A magnetic field imaging method embedded with a deep denoising network

Abstract

This invention discloses a magnetic field imaging method with an embedded deep denoising network. The implementation steps include: (1) building an experimental platform and deploying a main imaging sensor array and an auxiliary sensing sensor array in a preset two-dimensional measurement area to construct a magnetic field imaging system; (2) establishing a two-dimensional relative position mapping model between the main imaging sensor and the measured magnetic field source based on the two-dimensional grid parameters and the layout of the main imaging sensor; (3) simultaneously acquiring magnetic noise signals from the main imaging sensor and the auxiliary sensing sensor under the condition of only environmental noise interference to construct a deep denoising learning dataset; (4) constructing a deep denoising network with multi-channel auxiliary sensing sensor signals as input and main imaging sensor noise signals as output, and learning its mapping relationship through supervised training; (5) when the target magnetic field exists, using the trained network to reconstruct and suppress the noise components in the main imaging sensor signal; (6) mapping the denoised main imaging sensor signal to a two-dimensional magnetic field distribution according to the two-dimensional relative position mapping model to achieve magnetic field imaging. This invention achieves denoising and two-dimensional imaging of magnetic field signals under complex interference environments through multi-sensor collaborative deep learning noise modeling and two-dimensional spatial mapping.

Description

Technical Field

[0001] This invention relates to the field of magnetic field signal processing, specifically to a magnetic field imaging method with an embedded deep denoising network. Background Technology

[0002] Magnetic field detection, as a non-contact and non-destructive testing method, has shown promising application prospects in the field of engineering structure condition monitoring and fault diagnosis in recent years. During operation, a current-carrying conductor generates a magnetic field in its surrounding space that is closely related to the current distribution. This magnetic field not only reflects the geometry of the current-carrying structure but also contains important characteristics of its internal operating state and potential anomalies. By rationally arranging a magnetic sensor array to collect spatial magnetic fields at multiple points, and processing and reconstructing the obtained magnetic field signals, abstract magnetic field data can be transformed into intuitive two-dimensional magnetic field imaging images, thereby achieving imaging characterization of the structure and state of the current-carrying conductor. Compared with traditional destructive or contact testing methods, this magnetic field detection imaging method has the advantages of strong non-invasiveness, wide applicability, and good adaptability to complex structures, providing an effective approach for the non-destructive testing and visualization analysis of current-carrying structures.

[0003] However, in practical applications, magnetic field signals are typically weak and highly susceptible to environmental noise and electromagnetic interference. This is especially true in complex industrial settings, where interference from mechanical equipment operation, power system activity, and external electromagnetic radiation often couples strongly with the target magnetic signal in both the frequency and time domains, severely reducing the signal-to-noise ratio and spatial resolution of magnetic field imaging results. Existing magnetic signal denoising methods largely rely on traditional filtering or empirical parameter settings, making it difficult to simultaneously achieve noise suppression and preserve the complete structural characteristics of the magnetic field distribution in a noisy environment. This limits the effectiveness of magnetic field imaging in refined detection and condition assessment. Furthermore, these methods generally process the noisy signal from the main imaging sensor itself, lacking independent observation and modeling methods for environmental magnetic noise, making effective separation difficult when the target magnetic field and noise are highly coupled. Therefore, developing a magnetic signal denoising method that can effectively suppress interference and maintain the integrity of magnetic field imaging characteristics in complex noisy environments has become a key issue in improving the accuracy and practicality of magnetic field detection imaging.

