Ground-air electromagnetic detection data correction method and system

Abstract

The invention discloses a ground-air electromagnetic detection data correction method and system, and relates to the technical field of geophysical exploration. The method comprises the following specific steps: acquiring ground-air electromagnetic detection data; carrying out wavelet transform decomposition on the ground-air electromagnetic detection data, carrying out data reconstruction according to obtained decomposition parameters, and removing sky electricity noise to determine effective data segments; removing motion noise of the effective data segments based on a motion noise separation model, inhibiting attitude noise based on a BP neural network model, and removing low-frequency baseline interference through a Savitzky-Golay smoothing filtering method to obtain preprocessed electromagnetic data; and performing quality evaluation on the preprocessed electromagnetic data after preprocessing is completed. According to the method, multi-source data fusion and deep learning technologies are integrated, and the data preprocessing precision and correction efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of geophysical exploration, and particularly to a method and system for correcting geoelectric and electromagnetic detection data. Background Art

[0002] The geoelectric and electromagnetic detection technology is an electromagnetic exploration method that combines ground and aerial measurements, and is widely used in fields such as underground resource exploration and geological disaster monitoring. This method is mainly based on the principle of electromagnetic induction, using a high-power artificial electromagnetic source on the ground as the excitation field source, and realizing the rapid exploration of the earth's electrical structure by collecting magnetic field signals in the air. However, due to the influence of factors such as flight altitude, terrain undulation, and electromagnetic interference, there are often problems such as noise and drift in the original data, and correction is required to improve data accuracy. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for correcting geoelectric and electromagnetic detection data, which integrates multi-source data fusion and deep learning technologies, and significantly improves the data preprocessing accuracy and correction efficiency to solve the problems raised in the background art.

[0004] To achieve the above purpose, the present invention provides the following solutions: On the one hand, a method for correcting geoelectric and electromagnetic detection data is provided, and the specific steps include the following:

[0005] Obtain geoelectric and electromagnetic detection data;

[0006] Perform wavelet transform decomposition on the geoelectric and electromagnetic detection data, perform data reconstruction according to the obtained decomposition parameters, and remove telluric noise to determine the effective data segment;

[0007] Remove motion noise from the effective data segment based on the motion noise separation model, suppress attitude noise based on the BP neural network model, and remove low-frequency baseline interference through the Savitzky-Golay smoothing filtering method to obtain preprocessed electromagnetic data;

[0008] Perform quality assessment on the preprocessed electromagnetic data after preprocessing is completed.

[0009] Preferably, the geoelectric and electromagnetic detection data includes spectral data from two detection modes of time domain and frequency domain.

[0010] Preferably, the specific steps for removing motion noise based on the motion noise separation model are as follows:

[0011] Obtain time-domain secondary field sample data, its sampling frequency, and sample data period;

[0012] Extend the non-full-time sample data included in the secondary field sample data to full duration to restore continuity;

[0013] Construct an overdetermined system of equations using late-stage data and data from the pulse front end, and fit the noise with Fourier series;

[0014] Solve the overdetermined system of equations through the least squares inversion algorithm to obtain the Fourier series coefficients of the motion noise, thereby reconstructing the full-duration noise signal;

[0015] Remove the reconstructed full-duration noise signal obtained from the secondary field sample data.

[0016] Preferably, the specific steps for suppressing attitude noise based on the BP neural network model are as follows:

[0017] Construct a sample data set: Record the electromagnetic response data of the detection device at different attitude angles, generate noise profile samples according to the measured noise type, and superimpose the noise profile samples with the pure signal to form noisy data samples as inputs;

[0018] Preprocess the sample data set: Normalize the noisy data samples and divide them into a training set and a test set;

[0019] Construct a first BP neural network and a second BP neural network: Both the first BP neural network and the second BP neural network include an input layer, a hidden layer, and an output layer. A double hidden layer structure is adopted in the hidden layer, with 128 neurons in each layer;

[0020] Train the first BP neural network and the second BP neural network: Select the tanh function for the hidden layer training and the purelin function for the output layer activation function. Input the noisy data samples and the pure signal samples, as well as the sample classification labels of the noisy data samples and the pure signal samples, into the first BP neural network to train and obtain a BP signal-to-noise identification model. The input data of the BP signal-to-noise identification model is the signal, and the output data is the classification label of the signal; then input the noisy data samples and the noise profile samples into the second BP neural network to train and obtain a BP denoising model. The input data of the BP denoising model is the noise signal, and the output data is the noise profile of the noise signal;

[0021] Suppress attitude noise: Uniformly segment the measured electromagnetic data, input it into the BP signal-to-noise identification model to screen out the segments affected by attitude noise, input the interfering segments into the BP denoising model, output the predicted noise profile, and subtract the predicted noise from the interfering data to obtain the corrected signal.

