistent with the range of magnetic field strength, so as to avoid network training deviation due to uneven data distribution. At the same time, the data in different sets are independently numbered for easy tracking; and a network construction unit, which constructs the network structure of the generator and the discriminator according to the architectural requirements of the generative adversarial network. The generator adopts a multilayer perceptron structure, the number of input layer nodes is consistent with the parameter dimension of the three-dimensional components of the weak magnetic field, the hidden layer performs feature mapping on the input data through a nonlinear activation function, and the number of output layer nodes is the same as that of the input layer to generate reconstructed magnetic field data. The discriminator employs a convolutional neural network structure. It extracts spatial features from the input data through convolutional layers, reduces the feature dimensionality through pooling layers, and outputs the discrimination result through a fully connected layer. The parameter optimization unit receives training and validation set data from the dataset partitioning unit. It alternately updates the network parameters of the generator and discriminator using gradient descent. After each parameter update, it calculates the network loss value using validation set data. When the loss value stops decreasing for several consecutive rounds, the parameter update stops and the current network parameters are saved. The model evaluation unit receives the optimal network parameters and test set data from the parameter optimization unit. It inputs the test set data into the generator to generate reconstructed magnetic field data. It evaluates the network model's performance by calculating the error index between the reconstructed data and the original data. The evaluation result serves as the basis for deciding whether to use the model for subsequent data processing.

[0014] Beneficial Effects: This invention proposes a three-dimensional measurement and big data analysis system for weak magnetic fields based on the Faraday effect. This system achieves efficient and accurate three-dimensional measurement and analysis of weak magnetic fields through the coordinated operation of six modules and supporting technologies. The magnetic field excitation module outputs a stable and controllable excitation magnetic field, providing a benchmark for measurement; the polarization modulation module precisely modulates the laser to ensure its full interaction with the magnetic field; the optical signal acquisition module captures weak optical signals and suppresses noise through multi-stage processing, improving signal quality; the data processing module combines the Faraday effect physical model with a three-dimensional magnetic field calculation algorithm to accurately calculate the magnetic field components; the network training module optimizes the data using a generative adversarial network (GAN) to reduce bias; and the result output module ensures effective data output. The system reduces interference through the noise suppression unit of the optical signal acquisition module, and the correction mechanism of the data processing module and the optimization of the network training module reduce data bias and enhance anti-interference capabilities. The network training module efficiently processes massive amounts of data, and the GAN, after multiple rounds of training, possesses adaptive optimization capabilities, improving data processing efficiency and model generalization ability, adapting to different magnetic field scenarios, and ensuring system stability and consistency. Attached Figure Description

[0015] Figure 1 This is a diagram showing the system module composition of the present invention; Figure 2 This is a flowchart illustrating the system operation of the present invention. Detailed Implementation

[0016] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] like Figure 1 As shown, the three-dimensional measurement big data analysis system for weak magnetic fields based on the Faraday effect includes: The magnetic field excitation module is used to output an excitation magnetic field with a preset frequency and intensity through multiple coil arrays, and to transmit the excitation magnetic field parameters to the synchronization control module; Specifically, the magnetic field excitation module is the core component of the system for generating the reference magnetic field, and its technical parameters directly affect the accuracy of subsequent measurements. This module contains multiple coil arrays. The number of turns in the coils is determined according to the required intensity range of the excitation magnetic field, typically between 500 and 2000 turns. High-conductivity copper enameled wire is used as the coil material to reduce energy loss during current flow. The frequency of the excitation magnetic field can be adjusted within the range of 10Hz to 10kHz, and the intensity range is 1mT to 100mT, adapting to the needs of various weak magnetic field measurement scenarios. By outputting a stable and controllable excitation magnetic field, a known magnetic field environment is provided for the generation of the Faraday effect, allowing the measurement of weak magnetic fields to be calculated using the excitation magnetic field as a reference, thereby improving the accuracy and reliability of the measurement.

[0018] In the specific implementation process: First, based on the size of the area to be measured and the estimated intensity of the weak magnetic field, the arrangement of the coil array is determined, employing a three-dimensional orthogonal coil structure to ensure that an excitation magnetic field with uniformity better than 1% is generated within the area to be measured. Then, a specific current signal is input to the coil array via the control module. The magnitude of the current is calculated based on the required intensity of the excitation magnetic field. For example, when a 10mT excitation magnetic field is required, the input current is approximately 0.5A-2A, with the specific value determined by the number of turns and size of the coil. During the current input process, the coil temperature is monitored in real time. When the temperature exceeds 60℃, a heat dissipation device, such as a fan or water cooling system, is activated to prevent the coil from being damaged due to overheating. Simultaneously, a magnetic field sensor is used to collect real-time spatial distribution data of the excitation magnetic field within the area to be measured, with a sampling frequency of 1kHz. The collected data is transmitted to the synchronization control module to ensure that the parameters of the excitation magnetic field are synchronized with the operating timing of other modules in the system. In addition, in order to reduce the influence of external electromagnetic interference on the excitation magnetic field, the coil array is wrapped with a 2mm thick permalloy shielding layer. The magnetic permeability of the shielding layer is greater than 8000, which can effectively attenuate the interference of external magnetic fields and keep the fluctuation of the excitation magnetic field within 0.1%.

[0019] The polarization modulation module receives the timing signal from the synchronization control module and modulates the polarization state of the incident laser beam, so that the modulated polarized laser beam enters the area to be measured and interacts with the weak magnetic field and excitation magnetic field in the area. Specifically, the main function of the polarization modulation module is to modulate the polarization state of the incident laser beam. Its technical parameters are closely related to the characteristics of the laser and the modulation accuracy. The module uses a helium-neon laser as the laser source, with an output wavelength of 632.8 nm and a power stability better than 0.5% / h, ensuring stable intensity of the incident laser. The modulation element is an electro-optic modulator with a half-wave voltage of 300V-500V and a modulation bandwidth of 0-20kHz, enabling rapid modulation of the laser polarization state. The modulation accuracy of the polarization state can reach 0.1°, ensuring the detectability of subsequent changes in the optical signal. By modulating the laser polarization state, this module causes a identifiable change in polarization state after the laser interacts with a magnetic field, thereby converting the change in the magnetic field into a change in the optical signal, providing a detectable physical quantity for the measurement of weak magnetic fields.

[0020] In the specific implementation process: First, after the helium-neon laser is started, it undergoes a 30-minute preheating period to stabilize its output power within the range of 10mW-50mW. The laser beam is collimated using a collimating lens, resulting in a beam diameter of 2mm and a divergence angle of less than 0.5mrad, ensuring stable propagation along a preset path. Then, the synchronization control module sends a timing signal to the electro-optic modulator. The frequency of this signal is the same as the frequency of the excitation magnetic field and is adjustable within the range of 10Hz-10kHz. The electro-optic modulator adjusts the voltage applied across its terminals according to the timing signal, with a voltage range of 0-500V, thereby modulating the laser's polarization state. The modulated laser polarization direction changes periodically over time, with the period matching the period of the excitation magnetic field. The modulated polarized laser beam is then calibrated using a polarizer with a polarization direction accuracy of 0.5°, ensuring that the laser's polarization state meets the measurement requirements. Subsequently, the calibrated laser beam changes its propagation direction through a reflector and enters the measurement area at a 45° angle. During propagation, the laser beam spot size remains within 2mm ± 0.1mm, and the straightness error of the propagation path is less than 0.1mm / m, ensuring sufficient interaction between the laser and the magnetic field in the measurement area. During modulation, the polarization state of the laser is monitored in real time using a polarization state analyzer at a frequency of 1kHz. When the polarization state deviation exceeds 0.5°, the voltage parameters of the electro-optic modulator are adjusted promptly to maintain modulation accuracy.

