Self-adaptive magnetic anomaly signal detection method and system based on stochastic resonance

Abstract

The invention discloses a self-adaptive magnetic anomaly signal detection method and system based on stochastic resonance. According to the method, an EWSSR potential energy function is adopted and has very high flexibility, and the function shape can be changed from multiple dimensions; meanwhile, double magnetic detecting instruments are adopted, a static magnetometer is arranged locally to monitor local noise in real time, noise measurement and noise and target magnetic anomaly signal approximation are adopted to be more suitable for a potential energy function shape under the current noise background, and adjustment and optimization of system parameters are achieved in combination with a target function in an optimization algorithm. According to the method, self-real-time detection of abnormal signals is achieved by adjusting system parameters in a real-time and self-adaptive mode, the problem that parameter adjustment is difficult in a traditional method is solved, and effective detection of magnetic anomaly signals can be achieved in the environment with the lower signal-to-noise ratio. Moreover, due to the changeable function shape, different from a traditional method, an initial value setting problem needs to be considered, and under the potential energy function, wave crest and wave trough signals can be detected.

Description

Technical Field

[0001] The present invention relates to the technical field of weak signal detection, and in particular to an adaptive magnetic anomaly signal detection method and system based on stochastic resonance (SR), which can detect abnormal magnetic fields generated by ferromagnetic targets in real time under low signal-to-noise ratio conditions. Background Art

[0002] In the same geographical location within a short period of time, the earth itself has a relatively stable magnetic field, while ferromagnetic targets will cause abnormal distribution of the geomagnetic field within a certain range. Based on this phenomenon, a magnetic detector can be used to detect abnormal magnetic fields to determine the existence of magnetic targets. This process is called magnetic anomaly target detection. Magnetic anomaly target detection technology is not easily affected by media such as air and soil. It has the advantages of good stability, strong anti-interference ability, efficient execution, lightness and concealment. Therefore, it is widely used in civilian and military scenarios, such as oil surveys, mineral exploration, earthquake monitoring, mine clearance and explosive disposal, submarine detection and other scenarios. However, due to the complex composition of the geomagnetic field, the background magnetic field measured by the magnetic detector is unpredictable color noise; and in the case of long-distance detection, the size of the target magnetic anomaly signal is usually weaker than the background magnetic field and is submerged in the magnetic field noise. It belongs to the problem of weak signal detection and has become a research hotspot in recent years.

[0003] In order to detect weak magnetic anomaly signals, the research conducted by researchers can be divided into two categories: target-based and noise-based. The target-based magnetic anomaly signal detection method requires certain prior information about the target. According to the characteristics of the target magnetic anomaly signal that are different from the chaotic geomagnetic field noise, the target magnetic anomaly signal information is extracted. The researchers proposed a detection method based on orthonormal basis function (OBF). This method constructs three orthogonal basis functions according to the magnetic field expression of the magnetic target, and uses the components of the measured signal on the orthogonal basis as the detection quantity. It can achieve effective detection under low signal-to-noise ratio of white noise. However, this method constructs an orthogonal basis based on the magnetic field signal generated by the target, and requires certain prior information to achieve high-quality detection. In addition, the performance of this method is unstable in complex noise environments. Subsequently, a series of improved methods based on OBF were developed: such as the improved Karhunen–Loeve extended orthogonal basis algorithm, which modifies the classical orthogonal basis according to the statistical characteristics of geomagnetic noise, and improves the detection rate in a color noise environment; at the same time, some scholars have proposed a method of preprocessing the measurement signal through a whitening filter and a narrowband filter, and then using OBF detection, which has also improved the low detection rate problem in a complex noise environment to a certain extent. In addition, the singular value decomposition method, the principal component analysis method, and the time-frequency analysis method have all made corresponding improvements to the problem of magnetic anomaly signal detection, thereby achieving better detection results. However, in actual applications, it is difficult to obtain prior information on the target magnet in many scenarios, and the background noise is complex. To achieve reliable detection and minimize missed and misjudgment, the traditional OBF method is difficult to meet the detection needs.

[0004] The target-based magnetic anomaly signal detection methods all use the characteristics of the target abnormal magnetic field that are different from noise to achieve detection, which requires certain prior information, while the noise-based detection methods use the noise characteristics to achieve detection, and require less prior information. At first, some scholars keenly observed that the target magnetic field destroyed the chaotic pattern of noise, and proposed the minimum entropy (ME) detection method based on the principle that the entropy of the measured signal with the target abnormal signal is less than the entropy of the pure noise signal; afterwards, a series of algorithms were born, such as the energy detector based on the empirical mode decomposition (EMD) trend filter and the adaptive coherent noise suppression (ACS) method. The most widely used one is the magnetic anomaly signal detection method based on the SR theory proposed in 2018. The present invention belongs to this category and proposes a better stochastic resonance system, which can achieve a higher detection rate under adaptive conditions.