[0022] Preferably, the specific steps for quality assessment of the preprocessed electromagnetic data are as follows:

[0023] Randomly select an exploration flight line, and use the Haar wavelet strategy to extract the spike data segments in the data of this point in a point-by-point processing manner;

[0024] Calculate the average energy ratio of the peak data segment and the remaining data segment. After the average energy ratio of all data points on the entire route is calculated, evaluate the data quality of the route according to the average energy ratio threshold, classify and store the data that meets and does not meet the quality standard, and give the data qualification rate of the entire route;

[0025] After the quality evaluation of the data on all routes is completed, the average energy ratio distribution image of all data points in the entire survey area is drawn to display the overall data quality of the survey area.

[0026] Preferably, it also includes using an energy center correction method to extract spectrum information of ground-to-space frequency domain electromagnetic detection data.

[0027] On the other hand, a ground-to-air electromagnetic detection data correction system is provided, including a data acquisition module, a wavelet denoising module, a preprocessing module, and a quality assessment module; wherein:

[0028] The data acquisition module is used to acquire ground-to-air electromagnetic detection data;

[0029] The wavelet denoising module is used to perform wavelet transform decomposition on the ground-to-air electromagnetic detection data, reconstruct the data according to the decomposition parameters obtained, and remove the sky noise to determine the valid data segment;

[0030] The preprocessing module is used to remove motion noise from the valid data segment based on a motion noise separation model, suppress posture noise based on a BP neural network model, and remove low-frequency baseline interference through a Savitzky-Golay smoothing filter method to obtain preprocessed electromagnetic data;

[0031] The quality assessment module is used to perform quality assessment on the pre-processed electromagnetic data.

[0032] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: using a BP neural network to perform attitude correction on ground-to-air electromagnetic detection data can simplify the process of traditional attitude correction, improve data processing efficiency, and have high generalization ability. It has important application value in the field of rapid detection at large depths in complex terrain areas, and helps to explore deep mineral resources in complex terrain areas; after a series of denoising and correction operations are performed on the ground-to-air electromagnetic detection data, the average energy ratio of the peak data segment and the remaining data segment is used to evaluate the quality of the denoised data point by data point, and some data that does not meet the quality standards are eliminated before subsequent data interpretation and imaging, which can not only reduce the workload of data interpreters, but also obtain more accurate electromagnetic exploration results. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0034] Figure 1 is the flowchart of the method of the present invention;

[0035] Figure 2 is the system structure diagram of the present invention. Detailed implementation manners

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0037] The object of the present invention is to provide a method for correcting geoelectric magnetic detection data, as Figure 1 shown, the specific steps are as follows:

[0038] S1. Obtain geoelectric magnetic detection data;

[0039] S2. Perform wavelet transform decomposition on the geoelectric magnetic detection data, perform data reconstruction according to the obtained decomposition parameters, and remove atmospheric electricity noise to determine the effective data segment;

[0040] S3. Remove motion noise from the effective data segment based on the motion noise separation model, suppress attitude noise based on the BP neural network model, and remove low-frequency baseline interference through the Savitzky-Golay smoothing filtering method to obtain preprocessed electromagnetic data;

[0041] S4. After the preprocessing is completed, perform quality evaluation on the preprocessed electromagnetic data.

[0042] Ground-to-air electromagnetic detection is a geophysical exploration technology with artificial sources that can obtain electrical structural information of underground media by remote sensing. This technology establishes an active electromagnetic field by arranging a high-power electrical excitation source on the ground, and uses a signal acquisition system carried by a flying platform (such as a helicopter or a drone) to detect secondary fields that carry electrical information of underground structures. In the time domain (referred to as the time domain) ground-to-air electromagnetic working mode, a bipolar pulse with a fixed frequency is used as the excitation pulse signal of the ground excitation source, and the induced electromotive force captured by the air signal acquisition system is a time domain attenuation signal carrying shallow underground information. In the frequency domain (referred to as the frequency domain) ground-to-air electromagnetic working mode, a multi-frequency pseudo-random pulse is used as the excitation pulse of the ground excitation source, and the induced electromotive force captured by the air signal acquisition system is a frequency domain attenuation signal carrying deep underground information. In this embodiment, the ground-to-air electromagnetic detection data includes spectrum data from both time domain and frequency domain detection modes.

