Multi-frequency electromagnetic data intelligent reconstruction method and system for metal cultural relic detection

Abstract

The invention discloses a multi-frequency electromagnetic data intelligent reconstruction method and system for metal cultural relic detection, and relates to the technical field of electromagnetic detection, and the method comprises the steps: S1, constructing an initial reconstruction model; s2, acquiring a training data set; s3, importing the training data set into the initial reconstruction model, and carrying out training optimization on the initial reconstruction model; s4, obtaining original data to be reconstructed; s5, reconstructing the original data to be reconstructed by using the optimized reconstruction model to generate reconstructed multi-frequency electromagnetic data; through the method, the reconstruction module is trained and optimized, the model is endowed with strong robustness to abnormal values and measurement noise, and the problem of non-uniform distribution of measurement points in original data can be effectively solved. According to the technical scheme of the invention, after data in a dense region and data in a sparse region are processed by the two-stage reconstruction architecture, a regularized data set conforming to a preset interval can be generated, so that the space consistency and the data continuity are greatly improved, and high-quality basic data are provided for subsequent analysis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electromagnetic detection, and in particular to a multi-frequency electromagnetic data intelligent reconstruction method and system for metal cultural relic detection. BACKGROUND

[0002] In the field of metal cultural relic detection, the multi-frequency electromagnetic method has been widely used in archaeological exploration, cultural relic protection and underground heritage investigation due to its good response characteristics to underground metal targets, non-destructive, high detection efficiency and high sensitivity to shallow buried objects. The method transmits electromagnetic signals at multiple frequencies and receives the secondary induced field response generated by the underground medium to obtain original data reflecting the spatial distribution of geoelectric parameters (such as conductivity, magnetic permeability), and then infer the location, shape and burial depth of buried metal cultural relics.

[0003] However, in actual field detection, due to factors such as terrain undulation, obstacle blocking, human operation error or uneven detection device moving speed, the collected electromagnetic data often has problems of irregular spatial sampling, uneven density distribution and missing. Such non-uniform sampling not only destroys the spatial continuity of the data, but also significantly reduces the accuracy of subsequent imaging, inversion and interpretation, easily leading to image artifacts, resolution degradation and target misjudgment, which seriously restricts the application efficiency of multi-frequency electromagnetic technology in fine archaeological exploration.

[0004] In recent years, artificial intelligence technology has shown great potential in geophysical data processing. Deep learning models, especially fully connected neural networks (FCNN), have been widely used to extract deep features from complex electromagnetic responses and achieve data reconstruction due to their strong non-linear fitting ability. FCNN can learn the implicit mapping relationship between input coordinates and electromagnetic responses through multiple layers of non-linear transformation, and is suitable for processing non-uniform sampling and high-dimensional multi-frequency data. However, standard FCNN usually uses mean square error (MSE) as the loss function, which is essentially a least squares criterion, and is extremely sensitive to outliers and measurement noise. In actual detection data, due to environmental interference, instrument drift or human disturbance, there are usually a small number of outliers that have a significant impact. These abnormal samples will produce a large gradient under the MSE framework, causing the model parameters to deviate in the wrong direction, resulting in overfitting or unstable prediction, which seriously affects the robustness and reliability of the reconstruction results. SUMMARY

[0005] The purpose of the present application is to design a multi-frequency electromagnetic data intelligent reconstruction method and system for metal cultural relic detection to solve the above problems.

[0006] The application achieves the above-mentioned purpose through the following technical solutions:

[0007] The intelligent reconstruction method of multi-frequency electromagnetic data for metal cultural relic detection comprises:

[0008] S1, an initial reconstruction model is constructed, and the reconstruction model comprises an input layer, L full connection layers and an output layer from input to output;

[0009] S2, a training data set is obtained;

[0010] S3, the training data set is imported into the initial reconstruction model, and the initial reconstruction model is trained and optimized to obtain an optimized reconstruction model; specifically comprising:

[0011] S31, the training data is imported into the current reconstruction model to perform forward propagation, and a predicted value is obtained;

[0012] S32, the sample weight is dynamically adjusted according to the predicted value and a true value corresponding to the training data ;

[0013] S33, the sample weight is used to calculate a loss function Loss;

[0014] S34, the gradient of each parameter of the reconstruction model is calculated according to the loss function Loss by using the chain rule, and the network parameters of the reconstruction model are updated by applying the gradient descent method or the Adam optimizer to obtain an updated reconstruction model;

[0015] S35, it is judged whether the updated reconstruction model meets a stop condition, if yes, the updated reconstruction model is taken as the optimized reconstruction model, and S4 is entered; otherwise, S31 is returned to iteratively optimize the reconstruction model;

[0016] S4, original data to be reconstructed is obtained;

[0017] S5, the optimized reconstruction model is used to reconstruct the original data to be reconstructed to generate reconstructed multi-frequency electromagnetic data.