[0021] The optical signal acquisition module captures the polarized light signal after it passes through the area to be measured, converts the optical signal into an electrical signal and performs pre-amplification processing, and then transmits the processed electrical signal to the data processing module. Specifically, the optical signal acquisition module is responsible for capturing and initially processing the polarized light signal after it has been acted upon by the magnetic field. Its technical parameters directly affect the signal-to-noise ratio and the accuracy of subsequent processing. The module uses a silicon-based PIN photodiode as its photodetector, with a response wavelength range of 400nm-1100nm, matching the wavelength of the incident laser. Its quantum efficiency is greater than 80%, enabling efficient conversion of the optical signal into an electrical signal. The preamplifier's gain is adjustable from 100 to 10000 times, with a bandwidth of 10Hz-1MHz, effectively amplifying weak electrical signals. By converting the weak optical signal into an easily processed electrical signal and improving its quality through amplification and initial processing, the module provides a reliable signal source for subsequent data processing, ensuring that changes in the optical signal caused by the weak magnetic field can be accurately detected.

[0022] In the specific implementation process: First, the polarized light signal, after passing through the area to be measured, is focused onto the photosensitive surface of the photodetector by a focusing lens. The focal length of the focusing lens is 50mm, and the aperture is F / 1.8, ensuring that the light signal can be efficiently received by the detector. The photodetector converts the light signal into a current signal. The magnitude of the current signal is proportional to the intensity of the light signal, ranging from 1nA to 10μA. The current signal enters the preamplifier for amplification. The amplifier gain is automatically adjusted according to the signal intensity. When the input current is less than 100nA, the gain is set to 10000 times; when the input current is between 100nA and 1μA, the gain is set to 1000 times; when the input current is greater than 1μA, the gain is set to 100 times to avoid signal saturation. The amplified electrical signal is a voltage signal, ranging from 10mV to 10V. It is filtered by a low-pass filter with a cutoff frequency of 100kHz, which can effectively filter out high-frequency noise. The processed electrical signal is converted into a digital signal via an analog-to-digital converter (ADC). The ADC has a 16-bit resolution and a sampling frequency of 1MHz, ensuring that signal details are fully preserved. The digital signal is stored in a 128GB high-speed cache with a read / write speed greater than 100MB / s, meeting the requirements of high-speed data acquisition. Simultaneously, during data acquisition, the signal-to-noise ratio (SNR) is monitored in real time. When the SNR falls below 30dB, the exposure time of the photodetector is automatically adjusted. The exposure time can be adjusted within the range of 1μs-100μs to improve signal quality. Once the data storage is full, the data is automatically transmitted to the data processing module at a transmission rate of 1Gbps, ensuring real-time data processing.

[0023] The data processing module receives the electrical signal output by the optical signal acquisition module, combines it with the excitation magnetic field parameters provided by the magnetic field excitation module, and uses the polarization state change calculation formula derived from the Faraday effect physical model to calculate the original data of the three-dimensional components of the weak magnetic field in the area to be measured, and transmits the original data to the network training module. Specifically, the data processing module is the key component for analyzing and processing the acquired signals, and its technical parameters are related to processing speed and accuracy. This module uses a multi-core digital signal processor with a main frequency of 2GHz and a computing power of 100GFLOPS, capable of rapidly processing large amounts of raw data. It has 32GB of memory, sufficient for data caching and temporary computation. Based on the Faraday effect physical model and a three-dimensional magnetic field calculation algorithm, it converts electrical signals into three-dimensional components of a weak magnetic field, realizing the conversion from optical signals to magnetic field data. This provides raw data for subsequent network training and result output, serving as a bridge connecting signal acquisition and data application.

[0024] In the specific implementation process: First, the digital electrical signal transmitted by the optical signal acquisition module is received and classified according to time series and spatial coordinates to form a structured data matrix. The data matrix has dimensions of 1024×1024×100, where the first two dimensions represent spatial coordinates and the third dimension represents the time series. Then, the parameters of the Faraday effect physical model stored inside the module are called, including the Feld constant of the magneto-optical medium and the propagation path length of the laser in the medium. These parameters are obtained through prior calibration, with the error of the Feld constant being less than 1% and the measurement error of the propagation path length being less than 0.1 mm. Combined with the excitation magnetic field parameters provided by the magnetic field excitation module, including the intensity, frequency, and phase of the excitation magnetic field, a three-dimensional magnetic field calculation algorithm is used to solve the electrical signal. During the calculation process, the rotation angle of the polarized light is first calculated based on the amplitude of the electrical signal, with a calculation accuracy of 0.01°. Then, based on the rotation angle, excitation magnetic field parameters, and physical model, the component data of the weak magnetic field in the x, y, and z directions are calculated. An iterative algorithm is used during the calculation, with 100 iterations to ensure convergence of the results. The original data of the three-dimensional components of the weak magnetic field is in Tesla, with a resolution of 1 pT. The sampling interval is 1 mm × 1 mm × 1 mm, covering the entire measurement area. During the calculation, the calculation error is monitored in real time. When the error exceeds 5%, the physical model parameters are recalculated to ensure data accuracy. After calculation, the original data is converted to binary format to reduce storage space. The converted data packet size is approximately 100 MB. The data is then transmitted to the network training module, and a verification mechanism is used during transmission to ensure data integrity.

[0025] The network training module receives the raw data of the three-dimensional components of the weak magnetic field from the data processing module, and constructs a generative adversarial network model. The generator uses the raw data as input to generate reconstructed magnetic field data, and the discriminator judges the authenticity of the raw data and the reconstructed data. The network parameters are optimized through alternating training, and the optimized network model parameters are fed back to the data processing module. Specifically, the network training module optimizes the raw magnetic field data using a generative adversarial network (GAN). Its technical parameters are related to the network's performance and training effectiveness. The module's hardware configuration includes a graphics processor with 16GB of video memory, capable of supporting the training of large-scale neural networks. Both the generator and discriminator of the GAN are built using a deep learning framework, with 5-10 layers and 128-1024 neurons per layer. Through training with the GAN, noise and errors in the raw data can be suppressed, improving data quality and enhancing the system's adaptability to different magnetic field distribution scenarios, resulting in more reliable and stable measurement results.

[0026] In the specific implementation process: First, the raw data of the three-dimensional components of the weak magnetic field output from the data processing module is received. The data is divided into three sets: 70% as the training set, 15% as the validation set, and 15% as the test set. Random sampling is used in the division process to ensure that the distribution characteristics of each dataset are consistent. Then, the parameters of the generative adversarial network (GAN) are initialized. The initial weights of the generator and discriminator are randomly generated using a normal distribution with a mean of 0 and a standard deviation of 0.01. The bias term is initialized to 0. The generator takes the raw data from the training set as input and generates reconstructed magnetic field data through multi-layer neural network calculations. The generator uses the ReLU activation function, and the output layer uses a linear activation function to ensure that the range of the output data is consistent with the original data. The discriminator receives the raw data and the reconstructed data, and outputs the discrimination result through neural network calculations. The discriminator uses the LeakyReLU activation function, and the output layer uses the Sigmoid function. The output result is between 0 and 1, representing the probability that the input data is the original data. During training, an alternating training approach was adopted: the discriminator was trained 5 times, followed by the generator 1 time, with a batch size of 64 for each training iteration. The cross-entropy loss function was used, and network parameters were updated using gradient descent. The initial learning rate was set to 0.0002, decreasing to 0.9 times every 100 training epochs. Every 10 training epochs, the network loss was calculated using validation set data. Training was stopped when the validation set loss no longer decreased for 20 consecutive epochs, and the current network parameters were saved. The number of training iterations was set between 1000 and 5000. After training, the network model was evaluated using test set data, and the mean squared error (MSE) between the reconstructed and original data was calculated. When the MSE was less than 1%, the optimized network model parameters were fed back to the data processing module for subsequent data processing.

