A magnetic gradient tensor signal adaptive noise suppression method
An adaptive noise suppression method for magnetic gradient tensor signals, optimized by neuronal stochastic resonance and differential evolution algorithms, solves the problem of magnetic gradient tensor signals being susceptible to noise interference and achieves high-precision magnetic anomaly detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHONGQING RES INST OF SHANGHAI JIAOTONG UNIV
- Filing Date
- 2023-11-22
- Publication Date
- 2026-07-31
AI Technical Summary
In existing magnetic anomaly detection technologies, the magnetic gradient tensor signal is easily disturbed by changes in the geomagnetic field and noise interference, resulting in a weak signal that is difficult to suppress effectively, especially in the detection of underground or underwater targets.
An adaptive noise suppression method based on the magnetic gradient tensor signal of the neuron stochastic resonance is adopted. A forward model is established through the dipole model, the error function is optimized by the differential evolution algorithm, and a noise suppression model is established by combining the neuron model. The optimal parameters are optimized to achieve noise suppression.
It achieves high-precision reconstruction of magnetic gradient tensor signals without relying on prior knowledge, effectively suppresses noise, avoids signal deformation, and improves detection results.
Smart Images
Figure CN117607760B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of magnetic anomaly detection, and in particular to an adaptive noise suppression method for magnetic gradient tensor signals. Background Technology
[0002] Magnetic anomaly detection technology boasts advantages such as low cost and high speed, and has been widely applied in fields such as underground metal pipeline inspection and underwater submarine detection. However, traditional magnetic anomaly total field measurement signals contain the geomagnetic field, making them susceptible to disturbances caused by geomagnetic field variations. In contrast, the magnetic gradient tensor signal of a magnetic anomaly can effectively reduce the impact of geomagnetic field changes while containing more magnetic source information. Therefore, magnetic anomaly detection technology based on the magnetic gradient tensor has received widespread attention in recent years. However, because underground or underwater targets are generally far from the measurement point, and because the magnetic gradient tensor signal attenuates with the fourth power of distance, the measured magnetic gradient tensor signal is typically extremely weak and highly susceptible to interference from surrounding noise.
[0003] Current methods for noise reduction in magnetic anomaly detection signals mainly include the following: Wavelet decomposition-based noise removal methods, which require estimating the frequency range of the magnetic anomaly signal and eliminating noise signals outside the target frequency range. This method is highly dependent on the estimated frequency range, and errors in frequency range estimation can lead to incomplete noise removal or damage to useful signals. Deep learning-based adaptive noise reduction methods for magnetic anomaly signals are also effective, but these methods are highly dependent on the selection of training samples. When the representativeness of the selected training samples is low, their generalization performance is poor. In recent years, stochastic resonance methods have been successfully applied to magnetic anomaly detection; however, they are generally used for binary judgments of the presence or absence of magnetic anomalies, and their output signal is severely distorted compared to the original signal, failing to reflect the characteristics of the magnetic anomaly target. Summary of the Invention
[0004] Based on the problems existing in the prior art, the purpose of this invention is to provide a method for adaptive noise suppression of magnetic gradient tensor signals based on neuronal stochastic resonance, which can achieve adaptive noise suppression of magnetic gradient tensor signals under different scenarios and realize high-precision reconstruction of magnetic gradient tensor signals.
[0005] The objective of this invention is achieved through the following technical solution:
[0006] The adaptive noise suppression method for magnetic gradient tensor signals proposed in this invention mainly includes the following steps:
[0007] Step 1: Use a magnetic gradient tensor to measure the measured signal of the magnetic gradient tensor signal;
[0008] Step 2: Using the dipole model, establish a forward model of the magnetic gradient tensor signal;
[0009] Step 3: Use the differential evolution algorithm to optimize the error function, i.e. the objective function, of the forward modeling signal and the measured signal to obtain the inversion parameters of the magnetic anomaly target;
[0010] Step 4: Predict the magnetic gradient tensor signal using the inversion parameters, and calculate the forward reference signal of the magnetic gradient tensor signal.
