A method, system, device and storage medium for reconstructing sparse magnetic field data

Abstract

The application discloses a kind of sparse magnetic field data reconstruction method, system, equipment and storage medium, is related to magnetic field imaging and neural network image processing technical field, including by data acquisition department obtaining the initial sampling sparse magnetic field data of target detection area;Initial sampling sparse magnetic field data is input into sampling strategy department, generates sampling probability distribution graph, according to sampling probability distribution graph, obtains active sampling sparse magnetic field data;Active sampling sparse magnetic field data and initial sampling sparse magnetic field data are input into spatial superposition department, obtain final sampling sparse magnetic field data;Final sampling sparse magnetic field data is input into sparse data reconstruction module, and high-resolution reconstructed magnetic field data is output.The method described in the application is more good in improving sampling efficiency, optimizing reconstruction quality, reducing redundant data, enhancing reconstruction stability.

Description

Technical Field

[0001] This invention belongs to the field of magnetic field imaging and neural network image processing technology, specifically a method, system, device and storage medium for sparse magnetic field data reconstruction. Background Technology

[0002] With the rapid development of precision manufacturing and the new energy industry, the demand for non-destructive testing of power batteries, drive motors, and complex integrated circuits is increasing. In recent years, magnetic field detection technology, as a non-invasive and highly sensitive monitoring method, has shown great application potential in these fields. For example, the charging and discharging process of batteries and the operation of motors both generate characteristic magnetic fields. Therefore, by accurately capturing and analyzing the distribution and changes of these magnetic field signals, it is possible to effectively reflect the internal working status of equipment and potential early faults.

[0003] However, in the application of high-resolution magnetic field imaging, existing technologies face a core contradiction: the constraint between image resolution and acquisition efficiency. On the one hand, to obtain high-resolution, detailed magnetic field images, dense spatial sampling is required within the target area, i.e., acquiring a large number of data points. While this high-density sampling method ensures image quality, it significantly prolongs data acquisition time, reduces detection efficiency, and fails to meet the needs of rapid detection. On the other hand, reducing sampling points to improve acquisition speed, i.e., sparse sampling, results in low effective resolution of the acquired magnetic field data, poor image quality, and an inability to accurately characterize the fine structure and key features of the magnetic field, severely impacting the accuracy of subsequent condition assessment and fault diagnosis.

[0004] To improve detection efficiency and obtain higher-quality magnetic field data, researchers typically employ sparse sampling strategies combined with algorithms to attempt to reconstruct a complete, high-resolution magnetic field distribution from a small number of sampling points. Initially, some researchers used traditional interpolation and analytical methods, such as linear interpolation and bicubic interpolation. These methods only utilize spatial proximity relationships to fill in the data, but they cannot recover the complex nonlinear details in the magnetic field distribution and are prone to severe blurring and artifacts when the sparsity is high.

[0005] Another approach is passive reconstruction using neural network-based methods. This method often employs sensors to sample magnetic field signals at sparsely distributed points in a uniform distribution. This low-resolution data acquisition method based on uniform sampling has significant technical limitations: the bottleneck lies in the complete decoupling of the sampling layout from the characteristics of the magnetic field being measured. The network can only passively adapt to the existing data distribution and cannot dynamically adjust the sampling points according to the complexity of the magnetic field characteristics. This leads to redundant sampling in low-frequency information regions where the magnetic field changes gently, wasting time resources; while in regions with drastic magnetic field changes and containing critical defect information, the density of uniformly distributed sampling points is often insufficient, resulting in the loss of key feature information. If the sampling stage itself misses key singularities or high-frequency textures in the physical field due to insufficient resolution, the subsequent reconstruction algorithm can only "guess" or "smooth" based on statistical laws, easily producing artifacts, blurred edges, or geometric distortions, failing to accurately reconstruct complex nonlinear magnetic field distributions.

[0006] Therefore, how to break through the limitations of traditional uniform sampling, how to design a method that can actively select sampling positions based on the characteristics of magnetic field distribution, maximize information acquisition with a limited number of sampling points, and generate high-fidelity magnetic field images through joint optimization with the reconstruction network, are key challenges that urgently need to be overcome in the field of magnetic field imaging technology. Summary of the Invention

[0007] In view of the above-mentioned problems, the present invention is proposed.