[0004] On the other hand, deep learning-based denoising methods, by constructing data-driven nonlinear mapping models, can automatically learn the complex relationship between the magnetic signals of the auxiliary sensing sensor and the main imaging sensor from a large number of samples, providing a new technical path for noise suppression in magnetic field detection imaging. Compared with traditional denoising methods that rely on prior assumptions and fixed parameters, deep learning models have stronger feature representation capabilities and adaptive learning capabilities. They can jointly model the distribution characteristics of magnetic field signals at the time, frequency, and spatial domains, thereby effectively distinguishing target magnetic field information from multi-source interference components in a noisy environment. Especially in magnetic sensor array measurement scenarios, deep learning methods can fully exploit the correlation and redundancy information between multi-channel magnetic signals, improving denoising performance while maintaining the integrity of the magnetic field distribution structure and spatial continuity, providing strong support for high-quality magnetic field imaging. Therefore, deep learning-based magnetic signal denoising methods exhibit higher robustness and generalization potential in complex industrial environments, becoming an effective means to improve the accuracy and reliability of magnetic field detection imaging. Summary of the Invention

[0005] The technical solution of this invention is as follows: A magnetic field imaging method with an embedded deep denoising network is proposed. Magnetic field information is acquired by constructing a sensor array consisting of a main imaging sensor and an auxiliary sensing sensor. Utilizing the independent observation capability of the auxiliary sensing sensor for environmental magnetic noise, a deep learning model is used to characterize the mapping relationship between the noise components of the auxiliary sensing sensor and the main imaging sensor, achieving reconstructive suppression of noise from the main imaging sensor. Based on this, a relative displacement-driven two-dimensional spatial mapping mechanism is combined to reconstruct the denoised magnetic field signal into a two-dimensional magnetic field distribution image. This achieves effective denoising and intuitive presentation of magnetic field imaging results in complex magnetic field environments. The denoising process and the magnetic field imaging process are completed collaboratively within the same spatial mapping framework.

[0006] The technical solution adopted in this invention is as follows:

[0007] A magnetic field imaging method with an embedded deep denoising network includes the following steps:

[0008] S1: Construct a magnetic field signal denoising and imaging experimental platform, deploy a main imaging sensor array in a preset two-dimensional spatial region, and deploy auxiliary sensing sensor arrays on both sides of the main imaging sensor array, so that the auxiliary sensing sensor arrays avoid the area directly below the current-carrying conductor being measured in space.

[0009] S2: Based on the preset two-dimensional grid division parameters of the measurement area and the layout parameters of the main imaging sensor array, a two-dimensional relative position mapping model between the spatial coordinates of the main imaging sensor and the surface coordinates of the measured magnetic field source is established to determine the correspondence between the sensor acquisition points and the surface area of ​​the magnetic field source, providing a position calibration basis for the two-dimensional imaging of the one-dimensional magnetic field signal.

[0010] S3: Under the condition that the current-carrying conductor under test is not energized and only environmental interference sources exist, the magnetic field signals of the auxiliary sensing sensor array and the main imaging sensor array are collected simultaneously to construct an environmental magnetic noise learning dataset that does not contain the target magnetic field component.

[0011] S4: Construct a deep denoising network model, using the noise signal of the multi-channel auxiliary sensing sensor as input and the noise signal of each main imaging sensor as output, and perform supervised training on the network to learn the nonlinear mapping relationship between the auxiliary sensing sensor and the main imaging sensor.

[0012] S5: Under the condition that environmental interference sources exist and the current-carrying conductor being measured generates a target magnetic field, the magnetic field signals of the main imaging sensor and the auxiliary sensing sensor are acquired simultaneously, and the environmental magnetic noise component in the signals of each main sensor is reconstructed through a trained deep denoising network and removed from the magnetic field signal acquired by the main imaging sensor.

[0013] S6: After completing the denoising process for each main imaging sensor, according to the spatial mapping relationship established in step S2, the denoised magnetic field signals of each main imaging sensor are mapped into two-dimensional spatial magnetic field distribution data according to their positions, and the corresponding two-dimensional magnetic field imaging image is generated, thereby realizing magnetic field imaging with an embedded deep denoising network.

[0014] Furthermore, the magnetic field signal denoising and imaging experimental platform described in S1 includes a magnetic sensor array, a motion actuator for driving the magnetic sensor array to move relative to each other within a two-dimensional measurement area, a circuit board with etched wires of a specific configuration, a controllable magnetic noise source module, a data acquisition module, and a computer.