[0011] Step 5: Using a neuron model, establish a model for suppressing random resonance noise signals in neurons;
[0012] Step 6: Process the measured magnetic gradient tensor signal using neuronal stochastic resonance noise signal suppression models with different model parameters, and calculate the noise-suppressed signal of the magnetic gradient tensor signal.
[0013] Step 7: Optimize the fitness functions of the forward reference signal and the noise suppression signal using the differential evolution algorithm to calculate the optimal model parameters of the neuron stochastic resonance noise signal suppression model.
[0014] Step 8: Use the optimal neuron stochastic resonance noise signal suppression model to suppress the noise of the measured magnetic gradient tensor signal and output the optimal noise-suppressed signal of the magnetic gradient tensor signal.
[0015] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0016] This invention does not rely on prior knowledge, such as frequency range prediction of magnetic anomaly signals or selection of training datasets, effectively avoiding errors caused by human experience and achieving adaptive suppression of magnetic anomaly signal noise. Simultaneously, it effectively avoids deformation of the original magnetic gradient tensor signal, achieving noise suppression effects such as… Figure 4 As shown. Attached Figure Description
[0017] The accompanying drawings of this invention are described below:
[0018] Figure 1 This is a flowchart of an adaptive noise suppression method for magnetic gradient tensor signals according to an embodiment of the present invention;
[0019] Figure 2 This is a schematic diagram of an example data acquisition platform according to an embodiment of the present invention;
[0020] Figure 3 This is a flowchart of an adaptive noise suppression method for magnetic gradient tensor signals according to a preferred embodiment of the present invention;
[0021] Figure 4 This is a simulation result of the present invention after suppressing noise in the magnetic gradient tensor signal. Detailed Implementation
[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0023] Figure 1 This is a flowchart of an adaptive noise suppression method for magnetic gradient tensor signals according to an embodiment of the present invention; as shown below. Figure 1 As shown, the method includes:
[0024] Step 1: Use a magnetic gradient tensor to measure the measured signal of the magnetic gradient tensor signal;
[0025] In this embodiment, as Figure 2 As shown, a magnetic gradient tensor signal s is measured along the survey line l at a speed of V = 2 m / s using a magnetic gradient tensor instrument. m (t)=[Bm xx (t),Bm xy (t),Bm xz (t),Bm yy (t),Bm yz [(t)], where B xx (t), B xy (t),B xz (t),B yy (t),B yz (t) represents the gradients of the x-component of the magnetic induction intensity B in the x, y, and z directions, and the gradient of the y-component in the y and z directions, respectively. t is the time it takes for the magnetic gradient tensor to move; CPA is defined as the minimum approach distance, i.e., CPA is perpendicular to the survey line l, and the time it takes for the magnetic gradient tensor to move at this point is denoted as t0, also known as the overpass time; the magnetic moment M of the magnetic anomaly target = [M x M y M z The distance between the magnetic anomaly target and the magnetic gradient tensor is denoted as r = [x, y0, z0], where x = V × (t - t0). The unknown parameters in the above model are: the magnetic moment M of the magnetic anomaly target, the overpass time t0, and the distances y0 and z0.
[0026] Step 2: Using the dipole model, establish a forward model of the magnetic gradient tensor signal;
[0027] In this embodiment, a forward model s of the magnetic gradient tensor signal changing with time is established based on the dipole approximation. f (t):
[0028]
[0029] Where μ0 is the vacuum permeability, r represents the distance between the magnetic anomaly target and the magnetic gradient tensor, and r = ||r|| 2 r = [x, y0, z0], representing the three-dimensional coordinate distance between the magnetic anomaly target and the magnetic gradient tensor, and t is the movement time of the magnetic gradient tensor.
