High-precision magnetic measurement data processing method for aeromagnetic measurement

Abstract

The invention belongs to the field of data processing, and particularly relates to a high-precision magnetic measurement data processing method for aeromagnetic measurement. The method comprises the following steps: firstly, carrying out adaptive noise suppression on acquired aeromagnetic measurement data by using a convolutional neural network through an adaptive noise suppression network; secondly, de-noising is carried out by adopting the combination of enhanced wavelet packet transformation and a particle swarm optimization algorithm; and finally, dynamically correcting the magnetic measurement data through a deep reinforcement learning method. According to the method, environmental noise and equipment interference noise are effectively removed through multiple technical means, the denoising effect and the signal quality are improved, correction parameters can be adjusted according to real-time data feedback, the correction precision is improved, and finally high precision of magnetic measurement data is guaranteed.

Description

Technical Field

[0001] The invention belongs to the field of data processing, and in particular relates to a high-precision magnetic survey data processing method for aeromagnetic survey. Background Art

[0002] With the increasing demand for geological exploration and resource survey, aeromagnetic measurement has gradually become a widely used geophysical exploration method. By recording the spatial distribution characteristics of the geomagnetic field, aeromagnetic measurement provides important data for geological structure, mineral resource distribution, etc. However, since aeromagnetic measurement is usually carried out in complex terrain and electromagnetic environment, the data will be affected by various factors such as environmental noise, electromagnetic interference, equipment error, etc., which will reduce the measurement accuracy and increase the difficulty of data post-processing. Existing data processing methods mainly rely on signal filtering and noise suppression methods, such as low-pass filtering and wavelet decomposition. However, these methods are often difficult to achieve ideal results in the face of complex and changeable noise environments. In particular, when the measurement environment noise and equipment errors are nonlinearly superimposed, the error correction and noise suppression capabilities of the existing methods are insufficient, resulting in the measurement accuracy failing to meet high-precision requirements. In addition, the traditional method lacks an effective dynamic adjustment mechanism and cannot adapt to the noise characteristics and equipment deviations under different environmental conditions, thereby limiting the reliability and stability of aeromagnetic data processing. Summary of the invention

[0003] In view of the technical problems existing in the above-mentioned background technology, the present invention proposes a new high-precision magnetic measurement data processing method to cope with the noise and error problems in complex environments, and further improves the accuracy and efficiency of data processing through adaptive and intelligent means.

[0004] In order to achieve the above object, the technical solution adopted by the present invention comprises the following steps:

[0005] S1. The first step is data collection and preprocessing. Through the adaptive noise suppression network, the convolutional neural network is used to perform adaptive noise suppression on the collected aeromagnetic measurement data to remove the noise caused by environmental noise, sensor error and electromagnetic interference. The adaptive denoising formula is: where f θ (X) is the output of the convolutional neural network, is the denoised signal, X is the original input data;

[0006] S2, then the enhanced wavelet packet transform and particle swarm optimization algorithm are used to jointly remove noise. The specific formula is:

[0007] Y k =f(EWT(S k ,θ PSO )), where S kis the kth input signal, EWT(S k ,θ PSO ) represents the signal S transformed by enhanced wavelet packet transform k Perform wavelet packet decomposition, f function is the signal reconstruction function, Y k is the final signal;

[0008] S3. Finally, the magnetic survey data is dynamically corrected by a deep reinforcement learning method to adjust the deviation in data collection. The deep reinforcement learning method dynamically corrects by designing a reward and punishment mechanism and automatically adjusts the correction parameters to maximize the measurement accuracy.

[0009] The processing steps of the enhanced wavelet packet transform in step S2 are:

[0010] S21, first use enhanced wavelet packet transform to denoise the signal Perform multi-scale decomposition to obtain sub-signals in several frequency bands: Where W i Denoised signal The decomposition result on the i-th frequency subband, n is the number of decomposed frequency bands;

[0011] S22, in each sub-band, introduce an adaptive threshold τ i , and use the particle swarm algorithm to optimize the threshold selection. For the i-th subband signal W i , its adaptive denoising formula is: where β i The adaptive adjustment coefficient obtained by particle swarm algorithm optimization;

[0012] S23, the particle swarm algorithm searches for the best combination of threshold and adjustment coefficient, and the fitness function is defined as: Where S i , are the target signal and denoised signal of the i-th sample, is the jth frequency band amplitude of the denoised signal and the target signal in the frequency domain;

[0013] S24, after finding the optimal parameters, reconstruct each denoised subband to restore the denoised signal Y k :

[0014] Preferably, the step S1 is implemented by first using a convolutional neural network to use a large amount of noisy data and corresponding clean data in the training phase, and ensuring that the network can minimize the error between the output and the target signal through the loss function; then performing adaptive denoising, the adaptive denoising formula is: where f θ (X) is the output of the convolutional neural network, which represents the noise component predicted by the network. is the denoised signal and X is the original input data.