[0005] The SR system uses the displacement principle of Brownian particles in the approximate adiabatic theory to construct a motion model of Brownian particles in a potential well. A magnetic anomaly signal containing noise is input into the system. When the system matches it, the energy of the noise can be used to enhance the amplitude of the signal, and the state jump of different potential wells can be achieved. An output signal with a high signal-to-noise ratio can be obtained, such as Figure 1 As shown. The SR system was first proposed by R.Benzi to explain the periodicity of the ancient Earth Ice Age. It has now been widely used in the interpretation of natural physical phenomena and the detection of weak signals, especially in the field of bearing fault detection, solving many difficult problems and achieving remarkable results. Although some people have migrated the SR system to the field of magnetic anomaly signal detection, after analysis, the detection of magnetic anomaly signals is different from bearing signals and other signals in many aspects. First, the bearing signal contains more known factors, while the background noise of magnetic anomaly signal detection is more complex and has more unpredictable factors. Therefore, to build a random resonance system that matches the magnetic anomaly signal, it is necessary to combine the noise information more deeply; secondly, the bearing signal is a periodic signal. The purpose of the SR system built for the bearing signal is to detect whether there is a fault signal in the abnormal frequency band by improving the signal-to-noise ratio, while the magnetic anomaly signal is a non-periodic signal. Its purpose is to ensure that the false alarm rate is reasonable and achieve a higher detection rate as much as possible, that is, to build an SR system that can accurately realize the jump, and realize the state does not change in the case of pure noise but the state changes in the case of magnetic anomaly signals. Its focus is not on improving the signal-to-noise ratio.

[0006] In addition, the SR detection method is a model of Brownian motion vibrating in a potential energy function. The potential energy function models of existing SR detection methods include the classic bistable potential energy function, monostable potential energy function, and tristable potential energy function. These potential energy functions contain only 2-3 parameters, which results in the potential energy function model changing less and being relatively monotonous, making it difficult to design a random resonance system that can achieve the best match with the input signal. At the same time, before the SR system can detect, it is necessary to determine the system parameters and fix the system model based on the input measurement signal, so the parameter setting of the existing SR system is also a difficult problem, and it is difficult to achieve adaptive detection. In summary, the SR system can adapt to the situation where the overall noise level remains unchanged and achieve stable detection, because the parameters of the SR system remain unchanged during detection, but if the noise level changes, the state transition may be inaccurate, resulting in an excessively high false alarm rate or a low detection rate. Summary of the invention

[0007] In view of this, the present invention provides an adaptive magnetic anomaly signal detection method and system based on stochastic resonance, which can adaptively adjust system parameters in real time to achieve real-time and effective detection of magnetic anomaly signals in a lower signal-to-noise ratio environment.

[0008] The adaptive magnetic anomaly signal detection method based on stochastic resonance of the present invention comprises:

[0009] Step 1: Use dual magnetic detectors. Based on the strong spatial coherence of the geomagnetic field, one of the magnetic detectors is stationary around the test site to measure the local magnetic field information in real time as the measured reference noise N. ref Another magnetic detector moves on the test ground to measure the magnetic field information at different positions as the target detection signal S MAD ;

[0010] Step 2, construct the stochastic resonance system EWSSR, whose differential equation is:

[0011]

[0012] Where M(t) is the detection signal of the magnetic detector, x is the particle displacement; U(x) is the potential energy function, U(x)=U E (x)-U WS (x), a, b, v, r, c are parameters; a and b determine the slope of the potential well wall, and the slope of the well wall increases with the increase of a and b; c determines the slope of the potential barrier, and the slope of the potential barrier decreases with the increase of c; v determines the height of the potential barrier, and the height of the potential barrier increases with the increase of v; r determines the width of the potential well, and the width of the potential well decreases with the increase of r;

[0013] Step 3, synchronously input the detection signal of the dual magnetic detector in step 1 into the stochastic resonance system EWSSR, and use the optimization algorithm to optimize the potential energy function parameters a, b, v, r, c, wherein the optimization objective function is:

[0014]

[0015] Among them, SNR(EWSSR(S MAD ,a,b,c,v,r)) is M(t)=S MAD The signal-to-noise ratio of the stochastic resonance system EWSSR, SNR(EWSSR(N ref ,a,b,c,v,r)) is M(t)=N ref The signal-to-noise ratio of the stochastic resonance system EWSSR is:

[0016] Step 4: Transform the target detection signal S MAD Input into the stochastic resonance system EWSSR optimized in step 3 to detect magnetic anomaly signals.