[0008] Therefore, the technical problem solved by this invention is that existing magnetic field image data acquisition and reconstruction methods require a large amount of spatial sampling to obtain high-resolution magnetic field images, resulting in excessively long data acquisition time and low efficiency. Although reducing the number of sampling points improves acquisition efficiency, it leads to low-resolution and inaccurate reconstructed images, affecting the accuracy of subsequent fault diagnosis. To address this, a novel sparse magnetic field image data reconstruction method is proposed, which combines active sampling strategies with sparse data reconstruction. This method effectively improves reconstruction quality and acquisition efficiency by actively selecting key information regions for sampling while reducing the number of sampling points.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a sparse magnetic field data reconstruction method, comprising acquiring initial sampled sparse magnetic field data of the target detection area through a data acquisition unit.

[0010] The initial sampled sparse magnetic field data is input into the sampling strategy unit to generate a sampling probability distribution map. Based on the sampling probability distribution map, the actively sampled sparse magnetic field data is obtained.

[0011] The actively sampled sparse magnetic field data and the initially sampled sparse magnetic field data are input into the spatial superposition part to obtain the final sampled sparse magnetic field data.

[0012] The final sampled sparse magnetic field data is input into the sparse data reconstruction module, which outputs high-resolution reconstructed magnetic field data.

[0013] The sampling strategy unit includes features extraction, sampling strategy formulation, and sampling evaluation of the initial sampled sparse magnetic field data through the active sampling strategy module, and outputs a sampling probability distribution map.

[0014] Acquiring actively sampled sparse magnetic field data involves selecting target locations as active sampling points in the unsampled regions of the sampling probability distribution map.

[0015] The active sampling strategy module and the sparse data reconstruction module need to be trained in stages.

[0016] As a preferred embodiment of the sparse magnetic field data reconstruction method of the present invention, the data acquisition unit includes setting an initial sampling position, performing initial sampling of the real magnetic field through the initial sampling position, and obtaining initial sampled sparse magnetic field data.

[0017] As a preferred embodiment of the sparse magnetic field data reconstruction method of the present invention, the feature extraction includes performing convolutional encoding on the initial sampled sparse magnetic field data to extract a high-dimensional shared feature map containing the hidden physical field spatial topology and local correlation features.

[0018] The sampling strategy includes channel compression and nonlinear mapping of the high-dimensional shared feature map, and outputting a sampling probability heatmap with the same size as the target detection region.

[0019] The sampling evaluation includes a global analysis of the high-dimensional shared feature map and outputting a value score for the current observation state.

[0020] As a preferred embodiment of the sparse magnetic field data reconstruction method of the present invention, the step-by-step model training includes constructing a magnetic field training dataset and performing two-stage training.

[0021] The two-stage training includes the first stage, which uses the initial sampled sparse magnetic field data to supervise the training of the sparse data reconstruction module. By constructing and minimizing the reconstruction quality loss, the parameters of the sparse data reconstruction module are optimized through backpropagation.

[0022] The second phase of joint training establishes a closed-loop feedback between the active sampling strategy module and the pre-trained sparse data reconstruction module.

[0023] A reward function is constructed based on the reconstruction quality loss and the effectiveness of the sampling points, and a sampling strategy loss is constructed based on the reward function.

[0024] The active sampling loss is constructed by combining the sampling strategy loss, evaluation loss, and entropy regularization loss. The parameters of the active sampling strategy module are optimized by backpropagation by minimizing the active sampling loss, and the parameters of the sparse data reconstruction module are fine-tuned by backpropagation by minimizing the reconstruction quality loss.

[0025] As a preferred embodiment of the sparse magnetic field data reconstruction method of the present invention, the parameters of the optimized active sampling strategy module include calculating the difference between the reward function and the value score of the sampling evaluation output to obtain the sampling advantage value.

[0026] The sampling dominance value is used to characterize the extent to which the actual reconstruction benefits brought about by the current sampling action exceed expectations.

[0027] The sampling policy loss is calculated using the sampling advantage value, and the parameters of the sampling policy are updated by combining the entropy regularization loss.

[0028] As a preferred embodiment of the sparse magnetic field data reconstruction method of the present invention, the parameters of the updated sampling strategy include establishing a gradient channel between discrete sampling actions and network weights based on the sampling strategy loss.

[0029] The information entropy of the sampling probability distribution is maximized by using entropy regularization loss.

[0030] The evaluation loss is calculated using the mean squared error between the value score and the actual reward. The parameters of the sampled evaluation are then updated, and the value score is calibrated as a baseline for strategy optimization.