[0015] As a preference, the loss function is defined as: total =λ1Γ MSE +λ2Γ smooth +λ3Γ noise , where λ1, λ2, λ3 are the weight coefficients of each item of the loss function;

[0016] where Γ MSE is the mean square error loss, which measures the similarity between the denoised signal and the target signal. Where S i , Represent the i-th sample value of the target signal and the denoised signal respectively, and n is the total number of samples;

[0017] where Γ smooth is the frequency domain smoothness loss, which measures the frequency domain smoothness of the denoised signal. where F j represents the amplitude of the jth frequency band of the denoised signal in the frequency domain, and m is the total number of frequency bands;

[0018] where Γ noise In order to constrain the residual noise in the denoised signal and reduce the residual noise component in the network output, in represents the residual noise after denoising, and p is the exponential parameter that controls the penalty for the residual noise.

[0019] Preferably, the specific implementation method of evaluating the correction strategy by designing a reward and punishment mechanism in step S3 is:

[0020] S31. First, the reward and punishment mechanism is divided into short-term rewards and long-term rewards. The short-term reward function is: short-term = -α·|E current -E expected |, the long-term reward function is: Where α is the short-term error weight, β is the long-term error weight, E current is the current correction error, E expected is the ideal error, T is the length of the historical time window, and E trend,t Represents the trend error within the time window;

[0021] S32, dynamically adjust the weight coefficients of short-term and long-term rewards and penalties according to the complexity of the current data, α adaptive =α·(1+γ·E complexity ), β adaptive =β·(1-γ·E complexity ), where E complexityrepresents the complexity of the data, and γ is the adjustment factor;

[0022] S33. Finally, the reward and punishment weights are designed for different error sources. The error source reward and punishment formula is: where ω k is the adaptive weight of the k-th error source, The correction error value of each error source is assigned different weights to enhance the sensitivity of the correction strategy to the key error sources.

[0023] Compared with the prior art, the advantages and positive effects of the present invention are that, firstly, the present invention can effectively remove environmental noise and equipment interference noise through adaptive noise suppression network and deep learning technology, significantly improve the denoising effect, and make the signal-to-noise ratio more significantly improved. Secondly, the present invention introduces a joint denoising method of enhanced wavelet packet transform and particle swarm optimization algorithm, performs multi-scale decomposition of the signal, combines adaptive threshold optimization, effectively processes the noise of complex frequency components, and ensures higher signal quality. In addition, through the dynamic correction strategy of deep reinforcement learning, the present invention can adjust the correction parameters according to real-time data feedback, significantly improve the correction accuracy, and achieve more accurate data correction in a multi-noise source environment. DETAILED DESCRIPTION

[0024] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0025] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0026] Embodiment, in order to achieve high-precision denoising and dynamic correction of aeromagnetic measurement data, a multi-stage optimized data processing method is proposed, which can significantly reduce complex noise, system errors and environmental interference through adaptive noise suppression, enhanced wavelet packet transform and deep reinforcement learning correction strategy, achieve accurate data correction, and improve the accuracy and reliability of measurement results. First, in order to reduce environmental noise and sensor errors in the data and improve the quality of the original data, the present invention first performs data acquisition and preprocessing, and processes the collected aeromagnetic measurement data through an adaptive noise suppression network. The specific steps are as follows: A convolutional neural network is used to perform adaptive denoising on the data. The network is first trained using a large amount of noisy data and corresponding noise-free data in the training phase, and the loss function is used to ensure that the network can minimize the error between the output and the target signal. The loss function includes mean square error, frequency domain smoothness loss and noise penalty loss, and the formula is as follows: Γtotal =λ1Γ MSE +λ2Γ smooth +λ3Γ noise , where λ1, λ2, λ3 are the weight coefficients of each item in the loss function; where Γ MSE is the mean square error loss, which measures the similarity between the denoised signal and the target signal. Where S i , Represent the i-th sample value of the target signal and the denoised signal respectively, and n is the total number of samples; where Γ smooth is the frequency domain smoothness loss, which measures the frequency domain smoothness of the denoised signal. where F j represents the amplitude of the jth frequency band of the denoised signal in the frequency domain, m is the total number of frequency bands; where Γ noise In order to constrain the residual noise in the denoised signal and reduce the residual noise component in the network output, in represents the residual noise after denoising, and p is the exponential parameter that controls the penalty for the residual noise.