[0027] The results output module receives the three-dimensional measurement results of the weak magnetic field after processing by the data processing module through an optimized network model, performs format conversion and coordinate calibration on the measurement results, and generates three-dimensional magnetic field distribution data that is directly output.

[0028] Specifically, the results output module is responsible for organizing and outputting the processed magnetic field data. Its technical parameters are related to the format and accuracy of the output data. This module supports multiple data output formats, including text, image, and 3D model formats. The coordinate accuracy of the output data is 0.1 mm, and the magnetic field strength accuracy is 1 pT. It transforms complex magnetic field data into an intuitive and easy-to-use format, facilitating subsequent analysis and application by users, enabling the measurement results to directly serve practical production and research work.

[0029] In the specific implementation process: First, the data processing module receives the three-dimensional measurement results of the weak magnetic field after processing by the optimized network model. The measurement results are in the form of a three-dimensional array, containing the x, y, and z coordinates of each point in the measurement area and the corresponding magnetic field strength value. Then, the measurement results are calibrated by converting the relative coordinates of the data to absolute coordinates according to the actual physical size of the measurement area, with a conversion error of less than 0.1 mm. After calibration, the magnetic field strength value of each point is analyzed to remove obvious outliers. Outliers are judged based on the magnetic field strength exceeding three standard deviations from the normal range. The normal range is determined by statistical analysis of the training set data. Next, the data is converted to a specific format according to the user's requirements. If converted to text format, CSV format is used, with each line containing the x, y, and z coordinates and the x, y, and z components of the magnetic field strength, separated by commas. If converted to image format, a three-dimensional isosurface map is generated, with an isosurface interval of 10 pT and a gradient fill color according to the magnitude of the magnetic field strength, from blue (weak magnetic field) to red (strong magnetic field). If converted to a three-dimensional model format, an STL format model file is generated, with a mesh accuracy of 1 mm. After format conversion, the output data undergoes a quality check to verify its completeness and accuracy, ensuring that the coordinates and magnetic field strength values ​​of all points are valid and free of missing or erroneous data. Finally, the processed output data is transmitted via an interface to an external device, such as a computer, printer, or storage device, at a transmission rate of 100Mbps. Simultaneously, a real-time preview image of the data is displayed on the module's screen, allowing users to intuitively understand the measurement results. Upon completion, a data report is automatically generated, containing information such as measurement time, measurement area, data format, and data volume. This report is stored in the module's local storage, which has a capacity of 1TB, capable of storing a large number of measurement reports and data files.

[0030] Preferably, when the data processing module calculates the original data of the three-dimensional components of the weak magnetic field, it employs an improved three-dimensional magnetic field calculation algorithm. This algorithm, combined with the Faraday effect physical model, constructs the following formula: ,in, For a point within the area to be measured The total polarization rotation angle at that location; The Field constant of the magneto-optical medium The propagation path length of the laser beam within the area to be measured; To excite the magnetic field at point Time The changing magnetic field strength; For a weak magnetic field at point The magnetic field strength at that location; The length of the infinitesimal element along the laser propagation path; The polarization rotation angle deviation is due to noise introduced during the measurement process; simultaneously, during the generative adversarial network training process, the generator's loss function adopts an improved formula based on the three-dimensional measurement parameters of the weak magnetic field. ,in, This represents the loss value of the generator; and This is the weighting coefficient, with a value ranging from 0 to... The reconstructed magnetic field data output by the generator; This is the original three-dimensional component data of the weak magnetic field; It is an L2 norm; This represents the discrimination result of the discriminator on the reconstructed magnetic field data.

[0031] Specifically, the data processing module employs an improved three-dimensional magnetic field calculation algorithm, combined with the Faraday effect physical model, to account for the combined effects of the excitation magnetic field, weak magnetic field, and noise on the total polarization rotation angle. The Feld constant of the magneto-optical medium is selected from 0.5 rad / (T•m) to 2 rad / (T•m), with the specific value determined based on the characteristics of the magneto-optical material used. The propagation path length of the laser beam within the measurement area is set according to the area size, between 0.1 meters and 1 meter, ensuring a sufficient reflection of the spatial distribution of the magnetic field. The frequency of the excitation magnetic field changing with time is consistent with the modulation frequency of the polarization modulation module, ranging from 1 mT to 100 mT within the range of 10 Hz to 10 kHz. The intensity of the weak magnetic field is set between 1 pT and 1 nT. The polarization rotation angle deviation caused by noise is controlled within 0.001 rad to ensure calculation accuracy. In the network training module, the generator's loss function incorporates weighting coefficients. The values ​​of α and β are adjusted based on the noise level of the data and the required reconstruction accuracy. Generally, α is between 0.6 and 0.8, and β is between 0.2 and 0.4. By balancing the deviation between the reconstructed data and the original data, as well as the discriminator's judgment result, the generator can more accurately generate reconstructed magnetic field data that closely approximates the original data. During implementation, the data processing module first determines parameters such as the Feld constant and propagation path length based on the magnetic field environment of the area to be measured. During the calculation process, it collects real-time data on the temporal variation of the excitation magnetic field and combines it with the polarization rotation angle obtained from the optical signal acquisition module to calculate the three-dimensional components of the weak magnetic field using an algorithm. The network training module sets the values ​​of α and β based on the characteristics of the original data and continuously adjusts the generator's parameters during training to gradually reduce the loss function value until a stable state is reached. This ensures that the generated reconstructed data accurately reflects the actual distribution of the weak magnetic field, improving the system's measurement accuracy and data reliability.

[0032] Preferably, when constructing the generative adversarial network model, the network training module introduces constraints based on the Faraday effect physical model, and the generator's output layer adopts the following activation function formula: ,in, For a point of output of the generator The reconstructed magnetic field strength at the location; and These are the training parameters for the generator output layer; The duration of one cycle of the excitation magnetic field; The polarization rotation angle as a function of time is obtained from simulations based on the Faraday effect physical model. The change in ; tanh is the hyperbolic tangent activation function; in addition, the discriminator uses the following discriminant function formula when distinguishing the authenticity of the original data and the reconstructed data: ,in, The value range for the judgment result is 0 to 1; The magnetic field data (raw data or reconstructed data) is input to the discriminator. The average value of the weak magnetic field data in the training set; and Adjustment parameters for the discriminator It is a natural constant.