[0031] Another objective of this invention is to provide a sparse magnetic field data reconstruction system that solves the problems of sampling point redundancy, information omission, and large reconstruction errors in current traditional uniform sampling methods by combining active sampling strategies and sparse data reconstruction modules.

[0032] As a preferred embodiment of the sparse magnetic field data reconstruction system of the present invention, it includes a magnetic field acquisition area, an active decision-making area, and a reconstruction area.

[0033] The magnetic field acquisition area is used to acquire initial sampled sparse magnetic field data of the target detection area through the data acquisition unit; the initial sampled sparse magnetic field data is input into the sampling strategy unit to generate a sampling probability distribution map, and active sampled sparse magnetic field data is acquired based on the sampling probability distribution map.

[0034] The active decision region is used to input the actively sampled sparse magnetic field data and the initial sampled sparse magnetic field data into the spatial superposition region to obtain the final sampled sparse magnetic field data.

[0035] The reconstruction region is used to input the final sampled sparse magnetic field data into the sparse data reconstruction module and output high-resolution reconstructed magnetic field data.

[0036] Another object of the present invention is to provide a sparse magnetic field data reconstruction device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a sparse magnetic field data reconstruction method.

[0037] Another object of the present invention is to provide a sparse magnetic field data reconstruction storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a sparse magnetic field data reconstruction method.

[0038] The beneficial effects of this invention are as follows: The sparse magnetic field data reconstruction method provided by this invention intelligently selects key information regions for sampling, reducing redundant sampling points and effectively improving data acquisition efficiency. It also enhances the quality of the reconstruction results, avoiding the problem of insufficient sampling points in high-frequency information regions inherent in traditional uniform sampling methods. By combining the nonlinear mapping capability of the sparse data reconstruction module, it significantly improves the reconstruction accuracy of low-sampling-rate data, successfully reconstructing high-resolution magnetic field images. This reduces artifacts and blurring during the reconstruction process, effectively preventing excessive clustering of sampling points and ensuring a reasonable spatial distribution of sampling points, thereby improving the comprehensiveness of information acquisition. Through joint training and a reinforcement learning feedback mechanism, the sampling strategy and reconstruction quality are mutually optimized, ultimately achieving high-precision magnetic field data reconstruction with limited sampling points. This invention achieves better results in improving sampling efficiency, optimizing reconstruction quality, reducing redundant data, and enhancing reconstruction stability. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is an overall flowchart of a sparse magnetic field data reconstruction method provided in Embodiment 1 of the present invention.

[0041] Figure 2 The network structure diagram of the active sampling strategy module and the sparse data reconstruction module in the sparse magnetic field data reconstruction method provided in Embodiment 1 of the present invention.

[0042] Figure 3 The image shows the result of sparse magnetic field image reconstruction, which is a simulation of a sparse magnetic field data reconstruction method provided in Embodiment 2 of the present invention.

[0043] Figure 4The image shows a comprehensive reconstruction index of a simulated sparse magnetic field image dataset for a sparse magnetic field data reconstruction method provided in Embodiment 2 of the present invention.

[0044] Figure 5 The image shows the actual acquisition result of sparse magnetic field image data reconstruction for a sparse magnetic field data reconstruction method provided in Embodiment 2 of the present invention. Detailed Implementation

[0045] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0046] Example 1, referring to Figures 1-2 As an embodiment of the present invention, a method for reconstructing sparse magnetic field data is provided, comprising:

[0047] S1: Acquire the initial sampled sparse magnetic field data D1 of the target detection area through the data acquisition unit 100.

[0048] Set an initial sampling position, and perform initial sampling of the real magnetic field at the initial sampling position to obtain the initial sampled sparse magnetic field data D1.

[0049] Specifically, the real complete magnetic field data is initially discretized and observed using a preset initial uniform sampling mask.

[0050] The simulated sensor performs a low-cost, low-resolution preliminary scan in the physical field to acquire initial sampled sparse magnetic field data D1.

[0051] A preferred method for obtaining the initial sampled sparse magnetic field data D1 is as follows:

[0052] A magnetic field acquisition platform was constructed, which included a two-dimensional motion platform composed of two stepper motors, a sensor probe unit composed of an MR2103 magnetic sensor chip, an NI USB-6289 data acquisition card, a computer, a programmable DC power supply DP800, and a numerically controlled DC power supply SPE3102.