[0018] The intelligent reconstruction system of multi-frequency electromagnetic data for metal cultural relic detection is used to realize the intelligent reconstruction method of multi-frequency electromagnetic data for metal cultural relic detection, and comprises:

[0019] An acquisition end is used to acquire training data or original data to be reconstructed;

[0020] A preprocessing module is used to perform denoising and normalization processing on the training data or the original data to be reconstructed, extract spatial coordinates of each measuring point and a corresponding multi-frequency electromagnetic response vector, and construct an irregular sampling data set; ​

[0021] a reconstruction module, the reconstruction module being configured to reconstruct the multi-frequency electromagnetic data to generate reconstructed multi-frequency electromagnetic data;

[0022] a data storage and management module, the data storage and management module being configured to store the training data set and the original data;

[0023] a visualization and interactive interface module, the visualization and interactive interface module being configured to visualize the reconstructed multi-frequency electromagnetic data and to interact with an operator.

[0024] The method has the beneficial effect that the reconstruction module is trained and optimized, the model is endowed with strong robustness to abnormal values and measurement noise, and the problem of uneven distribution of measurement points in the original data can be effectively solved. Whether the data is in a dense or sparse area, after processing by the two-level reconstruction architecture, regularized data sets conforming to the preset interval can be generated, which greatly improves the spatial consistency and data continuity, and provides high-quality basic data for subsequent analysis. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 a schematic diagram of the method for intelligent reconstruction of multi-frequency electromagnetic data for metal cultural relic detection;

[0026] Figure 2 multi-frequency electromagnetic simulation data for the test set;

[0027] Figure 3 reconstructed multi-frequency electromagnetic simulation data for the method for intelligent reconstruction of multi-frequency electromagnetic data for metal cultural relic detection;

[0028] Figure 4 multi-frequency electromagnetic simulation data for the reconstruction of field data by the method. DETAILED DESCRIPTION

[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0030] Therefore, the detailed description of the embodiments of the present application provided in the drawings below is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0031] It should be noted that similar reference numerals and letters refer to like items throughout the accompanying drawings, and once an item is defined in one drawing, it is not necessary to further define and explain it in subsequent drawings.

[0032] In the description of the present application, it should be understood that the terms "upper", "lower", "inner", "outer", "left", "right", and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship commonly understood by those skilled in the art, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0033] In addition, the terms "first", "second", and the like are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0034] In the description of the present application, it should be noted that, unless otherwise specified and limited, the terms "provided", "connected" and the like should be broadly understood, for example, "connected" can be fixedly connected, or detachably connected, or integrally connected; can be mechanically connected, or electrically connected; can be directly connected, or indirectly connected through an intermediate medium, or the internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0035] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0036] As shown in Figure 1 The multi-frequency electromagnetic data intelligent reconstruction method for metal cultural relic detection comprises:

[0037] S1, an initial reconstruction model is constructed, and the reconstruction model comprises an input layer, L full connection layers and an output layer from input to output;

[0038] The input layer receives a data set of N measuring points, and the coordinate scalar corresponding to each measuring point is x i (unit: m), electromagnetic response data e=[e(f1), e(f2),..., e(f N )] ⊤ As label data, the data set composed of N measuring points is z=[x1, x 2， ..., x N ];

[0039] Hidden layer: contains L fully connected layers, each with a number of neurons that can be set to 128-512, and the activation function is ReLU (Rectified Linear Unit) function: ReLU(a) = max(0, a);

[0040] Output layer: output the predicted value of electromagnetic parameters on the target regular grid points , consistent with the dimension and e.