[0030] Step 3: Use the differential evolution algorithm to optimize the error function, i.e. the objective function, of the forward modeling signal and the measured signal to obtain the inversion parameters of the magnetic anomaly target;
[0031] In this embodiment of the invention, the error function between the forward modeling signal and the measured signal is used as the objective function, and the objective function is solved using the differential evolution algorithm. The objective function f1 is expressed as:
[0032]
[0033] Where t0 is the moment when the magnetic gradient tensor is closest to the magnetic anomaly target, [y0,z0] represents the two-dimensional coordinates of the magnetic anomaly target and the magnetic gradient tensor, [M x M y M z ] represents the magnetic moment of the magnetic anomaly target; S f S represents the forward modeling signal of the magnetic gradient tensor signal. m This represents the measured signal of the magnetic gradient tensor.
[0034] In this embodiment, after solving the objective function using the differential evolution algorithm, the inversion parameters corresponding to minimizing the objective function are obtained, which are also the optimized parameters of the magnetic anomaly target: para1=[t0,y0,z0,M x M y M z ].
[0035] In this embodiment, the Differential Evolution Algorithm (DE) is a highly efficient global optimization algorithm. It is also a population-based heuristic search algorithm, where each individual in the population corresponds to a solution vector. In the DE, the gene of each individual represents a candidate solution to the problem. Each iteration first performs a mutation operation, selecting the genes of one or more individuals as the base, then selecting the differences between different individuals to form a differential gene, and finally adding the base gene to the differential gene to obtain a new individual. A crossover operation crosses the new individual with the corresponding individual from the parent generation, followed by a selection operation, comparing the crossover individual with the corresponding individual from the parent generation, and selecting the better individual to be retained for the next generation. After the iteration is complete, the gene of the best individual in the population is selected as the solution.
[0036] Step 4: Predict the magnetic gradient tensor signal using the inversion parameters, and calculate the forward reference signal of the magnetic gradient tensor signal.
[0037] In this embodiment of the invention, the above-mentioned optimized parameter para1=[t0,y0,z0,M x M y M z Substitute into the forward model s f (t) can then generate the corresponding forward reference signal s. f .
[0038] Step 5: Using a neuron model, establish a model for suppressing random resonance noise signals in neurons;
[0039] Stochastic resonance refers to the phenomenon in nonlinear systems where noise acts as a medium, causing a weak periodic signal to interact with the system itself, thereby enhancing the extraction of the weak signal. Clearly, stochastic resonance's handling of noise differs from other methods of noise suppression or elimination. Stochastic resonance does not eliminate noise; rather, it fully utilizes noise to amplify the weak signal. Suppression methods, on the other hand, aim to eliminate noise as much as possible. Recent research has shown that noise exists in biological neural systems, and stochastic resonance also exists. Therefore, the neuron model has been proposed. Based on the neuron model, a stochastic resonance noise signal suppression model can be established, expressed as:
[0040]
[0041]
[0042] Where: v is the cell membrane voltage, a fast variable; w is the intracellular ion concentration, a slow variable; A T Let ε be the threshold voltage; ε be the time parameter, determining the firing rate of the neuron; B be the distance from the signal amplitude to the threshold voltage; n be Gaussian white noise; and s be the input magnetic gradient tensor signal. Only an appropriate noise intensity can enhance the system's output for weak signals. Since the noise intensity is not adjustable in practice, the system parameters need to be adjusted to achieve resonance between the system and the noise, i.e., parameter-tuned stochastic resonance. For the above system, the adjustable parameters are [ε, B], and the two work together to improve the system's output for weak signals.
[0043] Step 6: Process the measured magnetic gradient tensor signal using neuronal stochastic resonance noise signal suppression models with different model parameters, and calculate the noise-suppressed signal of the magnetic gradient tensor signal.
[0044] In this embodiment, since the parameters of the above-mentioned neuronal stochastic resonance noise signal suppression model are variable, different model parameters correspond to different neuronal stochastic resonance noise signal suppression models. This embodiment uses different neuronal stochastic resonance noise signal suppression models to process the measured signal with the magnetic gradient tensor signal, obtaining different noise suppression signals for the magnetic gradient tensor signal. Subsequent processes will use the obtained noise suppression signals to optimize the model until the optimal result is achieved. Since the output of the above system causes deformation of the input signal, the noise suppression signal is established based on the fast variable v of the above model as follows:
[0045]
[0046] Where v is the cell membrane voltage, A T B is the threshold voltage; B is the distance from the signal amplitude to the threshold voltage.