[0053] The MR2103 magnetic sensor has a sensitivity of 5 mV / Gs.

[0054] The sensor probe is fixed on a specially customized non-magnetic material base.

[0055] The design aims to fundamentally eliminate the interference of the base itself on magnetic field measurements, thereby ensuring the stability and fidelity of the acquired data.

[0056] The circuit board under test is securely mounted directly below the sensor array.

[0057] The two-dimensional motion platform is mounted on two orthogonally fixed guide rails. The two-dimensional motion platform is driven by two stepper motors. The motors are driven by lead screws, which enable the magnetic field acquisition platform to make precise linear displacements along the x-axis and y-axis, respectively, thereby realizing grating scanning of the target area to systematically cover the entire imaging range.

[0058] The operation process of the two-dimensional motion platform is designed as an automated sequence of alternating step-acquisition cycles.

[0059] During the experiment, the stepper motor strictly followed the preset parameters to perform step-by-step movement.

[0060] During the brief pause after each step displacement is completed, the NI USB-6289 data acquisition card will immediately trigger and synchronously record the magnetic field strength data at the current spatial coordinate point.

[0061] Because the sensing directions of adjacent channels on the sensor chip are perpendicular to each other, the magnetic field acquisition platform can simultaneously capture magnetic field vector components in different directions, thereby obtaining more comprehensive multidimensional magnetic field information.

[0062] During initial sampling, the magnetic field acquisition platform automatically calculates the basic movement step size of the two-dimensional motion platform based on the preset full target resolution:

[0063] When the target resolution is set to 128×128, the platform's basic step size is 5mm.

[0064] Then, based on the input uniform sparse position mask, magnetic field data is sampled at the target location to be sampled.

[0065] The data collected at each sampling point is recorded in real time in a spreadsheet according to its position identified by a uniform sparse location mask, providing structured raw data for subsequent analysis and processing.

[0066] This coordinated process of movement and data acquisition will continue until all the preset uniform sampling points have been collected.

[0067] S2: Input the initial sampled sparse magnetic field data D1 into the sampling strategy unit 200 to generate a sampling probability distribution map, and obtain the active sampled sparse magnetic field data D2 based on the sampling probability distribution map.

[0068] The active sampling strategy module 201 performs feature extraction, sampling strategy, and sampling evaluation on the initial sampled sparse magnetic field data D1, and outputs a sampling probability distribution map.

[0069] The active sampling strategy module 201 includes a sampling feature extraction module 2011, a sampling strategy module 2012, and a sampling evaluation module 2013.

[0070] The sampling feature extraction module 2011 performs convolutional encoding on the initial sampled sparse magnetic field data D1 to extract a high-dimensional shared feature map containing the hidden physical field spatial topology and local correlation features.

[0071] The sampling strategy module 2012 performs channel compression and nonlinear mapping on the high-dimensional shared feature map, and outputs a sampling probability heatmap with the same size as the target detection region.

[0072] The sampling evaluation module 2013 performs a global analysis of the high-dimensional shared feature map and outputs a value score for the current observation state.

[0073] Acquiring active sampling sparse magnetic field data D2 includes selecting target points as active sampling points D2 in the unsampled region of the sampling probability distribution map;

[0074] The active sampling strategy module 201 and the sparse data reconstruction module 301 need to undergo step-by-step model training.

[0075] Specifically, a magnetic field training dataset is constructed, and two-stage training is performed.

[0076] The two-stage training includes the first stage using the initial sampled sparse magnetic field data D1 to supervise the training of the sparse data reconstruction module 301. By constructing and minimizing the reconstruction quality loss, the parameters of the sparse data reconstruction module 301 are optimized through backpropagation.

[0077] The second phase of joint training involves constructing a closed-loop feedback between the active sampling strategy module 201 and the pre-trained sparse data reconstruction module 301.

[0078] A reward function is constructed based on the reconstruction quality loss and the effectiveness of the sampling points, and a sampling strategy loss is constructed based on the reward function.

[0079] The active sampling loss is constructed by combining the sampling strategy loss, evaluation loss, and entropy regularization loss. The parameters of the active sampling strategy module 201 are optimized by backpropagation by minimizing the active sampling loss, and the parameters of the sparse data reconstruction module 301 are fine-tuned by backpropagation by minimizing the reconstruction quality loss.

[0080] Furthermore, the reward function is composed of a weighted sum of reconstruction quality reward and dispersion penalty term.