[0033] Specifically, the construction of the generative adversarial network model in the network training module introduces constraints based on the Faraday effect physical model, optimizing the activation function of the generator's output layer and the discriminant function of the discriminator. The activation function parameters k and b of the generator's output layer are determined based on the measurement range of the weak magnetic field and the polarization rotation angle variation characteristics of the Faraday effect. The value of k ranges from 0.1T / rad to 1T / rad, and the value of b ranges from -0.1T to 0.1T. By adjusting these two parameters, the reconstructed magnetic field strength output by the generator is matched with the actual weak magnetic field strength range. The period of the excitation magnetic field is consistent with the synchronization period of the system, set between 0.0001 seconds and 0.1 seconds. The simulated polarization rotation angle variation over time is calibrated using previous physical experimental data to ensure it conforms to the actual Faraday effect. In the discriminator's discrimination function, γ ranges from 1 / T to 10 / T, and δ ranges from 0.1T to 1T. The average value of the weak magnetic field data in the training set is statistically derived from a large amount of historical measurement data, and is generally between 10pT and 100pT. By setting these parameters, the discriminator can effectively distinguish between the original magnetic field data and the reconstructed data output by the generator, thereby improving the accuracy of the discrimination results. During implementation, the network training module first determines the initial values ​​of k, b, γ, and δ based on the approximate range of the weak magnetic field to be measured. During generator training, the output is adjusted through an activation function based on the period of the excitation magnetic field and the simulated change in polarization rotation angle, making the reconstructed magnetic field data closer to the real data in terms of numerical range and trend. The discriminator uses a pre-set discriminant function to extract and analyze features from the input data, and the output discriminant results are used to guide the parameter optimization of the generator. After multiple rounds of training, the generator and discriminator reach a dynamic balance, and the generated reconstructed data can more accurately reflect the three-dimensional distribution characteristics of the weak magnetic field, enhancing the system's adaptability to complex magnetic field environments.

[0034] Preferably, when the polarization modulation module modulates the polarization state of the incident laser beam, it uses the following modulation depth calculation formula, combined with the dynamic range parameter of the three-dimensional measurement of the weak magnetic field: ,in, for The polarization modulation depth at any given time; The maximum modulation depth, with a value ranging from 0 to... For modulation frequency, This is the initial phase; This represents the maximum possible value of the weak magnetic field within the area to be measured. To modulate the threshold magnetic field strength; simultaneously, during the calculation process, the data processing module employs the following component separation formula to address the coupling problem of the three-dimensional magnetic field components: ,in, For weak magnetic fields in Corresponding to Components along the axis; For the first Polarization rotation angles measured in each polarization direction; For the first Each polarization direction and The angle between axes.

[0035] Specifically, in the modulation depth calculation formula of the polarization modulation module, the maximum modulation depth M0 is set to 0.8 to 0.95 to ensure sufficient modulation amplitude for detectable polarization state changes. The modulation frequency f is the same as the excitation magnetic field frequency, between 10 Hz and 10 kHz, and the initial phase... The synchronization requirement of the system is set to 0 to π / 2 radians. The maximum possible value of the weak magnetic field in the area to be measured is estimated based on the measurement scenario, and is between 1 pT and 1 nT. The modulation threshold magnetic field strength is then set. The setting is 10 to 100 times the maximum possible value of the weak magnetic field to ensure that the modulation depth can respond sensitively to changes in the weak magnetic field. In the component separation formula of the data processing module, the included angles of the three polarization directions... The polarization rotation angles are set to 0°, 60°, and 120° respectively, forming an orthogonal distribution to comprehensively capture changes in polarization state. By weighted calculation of the polarization rotation angles in the three directions, the components of the weak magnetic field in the x, y, and z directions are separated. During implementation, the polarization modulation module calculates the modulation depth variation curve based on the estimated maximum possible value of the weak magnetic field and the modulation threshold, and adjusts the polarization state modulation amplitude of the laser beam in real time, so that the modulated laser can more sensitively reflect changes in the weak magnetic field. The data processing module receives the polarization rotation angle data obtained by converting the optical signals in the three polarization directions, substitutes them into the component separation formula, and calculates the magnetic field components in each direction by combining the Feld constant and the propagation path length. In this way, the coupling problem of the three-dimensional magnetic field components is effectively solved, the independence and accuracy of the measurement of each component are improved, and the system can more accurately reproduce the spatial distribution of the weak magnetic field.

[0036] Preferably, when capturing polarized light signals, the optical signal acquisition module uses the following light intensity response formula, taking into account the light intensity changes caused by a weak magnetic field: ,in, for At all times The light intensity collected at that location; The initial intensity of the incident laser; This represents the initial polarization angle of the polarizer; For the optical signal acquisition module at point The photoelectric conversion efficiency at the point ranges from 0 to 1; the network training module uses the following learning rate adjustment formula during the alternating training process of the generative adversarial network: ,in, For the first The learning rate of each training round; For the first The learning rate of each training round; This is the learning rate adjustment factor; This represents the maximum value in the original weak magnetic field data.

[0037] Specifically, the light intensity response of the optical signal acquisition module and the learning rate adjustment of the network training module were optimized. In the light intensity response formula of the optical signal acquisition module, the initial light intensity I0 of the incident laser is set according to the power of the laser source, between 10mW and 50mW. The initial polarization angle θp of the polarizer is set to 0 degrees to ensure the stability of the initial polarization state. The photoelectric conversion efficiency η is determined according to the performance of the photodetector, between 0.7 and 0.9. This formula can accurately calculate the light intensity at different positions and times, reflecting the change in the polarization rotation angle. In the learning rate adjustment formula of the network training module, the initial learning rate ηk is set to 0.0001 to 0.001, the learning rate adjustment coefficient λ is between 0.5 and 2, and the maximum value in the original weak magnetic field data is determined according to the measurement scenario, between 1pT and 1nT. This formula can dynamically adjust the learning rate according to the deviation between the reconstructed data and the original data, making network training more efficient. During implementation, the optical signal acquisition module calculates the theoretical light intensity based on the initial light intensity of the incident laser and the preset initial angle of the polarizer, combined with the real-time measured polarization rotation angle, using the light intensity response formula. This theoretical light intensity is then compared with the actual acquired light intensity to calibrate the performance of the photodetector. After each round of training, the network training module calculates the deviation between the reconstructed data and the original data, and substitutes this deviation into the learning rate adjustment formula. When the deviation is large, the learning rate is appropriately increased to accelerate the convergence speed; when the deviation is small, the learning rate is decreased to improve accuracy. Through this dynamic adjustment mechanism, the generative adversarial network can reach its optimal state more quickly, and the generated reconstructed data is closer to the original data, improving the accuracy and stability of the system's measurement of weak magnetic fields.

[0038] Preferably, the excitation magnetic field output by the magnetic field excitation module satisfies the following time-varying characteristic formula: ,in, To the amplitude of the excitation magnetic field; The frequency of the excitation magnetic field; This represents the initial phase of the excitation magnetic field; To excite the DC component in the magnetic field, the data processing module uses the following filtering formula on the original electrical signal when employing the three-dimensional magnetic field calculation algorithm: ,in, The filtered electrical signal spectrum; The original electrical signal spectrum; For frequency variables; This is the cutoff frequency for the low-pass filter; This is the filter order; is the standard deviation of the Gaussian filter.

[0039] Specifically, in the excitation magnetic field output by the magnetic field excitation module, the amplitude B0(x,y,z) is spatially adjusted according to the size and location of the area to be measured, varying between 1mT and 100mT. The excitation magnetic field frequency fexc is consistent with the polarization modulation frequency, ranging from 10Hz to 10kHz. The initial phase φexc is set to 0 to π radians according to system synchronization requirements. The DC component Bdc(x,y,z) is controlled between 0.1mT and 1mT to ensure that the excitation magnetic field can effectively excite the Faraday effect while reducing interference to weak magnetic field measurements. In the filtering formula of the data processing module, the low-pass filter cutoff frequency fc is set to 2 to 5 times the excitation magnetic field frequency, between 20Hz and 50kHz. The filter order n is 4 to 8, and the standard deviation σ of the Gaussian filter is 0.1 to 0.3 times the excitation magnetic field frequency. Through this composite filtering method, high-frequency noise and non-excitation frequency components in the original electrical signal can be effectively filtered out, while retaining the effective signal related to the magnetic field. During implementation, the magnetic field excitation module adjusts the current output of each coil according to the spatial characteristics of the area to be measured, so that the amplitude and DC component of the excitation magnetic field present a specific distribution in space. At the same time, it precisely controls the frequency and phase to ensure synchronization with other modules of the system. After receiving the raw electrical signal, the data processing module first performs spectrum analysis to determine the main frequency components of the noise. Then, according to the filtering formula, it sets the corresponding cutoff frequency, order, and standard deviation parameters to filter the electrical signal. The signal-to-noise ratio of the processed electrical signal is improved by more than 30dB, providing a clearer and more reliable signal source for subsequent weak magnetic field calculations, and further improving the measurement accuracy and anti-interference capability of the system.