[0047] Step 7: Optimize the fitness functions of the forward reference signal and the noise suppression signal using the differential evolution algorithm to calculate the optimal model parameters of the neuron stochastic resonance noise signal suppression model.
[0048] In this embodiment of the invention, based on the forward reference signal s f and noise suppression signal s r The relationship between them can be used to establish a fitness function, which can be expressed as:
[0049]
[0050] in, and s f and s r The mean, s f The forward reference signal s represents the magnetic gradient tensor signal. r This represents the noise suppression signal for the magnetic gradient tensor signal.
[0051] By solving the fitness function, the optimal model parameters para2 = [ε0B0] can be obtained, where ε0 is the optimal time parameter, which determines the firing rate of neurons; and B is the optimal distance from the signal amplitude to the threshold voltage.
[0052] Step 8: Use the optimal neuron stochastic resonance noise signal suppression model to suppress the noise of the measured magnetic gradient tensor signal and output the optimal noise-suppressed signal of the magnetic gradient tensor signal.
[0053] In this embodiment, the optimal noise suppression signal of the magnetic gradient tensor signal can be obtained through the iterative optimization process of steps three to seven described above.
[0054] To better illustrate the iterative optimization process of this invention, Figure 3 This is a flowchart of a preferred embodiment of the magnetic gradient tensor signal noise adaptive suppression method of the present invention, as shown below. Figure 3 As shown, the method includes, before iteration, acquiring magnetic gradient tensor signals Sm along the survey line using a magnetic gradient tensor instrument, and establishing a forward model S of the magnetic gradient tensor signals. f = f(M, t0, y0, z0); During the formal iteration, the objective function f1 is optimized using the differential evolution algorithm, and the magnetic anomaly objective function para1 = [t0, y0, z0, M] is derived. x M y M z The measured signal is input into different stochastic resonance models for noise suppression, and the noise suppression signal S is output. r A fitness function f2 is established, and based on the fitness function, the differential evolution algorithm is used to maximize the fitness to determine the optimal model parameters para2 = [ε0B0]. The noise suppression signal f2 under the optimal model is output. The decision to enter the next iteration is made by judging whether the fitness function is greater than a preset threshold or whether the maximum number of iterations has been reached. If the next iteration is entered, the measured value of the magnetic gradient signal is set to equal the noise suppression signal S obtained by solving. r Continue to use the differential evolution algorithm to solve the objective function until the loop ends. If the loop does not enter the next iteration, that is, the loop ends, then the process ends directly.
[0055] Figure 4 This is a simulation result of the present invention after noise suppression of the magnetic gradient tensor signal, as shown in the figure. Figure 4 As shown, the left image represents the original noisy signal, and the right image represents the noise-suppressed signal. It can be seen that after adding the noise-suppressed signal, the deformation of the original magnetic gradient tensor signal can be effectively avoided, and adaptive suppression of magnetic anomaly signal noise is achieved.
[0056] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include ROM, RAM, disk, or optical disk, etc.
[0057] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. An adaptive noise suppression method for magnetic gradient tensor signals, characterized in that, The method includes the following steps: Step 1: Use a magnetic gradient tensor to measure the measured signal of the magnetic gradient tensor signal; Step 2: Using the dipole model, establish a forward model of the magnetic gradient tensor signal; Step 3: Use the differential evolution algorithm to optimize the error function, i.e. the objective function, of the forward modeling signal and the measured signal to obtain the inversion parameters of the magnetic anomaly target; Step 4: Predict the magnetic gradient tensor signal using the inversion parameters, and calculate the forward reference signal of the magnetic gradient tensor signal. Step 5: Using a neuron model, establish a model for suppressing random resonance noise signals in neurons; Step 6: Process the measured magnetic gradient tensor signal using neuronal stochastic resonance noise signal suppression models with different model parameters, and calculate the noise-suppressed signal of the magnetic gradient tensor signal. Step 7: Optimize the fitness functions of the forward reference signal and the noise suppression signal using the differential evolution algorithm to calculate the optimal model parameters of the neuron stochastic resonance noise signal suppression model. Step 8: Use the optimal neuron stochastic resonance noise signal suppression model to suppress noise in the measured magnetic gradient tensor signal and output the optimal noise-suppressed signal of the magnetic gradient tensor signal.