[0081] A preferred solution for the reconstruction quality loss is:

[0082] ,

[0083] in, Indicates global pixel loss. Indicating resistance to perceptual loss, The coefficients represent the global pixel loss term. This represents the coefficient of the adversarial perception loss term.

[0084] The mathematical expression for the two-sub-loss term is:

[0085] ,

[0086] ,

[0087] in, Indicates the reconstructed image. This represents the original, complete image. Indicates that the network reconstructs the image in The value at the point. Indicates the original complete image in The value at the point. and This represents the number of pixels in the image, both vertically and horizontally.

[0088] A preferred scheme for constructing the dispersion penalty term is:

[0089] ,

[0090] ,

[0091] ,

[0092] ,

[0093] in, Indicates the total number of sampling points. This is a local minimum value used to prevent gradient explosion. The weighting coefficients represent the dispersion penalty term. This represents the dispersion penalty term. This represents a mask matrix, which is used to mark the locations of sampling points. Indicates the mask matrix in The value at a point is either 0 (for non-sampled points) or 1 (for sampled points), and the nearest neighbor penalty operator is used. This represents a 3×3 matrix (with a center value of 0 and edge values ​​of 1) used to detect whether there are other sampling points in the nearest neighbor region of a sampling point. The second nearest neighbor penalty operator is used for this purpose. This represents a 5×5 circular matrix (with a 3×3 center value of 0 and an edge value of 1) used to detect whether there are other sampling points in the second nearest neighbor of a sampling point. and The weights of the nearest neighbor penalty term and the second nearest neighbor penalty term.

[0094] The learning results are guaranteed by reconstructing quality rewards, and excessive spatial clustering of sampling points in high-frequency information regions is suppressed by using a discreteness penalty term.

[0095] Suppressing excessive spatial clustering of sampling points in high-frequency information regions by using a discreteness penalty term includes performing convolution operations on the sampling mask using a fixed convolution kernel, calculating the sampling point density in the local neighborhood, and applying a negative penalty value to high-density regions, thereby forcing the sampling points generated by the active sampling strategy module 201 to maintain a discrete spatial distribution.

[0096] The active sampling loss is the total loss term of the active sampling network module, and its mathematical expression is:

[0097] ,

[0098] in The sampling strategy loss is calculated by multiplying the difference between the reward function and the value score output by the sampling evaluation to obtain the sampling advantage value, and then multiplying it by the log-likelihood of the sampling action. This represents the weighting coefficient corresponding to the entropy term. This represents the weighting coefficient corresponding to the evaluation item.

[0099] Its mathematical expression is:

[0100] ,

[0101] The mathematical expression for assessing loss is:

[0102] ,

[0103] in, This represents the value of the reward function for that round. This indicates the output value of the sampling evaluation module for this round.

[0104] The mathematical expression for entropy regularization loss is:

[0105] ,

[0106] in, Indicates the mask matrix in The value at the point, This indicates the output value of the sampling strategy module for this round.

[0107] The parameters of the optimized active sampling strategy module 201 include calculating the difference between the reward function and the value score output by the sampling evaluation to obtain the sampling advantage value.

[0108] The sampling dominance value is used to characterize the extent to which the actual reconstruction benefits brought about by the current sampling action exceed expectations.

[0109] The sampling policy loss is calculated using the sampling advantage value, and combined with the entropy regularization loss, the active sampling loss is calculated. The parameters of the sampling policy are then updated via backpropagation.

[0110] Furthermore, a gradient channel between discrete sampling actions and network weights is established based on the sampling strategy loss.

[0111] The information entropy of the sampling probability distribution is maximized by using entropy regularization loss.

[0112] The evaluation loss is calculated using the mean squared error between the value score and the actual reward. The parameters of the sampled evaluation are then updated, and the value score is calibrated as a baseline for strategy optimization.

[0113] S3: Input the actively sampled sparse magnetic field data D2 and the initial sampled sparse magnetic field data D1 into the spatial superposition part 300 to obtain the final sampled sparse magnetic field data D3.

[0114] Extract the corresponding sparse data from the determined active sampling points and the initial uniform sampling points.

[0115] S4: Input the final sampled sparse magnetic field data D3 into the sparse data reconstruction module 301, and output the high-resolution reconstructed magnetic field data D4.

[0116] Input the sparse magnetic field data into the sparse data reconstruction module 301.