[0017] Preferably, the measured reference noise N is used regularly. ref The parameters of the potential energy function are optimized, the stochastic resonance system EWSSR is updated, and the updated stochastic resonance system EWSSR is used to detect magnetic anomaly signals.

[0018] Preferably, in step 3, the optimization algorithm adopts an algorithm such as QPSO (quantum-behaved particle swarm algorithm), PSO (particle swarm algorithm) or GA (genetic algorithm).

[0019] The present invention also provides an adaptive magnetic anomaly signal detection system based on stochastic resonance, comprising a magnetic detector A, a magnetic detector B and a detection unit;

[0020] The magnetic detector A is statically set around the test site to measure the local magnetic field information in real time, which is used as the measured reference noise N. ref ;

[0021] The magnetic detector B moves on the test ground and measures the magnetic field information at different positions as the target detection signal S MAD ;

[0022] The detection unit adopts the stochastic resonance system EWSSR in the method as described in claims 1 to 3, and based on the measured reference noise N of the magnetic detector A ref and the target detection signal S of the magnetic detector B MAD Parameter optimization is carried out, and magnetic anomaly signals are detected based on the optimized stochastic resonance system EWSSR.

[0023] Beneficial effects:

[0024] (1) The present invention adopts the EWSSR potential energy function, which has high flexibility and can change the function shape from multiple dimensions; at the same time, a dual magnetic detector is used, a static magnetometer is set up locally to monitor the local noise in real time, and the measurement noise and the noise plus the target magnetic anomaly signal are used to approximate the potential energy function shape that is more suitable for the current noise background, and the objective function in the optimization algorithm is combined to achieve the tuning of the system parameters, and the system optimal stochastic resonance system that can only achieve state transition after the input of the noisy magnetic anomaly signal is constructed. After verification, the trained stochastic resonance system can obtain better detection effect and can achieve effective detection of magnetic anomaly signals in a lower signal-to-noise ratio environment. At the same time, the present invention realizes the self-real-time detection of abnormal signals by real-time adaptive adjustment of system parameters, which solves the problem of difficult parameter adjustment of traditional methods. In addition, due to the variable function shape, unlike the traditional method, the initial value setting problem needs to be considered. Under this potential energy function, both the peak and trough signals can be detected, which can adapt to the detection of magnetic anomaly signals under different motion trajectories (different signal waveforms) of the magnetic detector, and there is no need to set the initial state of the stochastic resonance system.

[0025] (2) The present invention uses the QPSO algorithm to tune the system parameters. The algorithm simplifies the parameter setting. Only one parameter, the innovation index, needs to be set. The setting of parameters such as the individual learning factor, the group learning factor, and the speed inertia factor in the traditional PSO algorithm is omitted. In addition, the way of updating the position is more random and not easy to fall into the local optimal value. By comparison, it can show more stable effects under complex optimization problems and has good real-time performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Schematic diagram of a stochastic resonance system.

[0027] Figure 2 It is a flow chart of the present invention.

[0028] Figure 3 This is the potential energy function diagram of the classic bistable state.

[0029] Figure 4 Graphs of the EWSSR potential energy function of the present invention. (a) is a monostable potential energy function constructed by the EWSSR potential energy function of the present invention; (b) is a bistable potential energy function constructed by the EWSSR potential energy function of the present invention; (c) is a bistable potential energy function constructed by the EWSSR potential energy function of the present invention.

[0030] Figure 5 The influence of each parameter on the EWSSR potential energy function. Among them, (a) is the influence of parameter a on the EWSSR potential energy function; (b) is the influence of parameter b on the EWSSR potential energy function; (c) is the influence of parameter c on the EWSSR potential energy function; (d) is the influence of parameter v on the EWSSR potential energy function; (e) is the influence of parameter r on the EWSSR potential energy function.

[0031] Figure 6 This is the flow chart of QPSO algorithm.

[0032] Figure 7 Training graph for stochastic resonance system.

[0033] Figure 8 This is the output of the stochastic resonance system.

[0034] Fig. 9 It is the output after standard deviation detection.

[0035] Fig.10 This is a graph showing the objective function changing with the number of iterations.