[0043] Furthermore, the specific steps of removing motion noise based on the motion noise separation model are:

[0044] S311, obtaining time domain secondary field sample data, and its sampling frequency and sample data period;

[0045] S312, extending the non-full-time sample data included in the secondary field sample data to the full-time length to restore continuity;

[0046] S313, using the late data and the data of the pulse front end to construct an overdetermined system of equations, and fitting the noise with Fourier series;

[0047] S314, solving the overdetermined equations by a least squares inversion algorithm to obtain the Fourier series coefficients of the motion noise, thereby reconstructing the full-duration noise signal;

[0048] S315 , removing the obtained reconstructed full-duration noise signal from the secondary field sample data.

[0049] Furthermore, the specific steps of suppressing posture noise based on the BP neural network model are as follows:

[0050] S321, constructing a sample data set: recording the electromagnetic response data of the detection device at different posture angles, generating noise profile samples according to the measured noise type, and superimposing the noise profile samples with the pure signal to form noisy data samples as input;

[0051] S322, sample data set preprocessing: normalize the noisy data samples and divide them into a training set and a test set;

[0052] S323. Construct the first BP neural network and the second BP neural network: Both the first BP neural network and the second BP neural network include an input layer, a hidden layer, and an output layer. A double hidden layer structure is adopted in the hidden layer, with 128 neurons in each layer.

[0053] S324. Train the first BP neural network and the second BP neural network: Select the tanh function for the training hidden layer and the purelin function for the output layer activation function. Input the noisy data samples and the pure signal samples, as well as the sample classification labels of the noisy data samples and the pure signal samples, into the first BP neural network to train the BP signal-to-noise identification model. The input data of the BP signal-to-noise identification model is the signal, and the output data is the classification label of the signal. Then, input the noisy data samples and the noise profile samples into the second BP neural network to train the BP denoising model. The input data of the BP denoising model is the noise signal, and the output data is the noise profile of the noise signal.

[0054] S325. Suppress the attitude noise: Segment the measured electromagnetic data evenly, input it into the BP signal-to-noise identification model to screen out the segments affected by the attitude noise, input the interfering segments into the BP denoising model, output the predicted noise profile, and subtract the predicted noise from the interfering data to obtain the corrected signal.

[0055] Furthermore, in the signal-to-noise identification model of step S324, the input is: the noisy data segment and the corresponding pure signal segment; the output is: the label (binary classification) indicating whether the data segment is affected by the attitude noise; the objective is: to distinguish the interfered and non-interfered data segments to provide a screening basis for subsequent denoising. The input of the BP denoising model is: the noisy data segment and its corresponding noise profile sample; the output is: the predicted noise component; the objective is: to optimize the network weights through backpropagation to minimize the error between the predicted noise and the true noise.

[0056] Further, the specific steps for quality assessment of the preprocessed electromagnetic data are as follows:

[0057] S41. Randomly select an exploration route and use the Haar wavelet strategy to extract the spike data segments in the data at this point in a point-by-point processing manner.

[0058] S42. Calculate the average energy ratio of the spike data segments and the remaining data segments. After calculating the average energy ratios of all data points on the entire route, evaluate the data quality of this route according to the average energy ratio threshold, classify and store the data that meets and does not meet the quality standards, and give the data qualification rate of the entire route.

[0059] S43. After the quality assessment of the data on all routes is completed, draw the distribution image of the average energy ratio of all data points in the entire survey area to display the overall data quality of the survey area.

[0060] Further, it also includes using the energy center correction method to extract the spectral information of the ground-air frequency-domain electromagnetic detection data.

[0061] On the other hand, a ground-air electromagnetic detection data correction system is provided, as Figure 2 shown, including a data acquisition module, a wavelet denoising module, a preprocessing module, and a quality assessment module; among them,

[0062] The data acquisition module is used to acquire the ground-air electromagnetic detection data;

[0063] The wavelet denoising module is used to perform wavelet transform decomposition on the ground-air electromagnetic detection data, reconstruct the data according to the obtained decomposition parameters, and remove the atmospheric electricity noise to determine the effective data segment;

[0064] The preprocessing module is used to remove the motion noise from the effective data segment based on the motion noise separation model, suppress the attitude noise based on the BP neural network model, and remove the low-frequency baseline interference through the Savitzky-Golay smoothing filtering method to obtain the preprocessed electromagnetic data;

[0065] The quality assessment module is used to perform quality assessment on the preprocessed electromagnetic data.