[0040] Preferably, the optical signal acquisition module includes: a polarization state detection unit, which integrates multiple sets of orthogonal polarizers and photodetectors to receive polarized light signals after passing through the area to be measured. Light signals with different polarization directions pass through their corresponding polarizers, and the photodetectors convert these signals into corresponding current signals. The responsivity of each photodetector is individually calibrated according to the polarization direction of the incident light. A signal amplification unit receives the current signal output from the polarization state detection unit and uses a multi-stage differential amplifier circuit to convert and amplify the current signal. The gain of the amplifier circuit is dynamically adjusted according to the intensity of the input current signal to avoid signal saturation. Simultaneously, electromagnetic interference is reduced through shielding and grounding. The process involves several steps: A noise suppression unit receives the amplified electrical signal from the signal amplification unit and uses an adaptive noise cancellation algorithm to suppress environmental and circuit noise in the signal. This unit uses real-time background noise signals as a reference to generate a compensation signal that is the inverse of the noise signal. The compensation signal is then superimposed on the amplified electrical signal to cancel out the noise components. A data buffer unit receives the processed electrical signal from the noise suppression unit and uses a high-speed buffer to temporarily store the signal. During storage, the signal is categorized and marked according to timestamps and spatial coordinates. When the buffered data reaches a preset threshold, the data is automatically transmitted in batches to the data processing module, while the buffer space is cleared to prepare for receiving new electrical signals.

[0041] Specifically, the four units of the optical signal acquisition module work together to ensure efficient capture and processing of optical signals. In the polarization state detection unit, the polarization direction angle of multiple sets of orthogonal polarizers is precisely controlled at 90 degrees to ensure that polarized light in different directions can be detected separately. The photodetector uses a silicon-based photodiode with fast response speed and low dark current, whose response wavelength covers the 632.8nm laser wavelength. The responsivity of each detector is individually calibrated, with the error controlled within ±2%, ensuring the consistency of optical signal conversion in different directions. The signal amplification unit adopts a multi-stage differential amplifier circuit. The first stage amplification factor is set to 100 times, the second stage is adjustable from 10 to 100 times, and the total gain range is 1000-10000 times. It can amplify nA-level current signals to mV-level voltage signals. The bandwidth of the amplifier circuit is 10Hz-1MHz to meet the amplification requirements of signals of different frequencies. At the same time, it is grounded through a metal shield shell with a grounding resistance of less than 1Ω, effectively reducing electromagnetic interference and improving the signal-to-noise ratio of the amplified signal by more than 20dB. The noise suppression unit's adaptive noise cancellation algorithm has a sampling frequency of 1kHz. The reference noise signal is collected from the background environment far from the measurement area, and the generation delay of the compensation signal is controlled within 10μs to ensure synchronous cancellation with the noise in the original signal. After processing by this unit, the 50Hz power frequency interference and high-frequency noise in the signal can be reduced to less than 1 / 10 of the original. The data buffer unit has a high-speed cache capacity of 128MB and a read / write speed of 100MB / s, capable of continuously storing 10 seconds of sampled data. When the cache data reaches 80% capacity, data transmission to the data processing module is automatically triggered. The transmission interface uses USB 3.0 with a transmission rate of 5Gbps, ensuring no data loss. The cache clearing time is less than 10ms, allowing for rapid preparation to receive new data. During implementation, the polarization state detection unit first performs beam splitting detection on the polarized light subjected to the magnetic field, converts it into a current signal, and sends it to the signal amplification unit. After amplification and noise suppression, it is temporarily stored and transmitted by the data buffer unit. The total delay of the entire process is controlled within 50μs, providing a high-quality original signal for subsequent data processing.

[0042] Preferably, the data processing module includes: a signal analysis unit, which receives the electrical signal transmitted by the optical signal acquisition module, extracts the waveform features of the electrical signal, including peak value, frequency, and phase parameters, and converts the feature parameters of the electrical signal into corresponding polarization rotation angle changes based on the correspondence between polarization state changes and magnetic field strength in the Faraday effect physical model, introducing the propagation path length and magneto-optical medium parameters for correction during the conversion process; and a magnetic field component calculation unit, which receives the polarization rotation angle change output by the signal analysis unit, combines it with the spatial distribution data of the excitation magnetic field provided by the magnetic field excitation module, and uses a three-dimensional magnetic field calculation algorithm to decompose the polarization rotation angle change to obtain the component data of the weak magnetic field in the x, y, and z coordinate axes, combining the laser beam propagation direction and... The influence of the included angle of the magnetic field direction on the measurement results; the data association unit receives the weak magnetic field three-dimensional component data output by the magnetic field component calculation unit, and integrates this data with the corresponding spatial coordinate information, excitation magnetic field parameters, and measurement time information to form a dataset containing multi-dimensional information. During the association process, timestamps and coordinate markers are used to ensure the spatiotemporal consistency of data from different sources; the outlier identification unit receives the integrated dataset output by the data association unit, and identifies outlier data points that exceed the normal range by analyzing the distribution range and changing trend of different parameters in the dataset. The identification is based on a normal data model constructed based on historical measurement data. For the identified outlier data points, their location and corresponding measurement parameters are marked, and they are not included in the data subsequently transmitted to the network training module.

[0043] Specifically, the data processing module comprises four units that collaborate in an orderly manner to convert and optimize electrical signals into magnetic field data. The signal analysis unit extracts characteristic parameters such as peak value, frequency, and phase of the electrical signal with extraction accuracies of ±1mV, ±0.1Hz, and ±0.1 degrees, respectively. When converting the polarization rotation angle, the introduced propagation path length error is less than 0.1mm. After temperature compensation, the error of the magneto-optical medium parameters is controlled within ±1%, ensuring that the converted polarization rotation angle accuracy reaches 0.01 degrees. In the magnetic field component calculation unit, the resolution of the excitation magnetic field spatial distribution data is 1mm×1mm×1mm. When calculating the three-dimensional magnetic field components, the angle between the laser beam propagation direction and the magnetic field direction is combined, with an angle measurement accuracy of ±0.5 degrees. After decomposition using the three-dimensional magnetic field calculation algorithm, the calculation errors of the x, y, and z components are all less than 5%, effectively distinguishing the magnetic field contribution from each direction. The data association unit associates magnetic field component data with spatial coordinates, excitation parameters, measurement time, and other information. The positioning error of the spatial coordinates is less than 0.5 mm, and the synchronization accuracy of the timestamp is ±1 μs. The associated dataset is stored in a structured format, with each data point containing 20 fields for easy subsequent retrieval and analysis. The accuracy of the data association reaches over 99.9%. The outlier identification unit constructs a normal data model based on historical data, containing 10,000 samples. The threshold for outlier identification is set at mean ± 3 times the standard deviation. Identified outlier data points are marked and stored in a separate outlier database, with the proportion of outlier data controlled below 0.5% to ensure the quality of data entering the network training module. During implementation, the signal analysis unit first processes the input electrical signal to obtain the polarization rotation angle, which is then passed to the magnetic field component calculation unit. The calculated three-dimensional components are integrated by the data association unit and then filtered by the outlier identification unit. The final output effective data rate remains above 99%, providing reliable raw data for network training.