[0117] The sparse data reconstruction module 301 includes a generator module 3011 and a discriminator module 3012.

[0118] The generator module 3011 is responsible for outputting reconstructed magnetic field data and comparing it with the original data for optimization.

[0119] The discriminator module 3012 is responsible for self-optimization by distinguishing between real data and reconstructed data.

[0120] The reconstruction module uses its generation capabilities to fill in data gaps and output the final high-resolution reconstructed magnetic field data D4.

[0121] After obtaining the final high-resolution reconstructed magnetic field data D4, it is compared with the original real magnetic field data, and the parameters of the active sampling strategy module 201 are optimized based on the comparison results.

[0122] Specifically, the optimization of the active sampling strategy module 201 encourages the selection of sampling points that significantly reduce reconstruction errors. This enables the module to effectively predict the locations of key information points in unknown regions based on initial sampling point information, ultimately achieving better reconstruction of complete magnetic field data. Simultaneously, the pre-trained sparse data reconstruction module is fine-tuned based on the reconstruction results, allowing the reconstruction network to better adapt to and utilize the non-uniform key information provided by the active sampling points, further improving reconstruction accuracy.

[0123] Through the closed-loop iterative training described above, the two modules co-evolve, ultimately maximizing the reconstruction quality of magnetic field data under the condition of limited sampling points.

[0124] Example 2, refer to Figures 3-5 This invention provides a method for reconstructing sparse magnetic field data. To verify the beneficial effects of the invention, scientific demonstration is conducted through economic benefit calculations and simulation experiments.

[0125] First, to test the effectiveness of the model, simulated magnetic field images were used for testing.

[0126] To quantitatively evaluate the improvement in reconstruction quality of the active sampling strategy compared to the non-active strategy, Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) were selected as evaluation metrics.

[0127] PSNR measures the fidelity of a reconstructed image by calculating the mean squared error (MSE). The larger the value, the smaller the reconstruction error and the higher the image quality.

[0128] SSIM focuses more on the overall reconstruction effect, comprehensively measuring the similarity of images from three dimensions: brightness, contrast, and structure. The closer the result is to 1, the more the reconstructed image matches the ground truth image in terms of structural features.

[0129] We used an initial sampling mode of 20×20 for sampling, and the sampling strategy module actively sampled to supplement 100 sparse sampling points, collecting a total of 500 sparse sampling points. We then compared the reconstruction results with those of various uniform sampling modes such as 20×20, 23×23, and 25×25, as well as the quantitative indicators.

[0130] like Figure 3 , 4As shown, it intuitively demonstrates the magnetic field data reconstruction results and quantitative indicators of this method. The indicators clearly show that, in terms of the PSNR of the reconstructed data, the magnetic field data using the active sampling method with only 500 total sampling points significantly outperforms the data reconstructed using 23×23 (529 sampling points), and is close to the data reconstructed using 25×25 (625 sampling points).

[0131] Furthermore, it outperformed all the comparison methods in terms of the SSIM index of the reconstructed data. However, regarding the stability of the reconstruction results, the use of passive uniform sampling may have overlooked some key information points, leading to poor reconstruction results in some areas and a large number of outliers appearing on the box plot.

[0132] Using active sampling can effectively capture important information areas for supplementary sampling, greatly reducing the probability of losing key information points. The reconstruction results have fewer outliers, reflecting that the active sampling method has better data reconstruction stability and robustness.

[0133] Secondly, to further verify the effectiveness of the method of the present invention in real physical scenarios, the present invention was tested using a physical circuit board with circuit patterns.

[0134] This embodiment constructs a magnetic field acquisition platform based on a mobile sensor. Through the precise coordination of multiple modules, the platform achieves stable magnetic field data acquisition.

[0135] The platform integrates a motion control system, a high-sensitivity sensor unit, and a data acquisition and control unit at the system level.

[0136] Specifically, its core components include: a two-dimensional motion platform consisting of two stepper motors; a sensor probe unit consisting of an MR2103 magnetic sensor chip; and a control and acquisition system consisting of an NI USB-6289 data acquisition card, a computer, a programmable DC power supply DP800, and a numerically controlled DC power supply SPE3102.

[0137] Among them, the MR2103 magnetic sensor has a sensitivity of up to 5 mV / GS, providing hardware support for accurately capturing subtle changes in magnetic fields.

[0138] In terms of physical structure design, the sensor probe is fixed on a specially customized non-magnetic material base.