[0066] In this article, specific examples are used to elaborate on the principles and implementation methods of the present invention. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for correcting ground-to-air electromagnetic detection data, characterized in that: The specific steps include the following: Acquire ground-to-air electromagnetic detection data; Performing wavelet transform decomposition on the ground-to-air electromagnetic detection data, reconstructing the data according to the decomposition parameters obtained, removing the sky noise and determining the effective data segment; The effective data segment is subjected to motion noise removal based on a motion noise separation model, posture noise is suppressed based on a BP neural network model, and low-frequency baseline interference is removed by a Savitzky-Golay smoothing filter method to obtain preprocessed electromagnetic data; After the preprocessing is completed, the quality of the preprocessed electromagnetic data is evaluated.

2. A method for correcting ground-to-air electromagnetic detection data according to claim 1, characterized in that: The ground-to-air electromagnetic detection data includes spectrum data from two detection modes: time domain and frequency domain.

3. The method for correcting ground-to-air electromagnetic detection data according to claim 1, characterized in that: The specific steps of removing motion noise based on the motion noise separation model are: Obtaining time domain secondary field sample data, and its sampling frequency and sample data period; Extending the non-full-time sample data included in the secondary field sample data to the full-time length to restore continuity; The overdetermined equations are constructed using the late data and the data from the pulse front end, and the noise is fitted using the Fourier series; Solving the overdetermined equations by a least squares inversion algorithm to obtain the Fourier series coefficients of the motion noise, thereby reconstructing the full-duration noise signal; The obtained reconstructed full-time noise signal is removed from the secondary field sample data.

4. The method for correcting ground-to-air electromagnetic detection data according to claim 1, characterized in that: The specific steps of suppressing posture noise based on the BP neural network model are: Construct a sample data set: record the electromagnetic response data of the detection equipment at different posture angles, generate noise profile samples according to the measured noise type, and superimpose the noise profile samples with the pure signal to form noisy data samples as input; Sample data set preprocessing: normalize the noisy data samples and divide them into training set and test set; Constructing a first BP neural network and a second BP neural network: the first BP neural network and the second BP neural network both include an input layer, a hidden layer, and an output layer, and the hidden layer adopts a double hidden layer structure, with 128 neurons in each layer; Training the first BP neural network and the second BP neural network: selecting the tanh function for the hidden layer training and the purelin function for the output layer activation function, inputting the sample classification labels of the noisy data samples and the pure signal samples, and the noisy data samples and the pure signal samples into the first BP neural network, and training to obtain a BP signal-noise identification model, wherein the input data of the BP signal-noise identification model is the signal, and the output data is the classification label of the signal; Then the noisy data samples and the noise profile samples are input into a second BP neural network to obtain a BP denoising model through training, wherein the input data of the BP denoising model is the noise signal and the output data is the noise profile of the noise signal; Suppressing attitude noise: Evenly segment the measured electromagnetic data, input the BP signal-noise identification model to filter out the sections affected by attitude noise, input the interference sections into the BP denoising model, output the predicted noise profile, subtract the predicted noise from the interference data, and obtain the corrected signal.

5. The method for correcting ground-to-air electromagnetic detection data according to claim 1, characterized in that: The specific steps of performing quality assessment on the pre-processed electromagnetic data are as follows: A survey route is randomly selected, and the peak data segment in the data of the point is extracted using the Haar wavelet strategy in a point-by-point processing manner; Calculate the average energy ratio of the peak data segment and the remaining data segment. After the average energy ratio of all data points on the entire route is calculated, evaluate the data quality of the route according to the average energy ratio threshold, classify and store the data that meets and does not meet the quality standard, and give the data qualification rate of the entire route; After the quality evaluation of the data on all routes is completed, the average energy ratio distribution image of all data points in the entire survey area is drawn to display the overall data quality of the survey area.

6. The method for correcting ground-to-air electromagnetic detection data according to claim 1, characterized in that: It also includes using the energy center correction method to extract the spectrum information of ground-to-space frequency domain electromagnetic detection data.

7. A ground-to-air electromagnetic detection data correction system, characterized in that: It includes data acquisition module, wavelet denoising module, preprocessing module and quality assessment module; among them, The data acquisition module is used to acquire ground-to-air electromagnetic detection data; The wavelet denoising module is used to perform wavelet transform decomposition on the ground-to-air electromagnetic detection data, reconstruct the data according to the decomposition parameters obtained, and remove the sky noise to determine the valid data segment; The preprocessing module is used to remove motion noise from the valid data segment based on a motion noise separation model, suppress posture noise based on a BP neural network model, and remove low-frequency baseline interference through a Savitzky-Golay smoothing filter method to obtain preprocessed electromagnetic data; The quality assessment module is used to perform quality assessment on the pre-processed electromagnetic data.

Images