[0015] Furthermore, in S2, the two-dimensional relative position mapping model constructs a two-dimensional measurement plane parallel to the surface of the current-carrying conductor being measured, and maps the magnetic field signals obtained by the main imaging sensor at each acquisition position to unique spatial coordinates in the two-dimensional measurement plane, so that the denoised magnetic field signal of the main imaging sensor serves as the numerical source of the corresponding pixel in the two-dimensional magnetic field imaging image; the vertical distance between the measurement plane and the surface of the current-carrying conductor being measured is fixed as... Thus forming a measurement coordinate system ,in shaft and The axis defines the two-dimensional coordinates of the measurement plane; based on the dimensions of the target being measured, a rectangular measurement area is delineated within the measurement plane, with dimensions of [missing information]. The rectangular measurement area is then divided into grids according to preset image resolution requirements; The number of sampling points in the direction is , The number of sampling points in the direction is Then adjacent sampling points are The spacing in the direction is ,exist The spacing in the direction is Each sampling point thus obtained corresponds to a unique two-dimensional coordinate. ,in , The two-dimensional coordinates serve as spatial indexes for pixels in subsequent two-dimensional images.

[0016] Furthermore, in S3, the ambient magnetic noise is generated by the controllable magnetic noise source module. During the training phase, by turning off the current excitation of the current-carrying conductor under test, the magnetic field signals collected by the auxiliary sensing sensor and the main imaging sensor contain only the ambient magnetic interference component.

[0017] Furthermore, the magnetic sensor arrays in S3 and S5 move point by point within the two-dimensional measurement area according to a preset scanning path under the drive of the motion actuator, and remain stationary at each acquisition position to complete the acquisition of magnetic field signals, thereby ensuring the spatial sampling consistency between the training phase and the imaging phase.

[0018] Furthermore, the input of the deep denoising network described in S4 is the one-dimensional noise signal from the multi-channel auxiliary sensing sensor, and the output is the environmental magnetic noise component corresponding to each main imaging sensor. The model training is completed using supervised learning.

[0019] Furthermore, the two-dimensional image conversion process of the main imaging sensor described in S6 involves first storing and managing the denoised magnetic field signals of each main imaging sensor according to their corresponding acquisition positions, and using the statistical characteristic value of the magnetic field signal of the main imaging sensor at each acquisition position to characterize the magnetic field strength at that position; then, mapping the magnetic field strength to pixel values; finally, based on the two-dimensional relative position mapping model established in step S2, arranging and combining the pixel values ​​corresponding to each acquisition position to generate a two-dimensional magnetic field imaging image that corresponds one-to-one with the spatial distribution of the main imaging sensor.

[0020] The advantages and positive effects of this invention are as follows:

[0021] (1) Embedded deep denoising network to achieve high-precision magnetic field signal purification: This invention embeds a deep denoising network in the magnetic field imaging process to construct a noise correlation model between the auxiliary sensing sensor and the main imaging sensor. This model can adaptively capture complex nonlinear interference in the magnetic field signal, accurately reconstruct noise features, and achieve efficient decoupling of signal and noise. Compared with traditional methods, this significantly improves the denoising accuracy of the magnetic field signal, can stably adapt to the magnetic field detection needs of multiple scenarios, and provides highly reliable raw data support for subsequent imaging and analysis, ensuring the quality of magnetic field imaging from the source.

[0022] (2) Innovative Imaging Conversion for Highly Visualized Magnetic Field Distribution: This invention breaks through the limitations of traditional one-dimensional signal analysis by innovatively converting the one-dimensional magnetic field signal before and after deep denoising into two-dimensional imaging data. Combined with the two-dimensional image stitching technology of multi-position master imaging sensors, it successfully transforms the abstract quantified magnetic field signal data into an intuitive visual image. This not only clearly highlights the differences in denoising effect of the deep denoising network, but also accurately presents the spatial distribution characteristics and strength gradient of the magnetic field, significantly reducing the difficulty of analyzing magnetic field detection results and providing a clear and intuitive basis for judgment in applications such as magnetic field anomaly identification and regional positioning.