[0044] Preferably, the network training module includes: a dataset partitioning unit, which receives the raw data of the three-dimensional components of the weak magnetic field output by the data processing module, and randomly partitions the raw data into a training set, a validation set, and a test set according to a preset ratio. During the partitioning process, it ensures that the spatial distribution of data in different sets is consistent with the range of magnetic field strength, avoiding network training bias due to uneven data distribution. Simultaneously, the data in different sets are independently numbered for easy tracking. A network construction unit, according to the architectural requirements of a generative adversarial network, constructs the network structure of a generator and a discriminator. The generator adopts a multilayer perceptron structure, with the number of input layer nodes consistent with the parameter dimensions of the three-dimensional components of the weak magnetic field. The hidden layer performs feature mapping on the input data through a nonlinear activation function, and the number of output layer nodes is the same as the input layer to generate reconstructed magnetic field data. The discriminator... The discriminator employs a convolutional neural network structure. It extracts spatial features from the input data through convolutional layers, reduces the feature dimensionality through pooling layers, and outputs the discrimination result through a fully connected layer. The parameter optimization unit receives training and validation set data from the dataset partitioning unit. It uses gradient descent to alternately update the network parameters of the generator and discriminator. After each parameter update, it calculates the network loss value using validation set data. When the loss value stops decreasing for several consecutive rounds, it stops updating the parameters and saves the current network parameters. The model evaluation unit receives the optimal network parameters and test set data from the parameter optimization unit. It inputs the test set data into the generator to generate reconstructed magnetic field data. It evaluates the network model's performance by calculating the error index between the reconstructed data and the original data. The evaluation result serves as the basis for deciding whether to use the model for subsequent data processing.

[0045] Specifically, the four units of the network training module work together to build and optimize the generative adversarial network. The dataset partitioning unit divides the original data into training, validation, and test sets in a 7:1.5:1.5 ratio. The partitioning process uses random number seeds to ensure consistency in each partition. The spatial distribution deviation of the data in each set is less than 5%, and the magnetic field strength range covers more than 95% of the original data range to avoid training bias caused by uneven data distribution. The total sample size of the dataset is no less than 10,000 sets to ensure sufficient network training. In the network building units, the generator's input layer has 3 nodes (corresponding to the three-dimensional magnetic field components), and the hidden layer contains 3 layers with 256, 512, and 256 neurons per layer, respectively, using the ReLU activation function. The output layer uses a linear activation function to ensure that the output data range is consistent with the input. The discriminator's convolutional layer contains 3 layers with kernel sizes of 3×3, 5×5, and 3×3, with a stride of 1. The pooling layer uses 2×2 max pooling. The fully connected layer contains 2 layers with 128 and 1 neurons, respectively. The output layer uses the Sigmoid activation function. The network parameters are initialized using the Xavier initialization method to ensure stable training. The parameter optimization unit uses the Adam optimizer with an initial learning rate of 0.0002, β1=0.5, β2=0.999, and a batch size of 64. Each iteration trains the discriminator 5 times and the generator once. The validation set loss is calculated every 10 epochs. Training stops when the loss value changes by less than 0.001 for 20 consecutive epochs. The total number of training epochs is set between 1000 and 5000, and the saved optimal parameters occupy less than 100MB of storage space. The model evaluation unit uses mean squared error (MSE) and structural similarity index (SSIM) as evaluation metrics. MSE is controlled below 1e-6, and SSIM is greater than 0.95. The evaluation process uses 500 sets of test data for validation. The visual comparison error between the generated reconstructed data and the original data is less than 10%. Models that pass the evaluation are marked as usable for subsequent data processing. During implementation, the dataset partitioning unit first processes the input data, and the partitioned dataset is then fed into the network building unit to build the network structure. The parameter optimization unit performs multiple rounds of training, and finally the model evaluation unit verifies the model. The training time for the entire process depends on the amount of data and is set between 24 and 48 hours to ensure that the generated network model can effectively improve the quality of the magnetic field data.

[0046] The Faraday effect physical model in this invention is based on the principle of the effect of a magnetic field on polarized light. Specifically, it describes the phenomenon that the polarization direction of polarized light rotates when it propagates in a magnetic field. The rotation angle is related to the magnetic field strength, the path length of light propagation in the magnetic field, and the Feld constant of the magneto-optical medium. The realization of this model relies on a precise characterization of the interaction process between polarized light and the magnetic field. By introducing the superposition effect of the excitation magnetic field and the weak magnetic field, and combining the influence of noise on the polarization rotation angle during the measurement process, the model can more comprehensively reflect the actual measurement scenario. In the system, the model's role is to establish a quantitative relationship between the polarization rotation angle information acquired by the optical signal acquisition module and the magnetic field strength, providing a theoretical basis for calculating magnetic field data from the optical signal. It provides a reliable physical basis for the measurement of weak magnetic fields, converting magnetic field quantities that are difficult to measure directly into polarization rotation angles that can be detected by optical means through clear physical laws. This overcomes the limitations of traditional magnetic field measurement methods in detecting weak magnetic fields, making accurate measurement of extremely weak magnetic fields possible and laying a solid theoretical foundation for subsequent data analysis and applications.

[0047] The three-dimensional magnetic field calculation algorithm is the core algorithm in this invention for converting polarized light signals into three-dimensional component data of weak magnetic fields. Based on the Faraday effect physical model, it combines the parameters of the excitation magnetic field and the polarization rotation angle information acquired from the light signal, and decomposes the weak magnetic field components in the x, y, and z directions through complex mathematical operations. The algorithm's implementation includes multiple steps: first, feature extraction is performed on the electrical signal transmitted by the light signal acquisition module to obtain the relevant parameters of the polarization rotation angle; then, combining the spatial distribution data of the excitation magnetic field and information such as the angle between the laser beam propagation path and the magnetic field direction, iterative calculation methods are used to decompose the polarization rotation angle, ultimately obtaining the magnetic field components in each direction. In the system, this algorithm transforms the original light signal data into three-dimensional magnetic field data with practical physical meaning, achieving the crucial signal-to-data conversion. It solves the coupling problem of the three-dimensional magnetic field components, accurately distinguishes the magnetic field contributions in different directions, improves the spatial resolution and accuracy of magnetic field measurement, and enables the system to comprehensively and accurately reconstruct the spatial distribution characteristics of the weak magnetic field within the measurement area, providing high-quality raw data for subsequent network training and result analysis.

[0048] In this invention, a Generative Adversarial Network (GAN) is a deep learning model composed of a generator and a discriminator. The generator generates reconstructed magnetic field data based on the original data of weak magnetic field three-dimensional components, while the discriminator distinguishes the authenticity of the original and reconstructed data. The implementation process includes dataset partitioning, network structure construction, parameter optimization, and model evaluation. First, the original data is divided into training, validation, and test sets. Then, the network structures for the generator and discriminator are constructed. The generator uses a multilayer perceptron structure for feature mapping, and the discriminator uses a convolutional neural network structure to extract spatial features. Next, the network parameters are optimized through alternating training, making the reconstructed data generated by the generator increasingly closer to the original data, and continuously improving the discriminator's discrimination ability. Finally, the performance of the network model is evaluated using test set data. In the system, the role of this network is to optimize the original magnetic field data, suppress noise and errors in the data, and improve data quality. Deep learning enhances the system's adaptability to complex magnetic field environments, improves the reliability and stability of weak magnetic field measurement data, and enables the system to maintain high measurement accuracy when facing weak magnetic fields of different intensities and distribution characteristics, providing better data support for the analysis and utilization of weak magnetic fields in practical applications.