[0027] In order to further remove the residual noise and improve the stability of the signal, the enhanced wavelet packet transform and particle swarm optimization algorithm are used to jointly denoise. This step can refine the noise suppression in the frequency domain. The steps are as follows: First, the denoised signal is denoised using the enhanced wavelet packet transform. Perform multi-scale decomposition to obtain sub-signals in several frequency bands: Where W i Denoised signal The decomposition result on the i-th frequency subband, n is the number of decomposed frequency bands; in each subband, an adaptive threshold τ is introduced i , and use the particle swarm algorithm to optimize the threshold selection. For the i-th subband signal W i , its adaptive denoising formula is: where β i is the adaptive adjustment coefficient optimized by the particle swarm algorithm; the particle swarm algorithm seeks the best combination of threshold and adjustment coefficient, and the adaptability function is defined as: Where S i , are the target signal and denoised signal of the i-th sample, is the jth frequency band amplitude of the denoised signal and the target signal in the frequency domain; finally, after finding the optimal parameters, the denoised sub-bands are reconstructed to restore the denoised signal Y k :

[0028] Finally, in order to further improve the correction accuracy of the data, the deep reinforcement learning method is used to dynamically correct the magnetic measurement data. This method evaluates the correction strategy by designing a reward and punishment mechanism, automatically adjusts the correction parameters, and ensures the optimization of measurement accuracy. The steps are as follows: First, the reward and punishment mechanism is divided into short-term rewards and long-term rewards. The short-term reward function is: short-term = -α·|E current -E expected |, the long-term reward function is: Where α is the short-term error weight, β is the long-term error weight, E current is the current correction error, E expected is the ideal error, T is the length of the historical time window, and E trend,t Represents the trend error within the time window; then dynamically adjusts the weight coefficients of short-term and long-term rewards and penalties according to the complexity of the current data, α adaptive =α·(1+γ·E complexity ), β adaptive =β·(1-γ·E complexity ), where E complexity represents the complexity of the data, and γ is the adjustment factor; finally, the reward and punishment weights are designed for different error sources, and the error source reward and punishment formula is: where ω k is the adaptive weight of the k-th error source, The correction error value of each error source is assigned different weights to enhance the sensitivity of the correction strategy to the key error sources.

[0029] In order to verify the effectiveness of the method of the present invention, a series of comparative experiments were designed to evaluate the performance of the method of the present invention in terms of denoising effect and correction accuracy. The experiment set reasonable noise levels and correction accuracy indicators. Table 1 is the evaluation of the denoising effect. In terms of denoising effect, the method of the present invention is significantly better than the traditional method. Compared with the low-pass filtering method and the wavelet decomposition denoising method, the signal-to-noise ratio of the method of the present invention is improved by 6.0dB, and the mean square error is significantly reduced to 0.07, indicating that the method has obvious advantages in suppressing noise and improving the signal-to-noise ratio.

[0030] Table 1: Denoising effect evaluation table

[0031]

[0032] In addition, the correction effect is evaluated by comparing the mean error before and after correction and the percentage of improvement in correction accuracy. The following Table 2 shows the experimental results. The results show that the method of the present invention is also superior to the traditional method in terms of correction accuracy. After using the method of the present invention, the mean error is reduced from 0.40mT to 0.125mT, and the correction accuracy is improved by 62.5%, which is significantly higher than the low-pass filtering and adaptive filtering methods, indicating that the method has extremely high accuracy in data correction.