[0041] The forward propagation process of the reconstruction model is as follows: let the output vector of the l-1 layer be o l-1 ∈R l-1 , then the forward propagation process of the l layer is as follows: z l =W l o (l-1) +b l , o l =f l (z l ), wherein: z l ∈R nl represents the weighted input vector of the l layer, W l is the weight matrix of the l layer; b l is the bias vector; f l is the activation function (ReLU for the hidden layer, and usually a linear function for the output layer);

[0042] S2, obtaining a training data set;

[0043] S3, importing the initial reconstruction model into the training data set, training and optimizing it to obtain an optimized reconstruction model; specifically including:

[0044] S31, importing the training data into the current reconstruction model to perform forward propagation and obtain a predicted value;

[0045] S32, dynamically adjusting the sample weight according to the predicted value and the true value corresponding to the training data ; specifically:

[0046] The residual error of the i-th measuring point is a scalar difference, , is the predicted scalar value of the i-th measuring point, is the true value of the i-th measuring point, and the first loss function is calculated according to the predicted value and the true value corresponding to the training data, and is expressed as: ; wherein c is a constant, and the sample weight is dynamically adjusted according to the first loss function, and is expressed as: , the value of c is 4.685;

[0047] S33, using the sample weight , calculate the loss function Loss, denoted as: Loss = L MSE_Tukey + λL smooth , where L smooth is a smooth regularization term, λ is a regularization parameter, L MSE_Tukey is a weighted mean square error term, denoted as: ; the loss function Loss combines the mean square error term, the spatial smoothing regularization term, and the Tukey's Biweight loss function, which simultaneously realizes spatial consistency, noise resistance, and robustness to outliers;

[0048] S34, calculate the gradient of each parameter of the reconstruction model using the chain rule according to the loss function Loss, and then apply the gradient descent method or Adam optimizer to update the network parameters of the reconstruction model to obtain an updated reconstruction model;

[0049] S35, determine whether the updated reconstruction model meets the stopping condition, if yes, the updated reconstruction model is used as the optimized reconstruction model, and the process goes to S4; otherwise, the process returns to S31 to iteratively optimize the reconstruction model;

[0050] S4, obtain the original data to be reconstructed;

[0051] The original data before inputting the reconstruction model is processed for denoising, normalization, and coordinate alignment, the spatial coordinates of each measuring point and the corresponding multi-frequency electromagnetic response vector are extracted, an irregular sampling data set is constructed, and then input to the reconstruction model to train and optimize the reconstruction model or generate reconstructed multi-frequency electromagnetic data;

[0052] S5, use the optimized reconstruction model to reconstruct the original data to be reconstructed to generate reconstructed multi-frequency electromagnetic data; based on the trained reconstruction model, high-density interpolation prediction is performed on the target area according to the preset resolution and sampling interval, and continuous, smooth, and physically meaningful electromagnetic response curves or grid data are output for subsequent imaging, inversion, or artifact identification analysis.

[0053] The reconstruction module of the method uses a fully connected neural network FCNN to perform nonlinear feature extraction and mapping learning on the preprocessed irregular data set. During network training, the traditional mean square error MSE loss function is abandoned, and the Tukey's Biweight loss function is introduced, and the training weights of each sample are dynamically adjusted through the IRLS iterative reweighting mechanism. Specifically, in each training iteration, the sample weight is calculated according to the current model prediction residual: samples with smaller residuals are given higher weights, and potential outliers with larger residuals are automatically downweighted or even ignored, thereby effectively suppressing the interference of noise and outliers on model training and improving the robustness of the model. After training, the FCNN model is used to infer on the preset spatial grid points to generate a regularized multi-frequency electromagnetic data set with uniform spatial distribution and consistent frequency points.

[0054] The multi-frequency electromagnetic data intelligent reconstruction system for metal cultural relic detection is used to realize the multi-frequency electromagnetic data intelligent reconstruction method for metal cultural relic detection as described above, and is characterized by comprising:

[0055] An acquisition end is configured to acquire training data or original data to be reconstructed.

[0056] A preprocessing module is configured to perform denoising and normalization processing on the training data or the original data to be reconstructed, extract spatial coordinates of each measuring point and a corresponding multi-frequency electromagnetic response vector, and construct an irregularly sampled data set.

[0057] A reconstruction module is provided with a reconstruction model, and is configured to reconstruct and generate reconstructed multi-frequency electromagnetic data.

[0058] A data storage and management module is configured to store the training data set and the original data.

[0059] A visualization and interactive interface module is configured to visually display the reconstructed multi-frequency electromagnetic data and perform human-computer interaction with an operator.