[0049] like Figure 2The aforementioned three-dimensional measurement big data analysis system for weak magnetic fields based on the Faraday effect comprises the following steps: Step S1: Activating the magnetic field excitation module, controlling multiple coil arrays to output excitation magnetic fields according to preset frequencies, intensities, and phases, and acquiring real-time spatial distribution data of the excitation magnetic field within the measurement area, sending this data to the synchronization control module to achieve time synchronization of all modules in the system; Step S2: Sending a modulation signal to the polarization modulation module through the synchronization control module, causing the polarization modulation module to periodically modulate the polarization state of the incident laser beam. The modulated polarized laser beam enters the measurement area along a preset path, interacting with the excitation magnetic field and weak magnetic field within the area during propagation, generating polarization state changes conforming to the Faraday effect; Step S3: Using the optical signal acquisition module to capture polarized light signals at the output end of the measurement area, converting the optical signals into electrical signals, performing pre-amplification and noise suppression processing, storing the processed electrical signals according to time series and spatial coordinates, and transmitting them to the data center. Processing Module; Step S4: The data processing module receives electrical signals and excitation magnetic field data, calculates the rotation angle of the polarized light signal based on the Faraday effect physical model, and calculates the original data of the three-dimensional components of the weak magnetic field using a three-dimensional magnetic field solution algorithm. During the solution process, laser beam propagation path and magneto-optical medium characteristic parameters are introduced for correction. Step S5: The original data of the three-dimensional components of the weak magnetic field is input into the generative adversarial network constructed by the network training module. The generator generates reconstructed magnetic field data based on the input data, and the discriminator compares and distinguishes the original data and the reconstructed data. Through multiple rounds of alternating training, the network parameters are optimized so that the deviation between the reconstructed data output by the generator and the original data is within a preset range. Step S6: After the network training is completed, the data processing module calls the optimized network model to process the original data of the three-dimensional components of the weak magnetic field, obtaining a denoised and enhanced three-dimensional magnetic field measurement result. The result output module performs coordinate transformation and format standardization on the measurement result, forming three-dimensional magnetic field distribution data that can be directly used for subsequent analysis.

[0050] This big data analysis system for three-dimensional measurement of weak magnetic fields based on the Faraday effect offers significant advantages in measurement accuracy. The magnetic field excitation module outputs a stable and controllable excitation magnetic field, providing a reliable benchmark for measurement; the polarization modulation module precisely modulates the laser to ensure its full interaction with the magnetic field; the optical signal acquisition module captures weak optical signals and suppresses noise through multi-stage processing, greatly improving signal quality; and the data processing module combines the Faraday effect physical model with a three-dimensional magnetic field calculation algorithm to accurately calculate the magnetic field components. The collaborative efforts of these multiple modules ensure the accuracy of the measurement data from the outset, laying a solid foundation for subsequent analysis.

[0051] It demonstrates outstanding performance in data processing. The network training module optimizes the data using generative adversarial networks, effectively reducing data bias. Simultaneously, this module can efficiently process massive amounts of data, continuously optimizing network parameters through multiple rounds of alternating training, significantly improving data processing speed. This advantage effectively solves the problem of low data processing efficiency in traditional technologies, enabling the system to quickly respond to the analytical needs of large amounts of measurement data.

[0052] The system also possesses strong adaptability. Addressing the issue of weak anti-interference capabilities, the noise suppression unit in the optical signal acquisition module reduces interference, while the correction mechanism in the data processing module and the optimization in the network training module further reduce data deviation, significantly enhancing the system's anti-interference ability. The generative adversarial network, after multiple rounds of training, possesses adaptive optimization capabilities, enabling the system to adapt to weak magnetic field scenarios with varying intensities and distribution characteristics, ensuring measurement stability and consistency.

[0053] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0054] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A three-dimensional measurement and big data analysis system for weak magnetic fields based on the Faraday effect, characterized in that, include: The magnetic field excitation module is used to output an excitation magnetic field with a preset frequency and intensity through multiple coil arrays, and to transmit the excitation magnetic field parameters to the synchronization control module; The polarization modulation module receives the timing signal from the synchronization control module and modulates the polarization state of the incident laser beam, so that the modulated polarized laser beam enters the area to be measured and interacts with the weak magnetic field and excitation magnetic field in the area. The optical signal acquisition module captures the polarized light signal after it passes through the area to be measured, converts the optical signal into an electrical signal and performs pre-amplification processing, and then transmits the processed electrical signal to the data processing module. The data processing module receives the electrical signal output by the optical signal acquisition module, combines it with the excitation magnetic field parameters provided by the magnetic field excitation module, and uses the polarization state change calculation formula derived from the Faraday effect physical model to calculate the original data of the three-dimensional components of the weak magnetic field in the area to be measured, and transmits the original data to the network training module. The network training module receives the raw data of the three-dimensional components of the weak magnetic field from the data processing module, and constructs a generative adversarial network model. The generator uses the raw data as input to generate reconstructed magnetic field data, and the discriminator judges the authenticity of the raw data and the reconstructed data. The network parameters are optimized through alternating training, and the optimized network model parameters are fed back to the data processing module. The results output module receives the three-dimensional measurement results of the weak magnetic field after processing by the data processing module through an optimized network model, performs format conversion and coordinate calibration on the measurement results, and generates three-dimensional magnetic field distribution data that is directly output.

2. The three-dimensional measurement big data analysis system for weak magnetic fields based on the Faraday effect according to claim 1, characterized in that, When processing the raw data of the three-dimensional components of the weak magnetic field, the data processing module employs an improved three-dimensional magnetic field calculation algorithm. This algorithm, combined with the Faraday effect physical model, constructs the following formula: ,in, For a point within the area to be measured The total polarization rotation angle at that location; The Field constant of the magneto-optical medium The propagation path length of the laser beam within the area to be measured; To excite the magnetic field at point Time The changing magnetic field strength; For a weak magnetic field at point The magnetic field strength at that location; The length of the infinitesimal element along the laser propagation path; The polarization rotation angle deviation is due to noise introduced during the measurement process; simultaneously, during the generative adversarial network training process, the generator's loss function adopts an improved formula based on the three-dimensional measurement parameters of the weak magnetic field. ,in, This represents the loss value of the generator; and This is the weighting coefficient, with a value ranging from 0 to... The reconstructed magnetic field data output by the generator; This is the original three-dimensional component data of the weak magnetic field; It is an L2 norm; This represents the discrimination result of the discriminator on the reconstructed magnetic field data.

3. The three-dimensional measurement big data analysis system for weak magnetic fields based on the Faraday effect according to claim 2, characterized in that, When constructing the generative adversarial network model, the network training module introduces constraints based on the Faraday effect physics model. The generator's output layer uses the following activation function formula: ,in, For a point of output of the generator The reconstructed magnetic field strength at the location; and These are the training parameters for the generator output layer; The duration of one cycle of the excitation magnetic field; The polarization rotation angle as a function of time is obtained from simulations based on the Faraday effect physical model. The change in ; tanh is the hyperbolic tangent activation function; in addition, the discriminator uses the following discriminant function formula when distinguishing the authenticity of the original data and the reconstructed data: ,in, The value range for the judgment result is 0 to 1; The magnetic field data (raw data or reconstructed data) is input to the discriminator. The average value of the weak magnetic field data in the training set; and Adjustment parameters for the discriminator It is a natural constant.