[0023] (3) The process is systematic and standardized, with high engineering operability and promotional value: This invention constructs a complete technical process covering the construction of a standardized experimental platform, synchronous acquisition of multi-sensor signals, training of a deep denoising network, and signal-to-image conversion. The sensor array configuration is scientific and reasonable, the training logic of the embedded deep denoising network is clear and the parameters are adjustable, and the technical path of the entire method is standardized and controllable, with strong engineering operability. At the same time, this process can be adapted to the magnetic field imaging needs in different scenarios, is easy to scale up and apply, and has broad practical promotional value and industrialization prospects. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the implementation of the method of the present invention;

[0025] Figure 2 This is a schematic diagram of a sensor array;

[0026] Figure 3 The time-domain waveform of the magnetic field signal collected by the auxiliary sensing sensor at the position coordinate (1,1) under the condition of no current-carrying conductor current excitation and only environmental noise interference.

[0027] Figure 4 The time-domain waveform of the magnetic field signal acquired by the main imaging sensor at position coordinates (1,1) under conditions of no current-carrying conductor current excitation and only environmental noise interference.

[0028] Figure 5 The time-domain waveform of the magnetic field signal collected by the auxiliary sensing sensor at the position coordinate (1,1) under the condition that the current-carrying conductor is energized and there is environmental magnetic noise interference.

[0029] Figure 6 The time-domain waveform of the magnetic field signal superimposed on the target magnetic field and the environmental magnetic noise, acquired by the main imaging sensor at position coordinates (1,1), under the condition that the current-carrying conductor is energized and there is environmental magnetic noise interference.

[0030] Figure 7 The waveform of the magnetic field noise signal at position coordinates (1,1) is the time-domain waveform of the signal reconstructed by a deep denoising network.

[0031] Figure 8 The waveform of the magnetic field signal at position coordinates (1,1) after processing by a deep denoising network is shown in the time domain.

[0032] Figure 9 The training process of the deep denoising network and the magnetic field signal prediction process are illustrated in the diagram.

[0033] Figure 10 This is a comparison of the two-dimensional magnetic field signal imaging results before and after noise reduction processing. Detailed Implementation

[0034] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0035] This invention provides a magnetic field imaging method with an embedded depth denoising network, such as... Figure 1 As shown, the method includes the following steps:

[0036] S1: Construct a magnetic field signal denoising and imaging experimental platform, deploy a main imaging sensor array in a preset two-dimensional spatial region, and deploy auxiliary sensing sensor arrays on both sides of the main imaging sensor array, so that the auxiliary sensing sensor arrays avoid the area directly below the current-carrying conductor being measured in space.

[0037] S2: Based on the preset two-dimensional grid division parameters of the measurement area and the layout parameters of the main imaging sensor array, a two-dimensional relative position mapping model between the spatial coordinates of the main imaging sensor and the surface coordinates of the measured magnetic field source is established to determine the correspondence between the sensor acquisition points and the surface area of ​​the magnetic field source, providing a position calibration basis for the two-dimensional imaging of the one-dimensional magnetic field signal.

[0038] S3: Under the condition that the current-carrying conductor under test is not energized and only environmental interference sources exist, the magnetic field signals of the auxiliary sensing sensor array and the main imaging sensor array are collected simultaneously to construct an environmental magnetic noise learning dataset that does not contain the target magnetic field component.

[0039] S4: Construct a deep denoising network model, using the noise signal of the multi-channel auxiliary sensing sensor as input and the noise signal of each main imaging sensor as output, and perform supervised training on the network to learn the nonlinear mapping relationship between the auxiliary sensing sensor and the main imaging sensor.