2. The adaptive noise suppression method for magnetic gradient tensor signals as described in claim 1, characterized in that, Step one includes using a magnetic gradient tensor to move along the measurement line at a speed of V and measuring the magnetic gradient tensor signal s. m The moment when the magnetic gradient tensor is closest to the magnetic anomaly target is defined as the overpass time, denoted as t0, and the magnetic moment of the magnetic anomaly target is M = [M x M y M z The distance between the magnetic anomaly target and the magnetic gradient tensor is denoted as r = [x, y0, z0], where x = V × (t - t0) and t is the time it takes for the magnetic gradient tensor to move.
3. The adaptive noise suppression method for magnetic gradient tensor signals as described in claim 1, characterized in that: The step two forward model s of the magnetic gradient tensor signal f (t) is given by the expression Among them, B xx (t), B xy (t),B xz (t),B yy (t),B yz (t) represents the x-component of magnetic flux density B in the x, y, and z directions, and the gradient of the y-component in the y and z directions, respectively. μ0 is the free permeability, and r represents the distance between the magnetic anomaly target and the magnetic gradient tensor, r = ||r|| 2 r = [x, y0, z0], representing the three-dimensional coordinate distance between the magnetic anomaly target and the magnetic gradient tensor, where x = V × (t - t0), V is the speed at which the magnetic gradient tensor moves along the survey line, and t is the movement time of the magnetic gradient tensor.
4. The adaptive noise suppression method for magnetic gradient tensor signals as described in claim 1, characterized in that: The objective function f1 in step three is as follows: Where t0 is the moment when the magnetic gradient tensor is closest to the magnetic anomaly target, [y0,z0] represents the two-dimensional coordinates of the magnetic anomaly target and the magnetic gradient tensor, [M x M y M z ] represents the magnetic moment of the magnetic anomaly target; S f S represents the forward modeling signal of the magnetic gradient tensor signal. m This represents the measured signal of the magnetic gradient tensor.
5. The adaptive noise suppression method for magnetic gradient tensor signals as described in claim 1, characterized in that: In step five, the neuronal stochastic resonance noise signal suppression model is as follows: Where: v is the cell membrane voltage, a fast variable; w is the intracellular ion concentration, a slow variable; A T ε is the threshold voltage; t is the movement time of the magnetic gradient meter; ε is a time parameter that determines the firing rate of the neuron; B is the distance from the signal amplitude to the threshold voltage; n is Gaussian white noise; and s is the input magnetic gradient tensor signal.
6. The adaptive noise suppression method for magnetic gradient tensor signals as described in claim 1, characterized in that: In step six, the noise suppression signal calculated based on the neuronal random resonance noise signal suppression model is as follows: Where v is the cell membrane voltage, A T B is the threshold voltage; B is the distance from the signal amplitude to the threshold voltage.
7. The adaptive noise suppression method for magnetic gradient tensor signals as described in claim 1, characterized in that: In step seven, the fitness function f2 is formulated as follows: in, and s f and s r The mean, s f The forward reference signal s represents the magnetic gradient tensor signal. r This represents the noise suppression signal for the magnetic gradient tensor signal.
8. The adaptive noise suppression method for magnetic gradient tensor signals as described in claim 1, characterized in that: In step seven, the optimization process includes determining whether the fitness function is greater than a preset threshold or has reached the maximum number of iterations. If not, the measured signal of the magnetic gradient tensor signal is set to equal the calculated noise suppression signal, and the process returns to step three. If yes, the process proceeds to step eight.