[0060] The method has significant advantages in processing irregular sampling and noise interference of original data, which are specifically shown in the following aspects:

[0061] High-precision data regularization is achieved: by using a reconstruction model of a fully connected neural network (FCNN) combined with an improved robust optimization mechanism (based on Tukey's Biweight loss function and iterative reweighted least squares, IRLS), the method can effectively solve the problem of uneven distribution of measuring points in the original data. Whether the data is in a dense or sparse area, after processing by the two-level reconstruction architecture, a regularized data set conforming to the preset interval can be generated, greatly improving the spatial consistency and data continuity, and providing high-quality basic data for subsequent analysis.

[0062] Significant enhancement of noise resistance: the introduction of Tukey's Biweight loss function and IRLS mechanism gives the model strong robustness to outliers and measurement noise. During the training process, this mechanism can dynamically adjust the sample weight, automatically reduce the influence of large residual samples (i.e. potential outliers), and avoid the overfitting or prediction distortion problem of traditional methods in noisy data. This enables the model to maintain high stability and accuracy in complex noise environments, especially suitable for low signal-to-noise ratio archaeological detection environments.

[0063] Retaining nonlinear features and improving reconstruction accuracy: The reconstruction model has strong nonlinear mapping ability, which can extract deep features from complex multi-frequency electromagnetic responses and perform effective preliminary reconstruction. Combined with the robust optimization mechanism, the generalization performance of the model under non-uniform sampling conditions is further enhanced. Compared with traditional interpolation methods or standard deep learning models, this method not only improves the reconstruction accuracy, but also better captures the nonlinear features and complex spatial structure of underground metal cultural relics responses, ensuring the authenticity and reliability of the results.

[0064] Improving the quality of subsequent imaging: Regularized data has higher spatial resolution and consistency, which is more suitable for various inversion imaging algorithms (such as Tikhonov regularization, neural network inversion, etc.). By providing high-quality, noise-free input data, this method significantly improves the spatial positioning accuracy and shape restoration ability of metal cultural relics, laying a solid foundation for subsequent high-resolution imaging and accurate identification.

[0065] Strong practicality and easy integration and deployment: The software system constructed adopts modular design, supports batch processing and visualization operation, and is convenient for users to quickly import, process and analyze multi-frequency electromagnetic detection data. The system has good scalability and compatibility, can flexibly adapt to different archaeological exploration needs, and is easy to deploy and apply in actual field or data center, greatly improving work efficiency and operation convenience.

[0066] In summary, this method proposes an innovative method that combines deep learning and robust optimization mechanism, solving the key problems of irregular sampling and noise interference of multi-frequency electromagnetic data in metal cultural relic detection. This method not only realizes high-precision and high-reliability data regularization reconstruction, but also significantly improves the accuracy and efficiency of subsequent imaging and target identification, with broad application prospects and promotional value.

[0067] Example 1: Simulated data processing process:

[0068] (1) Construct a simulated multi-frequency electromagnetic data set, as shown in Figure 2 The data set contains spatially irregularly distributed measurement points and their electromagnetic response values at multiple frequencies; generate non-uniform sampling data by superimposing Gaussian noise and random missing mechanisms on a regular grid; randomly divide the data set into training set and test set;

[0069] (2) Using the training set to train the reconstruction model to learn the nonlinear mapping relationship from the irregular sampling input to the target regular grid output. The reconstruction model architecture includes an input layer, five fully connected hidden layers (the number of neurons is 128, 256, 512, 256, and 128, respectively), and an output layer. The activation function uses ReLU. The Tukey's Biweight loss function is introduced to enhance the robustness of the model and reduce the influence of outliers on training. IRLS is used to dynamically adjust the sample weight during training, giving low residual points higher weight, and automatically reducing the weight of large residual samples or even ignoring them, thereby effectively suppressing the interference of noise and outliers on model training and improving model robustness.

[0070] (3) Using the trained FCNN model to perform forward inference on the test set data to obtain preliminary estimated values of electromagnetic parameters on the target regular grid points.

[0071] (4) Based on the trained model, directly reconstructing the original data to generate a regularized multi-frequency electromagnetic data set with uniform spatial distribution and consistent frequency points, as shown in Figure 3

[0072] Example 2: Field data processing of a certain archaeological bronze ware exploration

[0073] This example uses multi-frequency electromagnetic field data collected on site at a certain archaeological site. Due to the use of handheld multi-frequency electromagnetic instruments during exploration, factors such as terrain undulations, uneven operation speed, and underground complex interference affect the collected data, resulting in discontinuity in spatial distribution and data missing. Therefore, the collected data has obvious irregular sampling characteristics. This data set truly reflects the common data quality challenges in actual archaeological exploration and is suitable for verifying the regularized reconstruction ability and robustness of the method in complex field environments. The present application improves the reconstruction model to regularize the reconstruction of actual data, with a reconstruction data interval of 0.2m. As can be seen from Figure 4 , the predicted data curve generated by the present application is smooth and continuous, seamlessly connecting the repaired part of the missing data with the original data, and exhibits high reliability and consistency.