[0040] S5: Under the condition that environmental interference sources exist and the current-carrying conductor being measured generates a target magnetic field, the magnetic field signals of the main imaging sensor and the auxiliary sensing sensor are acquired simultaneously, and the environmental magnetic noise component in the signals of each main sensor is reconstructed through a trained deep denoising network and removed from the magnetic field signal acquired by the main imaging sensor.

[0041] S6: After completing the denoising process for each main imaging sensor, according to the spatial mapping relationship established in step S2, the denoised magnetic field signals of each main imaging sensor are mapped into two-dimensional spatial magnetic field distribution data according to their positions, and the corresponding two-dimensional magnetic field imaging image is generated, thereby realizing magnetic field imaging with an embedded deep denoising network.

[0042] Example 1:

[0043] The magnetic sensor array structure constructed in step S1 of the method according to the present invention is as follows: Figure 2 As shown, multiple main imaging sensors are arranged within a preset two-dimensional spatial region, and multiple auxiliary sensing sensors are arranged on the left and right sides of the main imaging sensors. A circuit board with specially shaped wires is placed below the main imaging sensors, and is placed as far below the auxiliary sensing sensors as possible to reduce the interference of the magnetic field generated by the powered circuit board on the measurement results of the auxiliary sensing sensors. Each magnetic sensor module consists of a tunnel magnetoresistive (TMR) sensor mounted on the circuit board. In order to achieve synchronous acquisition of magnetic field components in the X-axis and Y-axis directions, the TMR sensors in the same magnetic sensor module are arranged in a mutually orthogonal manner.

[0044] Example 2:

[0045] According to step S2 of the method of the present invention, multiple sampling points are divided in a preset two-dimensional plane according to regular spacing to form an X×Y rectangular sampling grid, and each sampling point is represented by two-dimensional coordinates. Characterization; wherein the height between the circuit board and the magnetic sensor array is a fixed value. ; using mutually orthogonally arranged tunnel magnetoresistive sensors at a fixed height Simultaneous collection of data from each sampling point The magnetic field strength data along the x and y axes of the location were obtained, and the corresponding magnetic field strength data along the z axis was calculated based on the Biot-Savart law, denoted as follows: , and .

[0046] Example 3:

[0047] According to steps S3 and S4 of the method of the present invention, magnetic field signals of the auxiliary sensing sensor and the main imaging sensor are first synchronously acquired at different locations under the condition that only environmental interference sources exist. By turning off the current in the current-carrying conductor at this stage, and retaining only the environmental interference source, the network training process focuses on learning the characteristics of environmental magnetic noise; wherein, the environmental interference source is a sinusoidal signal generated by the output of a digital signal generator and amplified by a energized solenoid and an operational amplifier module; when acquiring experimental data, the moving platform moves according to a preset path and stops for 0.2s after reaching each preset acquisition position; during the stop, the data acquisition card records the magnetic field strength data at the corresponding position; the data acquisition process continues until the entire area covered by the circuit board is scanned; the acquired data will be automatically stored in a spreadsheet for subsequent analysis and processing; in addition, the step size of the motor is adjustable, and by setting different step sizes, magnetic field images of different resolutions can be acquired. The magnetic field image acquired in this experiment has a resolution of 32×32; taking the signal acquired at position coordinates (1,1) as an example, the waveform of the noise signal acquired by the auxiliary sensing sensor and the main imaging sensor is as follows. Figure 3 and Figure 4 As shown; subsequently, according to step S3 of the method of the present invention, a deep learning network model is built, taking the noise signal of the multi-channel auxiliary sensing sensor as input and the noise signal of each main imaging sensor as output, and performing supervised training on the network to obtain the nonlinear mapping relationship between the auxiliary sensing sensor and the signal of each main imaging sensor. The network training process is as follows. Figure 9 As shown in (a).