[0033] Table 2: Comparison of calibration accuracy

[0034]

[0035] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A high-precision magnetic data processing method for aeromagnetic measurement, characterized in that: The following steps are involved: S1. The first step is data collection and preprocessing. Through the adaptive noise suppression network, the convolutional neural network is used to perform adaptive noise suppression on the collected aeromagnetic measurement data to remove the noise caused by environmental noise, sensor error and electromagnetic interference. The adaptive denoising formula is: where f θ (X) is the output of the convolutional neural network, is the denoised signal, X is the original input data; S2, then the enhanced wavelet packet transform and particle swarm optimization algorithm are used to jointly remove noise. The specific formula is: Y k =f(EWT(S k ,θ PSO )), where S k is the kth input signal, EWT(S k ,θ PSO ) represents the signal S transformed by enhanced wavelet packet transform k Perform wavelet packet decomposition, f function is the signal reconstruction function, Y k is the final signal; S3. Finally, the magnetic survey data is dynamically corrected by a deep reinforcement learning method to adjust the deviation in data collection. The deep reinforcement learning method dynamically corrects by designing a reward and punishment mechanism and automatically adjusts the correction parameters to maximize the measurement accuracy. The processing steps of the enhanced wavelet packet transform in step S2 are: S21, first use enhanced wavelet packet transform to denoise the signal Perform multi-scale decomposition to obtain sub-signals in several frequency bands: Where W i Denoised signal The decomposition result on the i-th frequency subband, n is the number of decomposed frequency bands; S22, in each sub-band, introduce an adaptive threshold τ i , and use the particle swarm algorithm to optimize the threshold selection. For the i-th subband signal W i , its adaptive denoising formula is: where β i The adaptive adjustment coefficient obtained by particle swarm algorithm optimization; S23, the particle swarm algorithm searches for the best combination of threshold and adjustment coefficient, and the fitness function is defined as: Where S i , are the target signal and denoised signal of the i-th sample, is the jth frequency band amplitude of the denoised signal and the target signal in the frequency domain; S24, after finding the optimal parameters, reconstruct each denoised subband to restore the denoised signal Y k :

2. A high-precision magnetic survey data processing method for aeromagnetic survey according to claim 1, characterized in that: The implementation of step S1 is to use a convolutional neural network to first use a large amount of noisy data and corresponding clean data in the training phase, and ensure that the network can minimize the error between the output and the target signal through the loss function; then perform adaptive denoising, and the adaptive denoising formula is: where f θ (X) is the output of the convolutional neural network, which represents the noise component predicted by the network. is the denoised signal and X is the original input data.

3. The high-precision magnetic data processing method for aeromagnetic measurement according to claim 2, characterized in that: The loss function is defined as: total =λ1Γ MSE +λ2Γ smooth +λ3Γ noise , where λ1, λ2, λ3 are the weight coefficients of each item of the loss function; where Γ MSE is the mean square error loss, which measures the similarity between the denoised signal and the target signal. Where S i , Represent the i-th sample value of the target signal and the denoised signal respectively, and n is the total number of samples; where Γ smooth is the frequency domain smoothness loss, which measures the frequency domain smoothness of the denoised signal. where F j represents the amplitude of the jth frequency band of the denoised signal in the frequency domain, and m is the total number of frequency bands; where Γ noise In order to constrain the residual noise in the denoised signal and reduce the residual noise component in the network output, in represents the residual noise after denoising, and p is the exponential parameter that controls the penalty for the residual noise.

4. The high-precision magnetic data processing method for aeromagnetic measurement according to claim 1, characterized in that: The specific implementation method of evaluating the correction strategy by designing a reward and punishment mechanism in step S3 is: S31. First, the reward and punishment mechanism is divided into short-term rewards and long-term rewards. The short-term reward function is: short-term = -α·|E current -E expected |, the long-term reward function is: Where α is the short-term error weight, β is the long-term error weight, E current is the current correction error, E expected is the ideal error, T is the length of the historical time window, and E trend,t Represents the trend error within the time window; S32, dynamically adjust the weight coefficients of short-term and long-term rewards and penalties according to the complexity of the current data, α adaptive =α·(1+γ·E complexity ), β adaptive =β·(1-γ·E complexity ), where E complexity represents the complexity of the data, and γ is the adjustment factor; S33. Finally, the reward and punishment weights are designed for different error sources. The error source reward and punishment formula is: where ω k is the adaptive weight of the k-th error source, The correction error value of each error source is assigned different weights to enhance the sensitivity of the correction strategy to the key error sources.

Images