[0139] The design aims to fundamentally eliminate the interference of the base itself on magnetic field measurements, thereby ensuring the stability and fidelity of the acquired data.

[0140] The circuit board under test is securely mounted directly below the sensor array. The moving platform is placed on two orthogonally fixed guide rails and is driven by two stepper motors. The fixed position rotation of the motors is driven by a lead screw, enabling the platform to make precise linear displacements along the x and y axes, respectively, thereby achieving raster-like scanning of the target area to systematically cover the entire imaging range.

[0141] The platform's operation is designed as an automated sequence of alternating stepping and data acquisition cycles. During the experiment, the stepper motor strictly follows preset parameters for stepping movement.

[0142] During the brief pause after each step displacement, the NI USB-6289 data acquisition card immediately triggers and synchronously records the magnetic field strength data at the current spatial coordinate point. Because the sensing directions of adjacent channels on the sensor chip are perpendicular to each other, this platform can simultaneously capture magnetic field vector components in different directions, thereby obtaining more comprehensive multidimensional magnetic field information.

[0143] During initial sampling, the platform automatically calculates the basic moving step size based on the preset full target resolution, and then samples the magnetic field data at the target location to be sampled based on the input uniform sparse position mask.

[0144] The data collected at each sampling point is recorded in real time in a spreadsheet according to its position identified by a uniform sparse location mask, providing structured raw data for subsequent analysis and processing.

[0145] This coordinated process of movement and data acquisition will continue until all the preset uniform sampling points have been collected.

[0146] The collected sparse magnetic field data is input into a trained active sampling strategy module. This module automatically predicts the information value distribution map of the remaining locations based on the existing location information, and generates an active sampling location mask based on the distribution map.

[0147] The active sampling location mask is transmitted to the magnetic field acquisition platform. Based on the location of the active sampling point, the acquisition platform automatically collects all the required magnetic field information of the points according to the above acquisition process and stores it in a spreadsheet.

[0148] The initial sampled magnetic field data is superimposed with the actively sampled magnetic field data to obtain the final sampled magnetic field data. This data is then input into the sparse data reconstruction module, which will automatically predict and complete the high-resolution magnetic field data based on the known sampling point information, thus achieving high-fidelity and high-resolution acquisition of magnetic field data.

[0149] like Figure 5As shown, it intuitively displays the image of the excitation circuit board, and finally the active sampling magnetic field data and the complete high-resolution magnetic field data reconstructed by the model.

[0150] Example 3, an embodiment of the present invention, provides a sparse magnetic field data reconstruction system, including a magnetic field acquisition area, an active decision-making area, and a reconstruction area.

[0151] The magnetic field acquisition area is used to acquire the initial sampled sparse magnetic field data D1 of the target detection area through the data acquisition unit 100.

[0152] The initial sampled sparse magnetic field data D1 is input into the sampling strategy unit 200 to generate a sampling probability distribution map. Based on the sampling probability distribution map, the actively sampled sparse magnetic field data D2 is obtained.

[0153] The active decision region is used to input the actively sampled sparse magnetic field data D2 and the initial sampled sparse magnetic field data D1 into the spatial superposition part 300 to obtain the final sampled sparse magnetic field data D3.

[0154] The reconstruction region is used to input the final sampled sparse magnetic field data D3 into the sparse data reconstruction module 301 and output high-resolution reconstructed magnetic field data D4.

[0155] This embodiment also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the sparse magnetic field data reconstruction method proposed in the above embodiment.

[0156] This embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the sparse magnetic field data reconstruction method proposed in the above embodiments.

[0157] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0158] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0159] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0160] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0161] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for reconstructing sparse magnetic field data, characterized in that, include: The initial sampled sparse magnetic field data (D1) of the target detection area is acquired by the data acquisition unit (100). The initial sampled sparse magnetic field data (D1) is input into the sampling strategy unit (200) to generate a sampling probability distribution map. Based on the sampling probability distribution map, the actively sampled sparse magnetic field data (D2) is obtained. The actively sampled sparse magnetic field data (D2) and the initial sampled sparse magnetic field data (D1) are input into the spatial overlay unit (300) to obtain the final sampled sparse magnetic field data (D3). The final sampled sparse magnetic field data (D3) is input into the sparse data reconstruction module (301), which outputs high-resolution reconstructed magnetic field data (D4); where, The sampling strategy unit (200) includes extracting features from the initial sampled sparse magnetic field data (D1), formulating a sampling strategy, evaluating the sampling, and outputting a sampling probability distribution map through the active sampling strategy module (201). Acquiring active sampling sparse magnetic field data (D2) involves selecting target points as active sampling points in the unsampled regions of the sampling probability distribution map; The active sampling strategy module (201) and the sparse data reconstruction module (301) need to be trained in stages.