[0048] Example 4:

[0049] According to step S5 of the method of the present invention, the circuit board with etched wires of a specific configuration is powered on. According to the Biot-Savart law, the circuit board generates a target magnetic field when powered on. Simultaneously, magnetic field signals from the auxiliary sensing sensor and the main imaging sensor are collected at different locations under the influence of interference sources. Since the circuit board is positioned below the main imaging sensor and avoids being positioned directly below the auxiliary sensing sensor as much as possible, the influence of the target magnetic field generated by the circuit board on the auxiliary sensing sensor is negligible. Under this arrangement, the magnetic field signal collected by the auxiliary sensing sensor mainly reflects the propagation characteristics of the environmental interference source in space, thus providing an observable basis for subsequent noise modeling based on deep learning. Under these conditions, the magnetic field signal collected by the main imaging sensor at its corresponding position is the superposition signal of the target magnetic field generated by the circuit board and the external noise generated by the interference source, while the magnetic field signal collected by the auxiliary sensing sensor at its corresponding position is only a spatial mapping of the external noise generated by the interference source. The waveforms of the magnetic field signals collected by the auxiliary sensing sensor and the main imaging sensor at position coordinates (1,1) are respectively as shown below. Figure 5 , Figure 6As shown; using a trained deep denoising network model, the noise signal of each main imaging sensor can be reconstructed from the noise signal of the auxiliary sensing sensor, such as... Figure 7 As shown; by subtracting the reconstructed noise signal from the one-dimensional magnetic field signal of the main imaging sensor under current excitation and interference from a noisy source, the denoised magnetic field signal waveform can be obtained, as shown. Figure 8 As shown, the network prediction process is as follows: Figure 9 As shown in (b).

[0050] Example 5:

[0051] According to step S6 of the method of the present invention, based on the two-dimensional relative position mapping model established in step S2, the one-dimensional magnetic field signal before and after denoising obtained under the current excitation condition are converted into two-dimensional images respectively; the two-dimensional coordinates of each sampling point are... Mapped to pixel coordinates of a two-dimensional image, where the sampling points The corresponding image line, number The pixels in the column; further, the pixels at the sampling points , and Magnetic field strength data is mapped to the R, G, and B channels of an image, and then processed using an image format. The three-channel data of each pixel are arranged according to their pixel coordinates to generate a corresponding standard two-dimensional image file, such as... Figure 10 As shown.

[0052] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

[0053] Although embodiments and drawings of the present invention have been disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, variations, and modifications are possible without departing from the spirit and scope of the invention and the appended claims. Therefore, the present invention should not be limited to the content disclosed in the embodiments and drawings.