[0074] The technical solution of the present application is not limited to the limitations of the above specific examples, and any technical variations made according to the technical solution of the present application fall within the scope of protection of the present application.​

Claims

1. A method for intelligent reconstruction of multi-frequency electromagnetic data for metal artifact detection, characterized in that, include: S1. Construct the initial reconstruction model, which consists of an input layer, L fully connected layers, and an output layer from input to output. S2. Obtain the training dataset; S3. Import the initial reconstruction model into the training dataset, train and optimize it to obtain the optimized reconstruction model; specifically including: S31. Import the training data into the current reconstruction model for forward propagation to obtain the predicted values; S32. Dynamically adjust sample weights based on the predicted values ​​and the corresponding true values ​​in the training data. ; S33, Utilizing Sample Weights Calculate the loss function Loss; S34. Calculate the gradient of each parameter of the reconstruction model using the chain rule based on the loss function Loss, and then apply gradient descent or Adam optimizer to update the network parameters of the reconstruction model to obtain the updated reconstruction model. S35. Determine whether the updated reconstruction model meets the stopping condition. If it does, the updated reconstruction model is used as the optimized reconstruction model and proceeds to S4; otherwise, return to S31 and iteratively optimize the reconstruction model. S4. Obtain the original data to be reconstructed; S5. Use the optimized reconstruction model to reconstruct the original data to be reconstructed and generate the reconstructed multi-frequency electromagnetic data.

2. The intelligent reconstruction method for multi-frequency electromagnetic data for metal artifact detection according to claim 1, characterized in that, In S32, a first loss function is calculated based on the predicted values ​​and the corresponding true values ​​in the training data, and the sample weights are dynamically adjusted based on the first loss function. .

3. The intelligent reconstruction method for multi-frequency electromagnetic data for metal artifact detection according to claim 2, characterized in that, First loss function Represented as: ;in, , Let i be the predicted scalar value for the i-th measurement point. Let c be the true value of the i-th measurement point, and c be a constant; dynamically adjust the sample weights. Represented as: .

4. The intelligent reconstruction method for multi-frequency electromagnetic data for metal artifact detection according to claim 3, characterized in that, The value of c is 4.

685.

5. The intelligent reconstruction method for multi-frequency electromagnetic data for metal artifact detection according to claim 1, characterized in that, The loss function Loss is expressed as: Loss = L MSE_Tukey +λL smooth , where L smooth For smoothing regularization, λ is the regularization parameter, and L MSE_Tukey The weighted mean square error term is expressed as: Where N is the total number of measurement points, and r is the residual of the i-th measurement point. i , These are the sample weights.

6. The intelligent reconstruction method for multi-frequency electromagnetic data for metal artifact detection according to claim 1, characterized in that, The original data before inputting into the reconstruction model is denoised and normalized. The spatial coordinates of each measurement point and its corresponding multi-frequency electromagnetic response vector are extracted to construct an irregular sampling dataset. This dataset is then input into the reconstruction model to train and optimize the model or to reconstruct and generate the reconstructed multi-frequency electromagnetic data.

7. A multi-frequency electromagnetic data intelligent reconstruction system for metal artifact detection, used to implement the multi-frequency electromagnetic data intelligent reconstruction method for metal artifact detection as described in any one of claims 1-6, characterized in that, include; Acquisition end; the acquisition end is used to acquire training data or raw data to be reconstructed; Preprocessing module; The preprocessing module is used to denoise and normalize the training data or the original data to be reconstructed, extract the spatial coordinates of each measurement point and its corresponding multi-frequency electromagnetic response vector, and construct an irregular sampling dataset. Rebuild module; The reconstruction module is equipped with a reconstruction model and is used to reconstruct and generate reconstructed multi-frequency electromagnetic data. Data storage and management module; The data storage and management module is used to store training datasets and raw data; Visualization and interactive interface module; It is used to visualize the reconstructed multi-frequency electromagnetic data and to enable human-computer interaction with the operator.

Images