4. The three-dimensional measurement big data analysis system for weak magnetic fields based on the Faraday effect according to claim 2, characterized in that, When modulating the polarization state of the incident laser beam, the polarization modulation module uses the following formula to calculate the modulation depth, taking into account the dynamic range parameters of the three-dimensional measurement of the weak magnetic field: ,in, for The polarization modulation depth at any given time; The maximum modulation depth, with a value ranging from 0 to... For modulation frequency, This is the initial phase; This represents the maximum possible value of the weak magnetic field within the area to be measured. To modulate the threshold magnetic field strength; simultaneously, during the calculation process, the data processing module employs the following component separation formula to address the coupling problem of the three-dimensional magnetic field components: ,in, For weak magnetic fields in Corresponding to Components along the axis; For the first Polarization rotation angles measured in each polarization direction; For the first Each polarization direction and The angle between axes.

5. The three-dimensional measurement big data analysis system for weak magnetic fields based on the Faraday effect according to claim 2, characterized in that, When capturing polarized light signals, the optical signal acquisition module combines the light intensity changes caused by a weak magnetic field and uses the following light intensity response formula: ,in, for At all times The light intensity collected at that location; The initial intensity of the incident laser; This represents the initial polarization angle of the polarizer; For the optical signal acquisition module at point The photoelectric conversion efficiency at the point ranges from 0 to 1; the network training module uses the following learning rate adjustment formula during the alternating training process of the generative adversarial network: ,in, For the first The learning rate of each training round; For the first The learning rate of each training round; This is the learning rate adjustment factor; This represents the maximum value in the original weak magnetic field data.

6. The three-dimensional measurement big data analysis system for weak magnetic fields based on the Faraday effect according to claim 1, characterized in that, The excitation magnetic field output by the magnetic field excitation module satisfies the following time-varying characteristic formula: ,in, To the amplitude of the excitation magnetic field; The frequency of the excitation magnetic field; This represents the initial phase of the excitation magnetic field; To excite the DC component in the magnetic field, the data processing module uses the following filtering formula on the original electrical signal when employing the three-dimensional magnetic field calculation algorithm: ,in, The filtered electrical signal spectrum; The original electrical signal spectrum; For frequency variables; This is the cutoff frequency for the low-pass filter; This is the filter order; is the standard deviation of the Gaussian filter.

7. The three-dimensional measurement big data analysis system for weak magnetic fields based on the Faraday effect according to claim 1, characterized in that, The optical signal acquisition module includes: a polarization state detection unit, which incorporates multiple sets of orthogonal polarizers and photodetectors to receive polarized light signals after they have passed through the area to be measured. Light signals with different polarization directions pass through their corresponding polarizers, and the photodetectors convert these signals into corresponding current signals. The responsivity of each photodetector is individually calibrated according to the polarization direction of the incident light. A signal amplification unit receives the current signal output from the polarization state detection unit and uses a multi-stage differential amplifier circuit to convert and amplify the current signal. The gain of the amplifier circuit is dynamically adjusted according to the intensity of the input current signal to avoid signal saturation. Simultaneously, shielding and grounding reduce electromagnetic interference during the amplification process. The noise suppression unit receives the amplified electrical signal output from the signal amplification unit and uses an adaptive noise cancellation algorithm to suppress environmental and circuit noise in the electrical signal. This unit uses real-time background noise signals as a reference to generate a compensation signal that is the opposite of the noise signal. The compensation signal is then superimposed on the amplified electrical signal to cancel out the noise components. The data buffer unit receives the processed electrical signal output from the noise suppression unit and uses a high-speed buffer to temporarily store the electrical signal. During storage, the electrical signal is classified and marked according to timestamps and spatial coordinates. When the buffered data reaches a preset threshold, the data is automatically transmitted in batches to the data processing module, and the buffer space is cleared to prepare for receiving new electrical signals.

8. The three-dimensional measurement big data analysis system for weak magnetic fields based on the Faraday effect according to claim 1, characterized in that, The data processing module includes: a signal analysis unit, which receives the electrical signal transmitted by the optical signal acquisition module, extracts the waveform features of the electrical signal, including peak value, frequency, and phase parameters, and converts the characteristic parameters of the electrical signal into corresponding polarization rotation angle changes based on the correspondence between polarization state changes and magnetic field strength in the Faraday effect physical model. During the conversion process, the propagation path length and magneto-optical medium parameters are introduced for correction. A magnetic field component calculation unit receives the polarization rotation angle change output by the signal analysis unit, combines it with the spatial distribution data of the excitation magnetic field provided by the magnetic field excitation module, and uses a three-dimensional magnetic field calculation algorithm to decompose the polarization rotation angle change, obtaining the component data of the weak magnetic field in the x, y, and z coordinate axes. During the calculation process, the laser beam propagation direction and magnetic field are considered. The influence of the included angle on the measurement results; the data association unit receives the weak magnetic field three-dimensional component data output by the magnetic field component calculation unit, and integrates this data with the corresponding spatial coordinate information, excitation magnetic field parameters, and measurement time information to form a dataset containing multi-dimensional information. During the association process, timestamps and coordinate markers are used to ensure the spatiotemporal consistency of data from different sources; the outlier identification unit receives the integrated dataset output by the data association unit, and identifies outlier data points that exceed the normal range by analyzing the distribution range and changing trend of different parameters in the dataset. The identification is based on a normal data model constructed based on historical measurement data. For the identified outlier data points, their positions and corresponding measurement parameters are marked, and they are not included in the data subsequently transmitted to the network training module.

9. The three-dimensional measurement big data analysis system for weak magnetic fields based on the Faraday effect according to claim 1, characterized in that, The network training module includes: a dataset partitioning unit, which receives the raw data of the three-dimensional components of the weak magnetic field output by the data processing module, and randomly partitions the raw data into training set, validation set, and test set according to a preset ratio. During the partitioning process, it ensures that the spatial distribution of the data in different sets is consistent with the range of magnetic field strength, avoiding network training bias caused by uneven data distribution. At the same time, the data in different sets are independently numbered for easy tracking; a network construction unit, which constructs the network structure of generator and discriminator according to the architectural requirements of generative adversarial network. The generator adopts a multilayer perceptron structure, with the number of input layer nodes consistent with the parameter dimension of the three-dimensional components of the weak magnetic field. The hidden layer performs feature mapping on the input data through a nonlinear activation function, and the number of output layer nodes is the same as that of the input layer to generate reconstructed magnetic field data; the discriminator adopts a convolutional neural network structure, which extracts the spatial features of the input data through convolutional layers, reduces the feature dimension through pooling layers, and outputs the discrimination result through a fully connected layer.

10. The three-dimensional measurement big data analysis system for weak magnetic fields based on the Faraday effect according to claim 1, characterized in that, The network training module further includes: a parameter optimization unit, which receives training and validation set data output by the dataset partitioning unit, alternately updates the network parameters of the generator and discriminator using the gradient descent algorithm, calculates the network loss value using validation set data after each parameter update, and stops parameter updates and saves the current network parameters when the loss value no longer decreases for several consecutive rounds; and a model evaluation unit, which receives the optimal network parameters and test set data output by the parameter optimization unit, inputs the test set data into the generator to generate reconstructed magnetic field data, evaluates the performance of the network model by calculating the error index between the reconstructed data and the original data, and uses the evaluation result as the basis for whether the model is used for subsequent data processing.

Images