Claims

1. A magnetic field imaging method with an embedded depth denoising network, characterized in that, Includes the following steps: S1: Construct a magnetic field signal denoising and imaging experimental platform, deploy a main imaging sensor array in a preset two-dimensional spatial region, and deploy auxiliary sensing sensor arrays on both sides of the main imaging sensor array, so that the auxiliary sensing sensor arrays avoid the area directly below the current-carrying conductor being measured in space. S2: Based on the preset two-dimensional grid division parameters of the measurement area and the layout parameters of the main imaging sensor array, a two-dimensional relative position mapping model is established between the spatial coordinates of the main imaging sensor and the surface coordinates of the measured magnetic field source. This determines the correspondence between the sensor acquisition points and the surface area of ​​the magnetic field source, providing a position calibration basis for the two-dimensional imaging of the one-dimensional magnetic field signal. The two-dimensional relative position mapping model uses the surface of the current-carrying conductor under test as a reference, constructing a two-dimensional measurement plane parallel to it. The magnetic field signals obtained by the main imaging sensor at each acquisition position are mapped one-to-one with the unique spatial coordinates in the two-dimensional measurement plane, making the denoised magnetic field signal of the main imaging sensor the numerical source of the corresponding pixel in the two-dimensional magnetic field imaging image. The vertical distance between the measurement plane and the surface of the current-carrying conductor under test is fixed as... Thus forming a measurement coordinate system ,in shaft and The axis defines the two-dimensional coordinates of the measurement plane; based on the dimensions of the target being measured, a rectangular measurement area is delineated within the measurement plane, with dimensions of [missing information]. The rectangular measurement area is then divided into grids according to preset image resolution requirements; The number of sampling points in the direction is , The number of sampling points in the direction is Then adjacent sampling points are The spacing in the direction is ,exist The spacing in the direction is Each sampling point thus obtained corresponds to a unique two-dimensional coordinate. ,in , The two-dimensional coordinates serve as spatial position indices for pixels in subsequent two-dimensional images; S3: Under the condition that the current-carrying conductor under test is not energized and only environmental interference sources exist, the magnetic field signals of the auxiliary sensing sensor array and the main imaging sensor array are collected simultaneously to construct an environmental magnetic noise learning dataset that does not contain the target magnetic field component. S4: Construct a deep denoising network model, using the noise signal of the multi-channel auxiliary sensing sensor as input and the noise signal of each main imaging sensor as output, and perform supervised training on the network to learn the nonlinear mapping relationship between the auxiliary sensing sensor and the main imaging sensor. S5: Under the condition that environmental interference sources exist and the current-carrying conductor being measured generates a target magnetic field, the magnetic field signals of the main imaging sensor and the auxiliary sensing sensor are acquired simultaneously, and the environmental magnetic noise component in the signals of each main sensor is reconstructed through a trained deep denoising network and removed from the magnetic field signal acquired by the main imaging sensor. S6: After completing the denoising process for each main imaging sensor, according to the spatial mapping relationship established in step S2, the denoised magnetic field signals of each main imaging sensor are mapped into two-dimensional spatial magnetic field distribution data according to their positions, and the corresponding two-dimensional magnetic field imaging image is generated, thereby realizing magnetic field imaging with an embedded deep denoising network.

2. The magnetic field imaging method with an embedded depth denoising network according to claim 1, characterized in that: The magnetic field signal denoising and imaging experimental platform described in S1 includes a magnetic sensor array, a motion actuator for driving the magnetic sensor array to move relative to each other within a two-dimensional measurement area, a circuit board with etched wires of a specific configuration, a controllable magnetic noise source module, a data acquisition module, and a computer.

3. The magnetic field imaging method with an embedded depth denoising network according to claim 2, characterized in that: In S3, the ambient magnetic noise is generated by the controllable magnetic noise source module. During the training phase, by turning off the current excitation of the current-carrying conductor under test, the magnetic field signals collected by the auxiliary sensing sensor and the main imaging sensor contain only the ambient magnetic interference component.

4. The magnetic field imaging method with an embedded depth denoising network according to claim 1, characterized in that: In S3 and S5, the magnetic sensor arrays move point by point within the two-dimensional measurement area according to a preset scanning path driven by the motion actuator, and remain stationary at each acquisition position to complete the acquisition of magnetic field signals, thereby ensuring the spatial sampling consistency between the training and imaging stages.

5. The magnetic field imaging method with an embedded depth denoising network according to claim 1, characterized in that: The deep denoising network described in S4 takes a one-dimensional noise signal from a multi-channel auxiliary sensing sensor as input and outputs an environmental magnetic noise component corresponding to each main imaging sensor. The model is trained using supervised learning.

6. The magnetic field imaging method with an embedded depth denoising network according to claim 1, characterized in that: The two-dimensional image conversion process of the main imaging sensor described in S6 is as follows: First, the magnetic field signals of each main imaging sensor after noise reduction are stored and managed according to their corresponding acquisition positions, and the magnetic field strength at each acquisition position is characterized by the statistical characteristic value of the magnetic field signal of the main imaging sensor at each acquisition position; then, the magnetic field strength is mapped to pixel values; finally, according to the two-dimensional relative position mapping model established in step S2, the pixel values ​​corresponding to each acquisition position are arranged and combined to generate a two-dimensional magnetic field imaging image that corresponds one-to-one with the spatial distribution of the main imaging sensor.

Images