2. The sparse magnetic field data reconstruction method as described in claim 1, characterized in that: The data acquisition unit (100) includes, Set an initial sampling position, and perform initial sampling of the real magnetic field at the initial sampling position to obtain the initial sampled sparse magnetic field data (D1).

3. The sparse magnetic field data reconstruction method as described in claim 1, characterized in that: The feature extraction includes, Convolutional encoding is performed on the initial sampled sparse magnetic field data (D1) to extract a high-dimensional shared feature map containing the hidden physical field spatial topology and local correlation features; The sampling strategy includes channel compression and nonlinear mapping of the high-dimensional shared feature map, and outputting a sampling probability heatmap with the same size as the target detection region. The sampling evaluation includes a global analysis of the high-dimensional shared feature map and outputting a value score for the current observation state.

4. The sparse magnetic field data reconstruction method as described in claim 1 or 3, characterized in that: The step-by-step model training includes, Construct a magnetic field training dataset and perform two-stage training; The two-stage training includes the first stage using the initial sampled sparse magnetic field data (D1) to supervise the training of the sparse data reconstruction module (301), and optimizing the parameters of the sparse data reconstruction module (301) by constructing and minimizing the reconstruction quality loss; The second stage of joint training constructs a closed-loop feedback between the active sampling strategy module (201) and the pre-trained sparse data reconstruction module (301); A reward function is constructed based on the reconstruction quality loss and the effectiveness of the sampling points, and a sampling strategy loss is constructed based on the reward function; The active sampling loss is constructed by combining the sampling strategy loss, evaluation loss and entropy regularization loss. The parameters of the active sampling strategy module (201) are optimized by backpropagation by minimizing the active sampling loss, and the parameters of the sparse data reconstruction module (301) are fine-tuned by backpropagation by minimizing the reconstruction quality loss.

5. The sparse magnetic field data reconstruction method as described in claim 4, characterized in that: The reward function is composed of a weighted sum of a reconstruction quality reward and a dispersion penalty term.

6. The sparse magnetic field data reconstruction method as described in claim 4, characterized in that: The parameters of the optimized active sampling strategy module (201) include, The difference between the reward function and the value score output by the sampling evaluation is calculated to obtain the sampling advantage value; The sampling dominance value is used to characterize the extent to which the actual reconstruction benefits brought about by the current sampling action exceed expectations; The sampling policy loss is calculated using the sampling advantage value, and combined with the entropy regularization loss, the active sampling loss is calculated. The parameters of the sampling policy are then updated via backpropagation.

7. The sparse magnetic field data reconstruction method as described in claim 6, characterized in that: The parameters of the updated sampling strategy include, Establish a gradient channel between discrete sampling actions and network weights based on the sampling strategy loss; Maximize the information entropy of the sampling probability distribution by using entropy regularization loss; The evaluation loss is calculated using the mean squared error between the value score and the actual reward. The parameters of the sampled evaluation are then updated, and the value score is calibrated as a baseline for strategy optimization.

8. A sparse magnetic field data reconstruction system, employing the sparse magnetic field data reconstruction method as described in any one of claims 1 to 7, characterized in that: It includes the magnetic field acquisition area, the active decision-making area, and the reconstruction area; The magnetic field acquisition area is used to acquire the initial sampled sparse magnetic field data (D1) of the target detection area through the data acquisition unit (100). The initial sampled sparse magnetic field data (D1) is input into the sampling strategy unit (200) to generate a sampling probability distribution map. Based on the sampling probability distribution map, the actively sampled sparse magnetic field data (D2) is obtained. The active decision region is used to input the actively sampled sparse magnetic field data (D2) and the initial sampled sparse magnetic field data (D1) into the spatial overlay unit (300) to obtain the final sampled sparse magnetic field data (D3). The reconstruction region is used to input the final sampled sparse magnetic field data (D3) into the sparse data reconstruction module (301) and output high-resolution reconstructed magnetic field data (D4).

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the sparse magnetic field data reconstruction method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the sparse magnetic field data reconstruction method according to any one of claims 1 to 7.

Images