[0036] Fig.11 The figure is a comparison of the detection rates of the method of the present invention and the original method. DETAILED DESCRIPTION

[0037] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0038] The invention provides a method and system for adaptive magnetic anomaly signal detection based on stochastic resonance.

[0039] Due to the many application scenarios and complex noise background of magnetic anomaly signal detection system, the present invention adopts the method of dual detectors. In addition to the target magnetic detector, a stationary magnetic detector is set in the detection environment to monitor the local noise situation in real time, and the random resonance system is designed by integrating noise information, so that the parameters of the random resonance system can be accurately and effectively determined, and the random resonance system is constructed as auxiliary information, so that the system parameters can be updated regularly to ensure that the detection system can match the latest input signal well. In theory, the system is in a state where the input pure noise signal does not change state but the input training signal sample changes. At the same time, in view of the problem that the potential energy function is monotonous and the adjustment parameters are difficult in the existing random resonance detection method, the exponential function joint Woods-Saxon function (Exponential function joint Woods-Saxon function, EWSfunction) is used as the potential energy function for detecting magnetic anomaly signals. The function includes five parameters, which can change the shape from different dimensions and adapt to different input functions; and the quantum-behaved particle swarm optimization (Quantum-behaved ParticleSwarm Optimization, QPSO) is combined to find the optimal parameters of the potential energy function to achieve full-process adaptive detection.

[0040] Figure 2 It is a flow chart of the present invention. The overall system is designed into two parts: training parameters and application system, including two magnetic detectors and a detection unit; the detection unit adopts a stochastic resonance system EWSSR.

[0041] Stochastic resonance is different from other noise suppression methods. It can use the amplitude of noise to enhance the amplitude of effective signal. The principle of this method is gradually evolved from Brownian motion. The differential equation that originally expressed Brownian thermal motion is:

[0042]

[0043] Where m represents the mass of the particle, x represents the displacement of the particle, is the resistance of the particle, and F(t) is the noise force generated by irregular collisions.

[0044] Based on the above physical model, Benzi et al. proposed a stochastic resonance system, whose differential equation is described as:

[0045]

[0046] Among them, x represents the motion trajectory of Brownian particles in the approximate adiabatic theory, U(x) represents the potential energy function, and M(t) is the detection signal of the magnetic detector. Generally, M(t)=S(t)+Γ(t), where S(t) represents the target magnetic anomaly signal and Γ(t) is the noise signal. The physical meaning of the stochastic resonance system is the random vibration trajectory of Brownian particles in the potential energy function. If the potential energy function matches the input detection signal, a state transition can be generated and the signal detection can be realized. Therefore, the selection of the potential energy function is crucial to the entire system, and the selection of the potential energy function includes two levels: potential energy function design and function parameter setting.

[0047] From the perspective of potential energy function selection, the classical stochastic resonance system is a bistable system, and its potential energy function expression is:

[0048] U BSR (x) = -ax 2 / 2+bx 4 / 4 (3)

[0049] like Figure 3 As shown, this bistable potential energy function realizes signal detection by the jumping of particles between two steady states. It contains two parameters, which adjust the depth and width of the potential well respectively. However, it is still not possible to adjust the function shape accurately enough. In addition, the bistable potential energy function still has the problem of initial value setting. If the initial value is a negative steady state, it can only jump in the positive direction to detect the peak signal. Similarly, if the initial value is a positive steady state, it can only detect the trough signal.

[0050] Therefore, the present invention selects an exponential function combined with a Woods-Saxon function as a potential energy function, which includes an exponential function and a Woods-Saxon function. The five parameters can change the shape of the potential energy function from different dimensions, so as to design a suitable potential energy function according to the input signal to achieve precise matching and accurate detection. The expression of the potential energy function is:

[0051]

[0052] Among them U E (x) is an exponential function, U WS (x) is the Woods-Saxon function.

[0053] The potential energy function system is more complex and contains more parameters, so it also has richer dynamic characteristics. Different monostable, bistable, and tristable systems can be designed by setting different parameters, such as Figure 4Therefore, the potential energy function used in the present invention can better adapt to various input signals through parameter setting, and achieve the best adaptation between the input signal and the design system; at the same time, it can also solve the problem of difficult initial state setting of the above-mentioned classic system. The initial state is set to 0, and through appropriate parameter adjustment, it is possible to find a potential energy function that can successfully detect both the peak and the trough at the same time.

[0054] Therefore, by substituting equation (4) into equation (2), we can obtain the basis of the present invention, that is, the differential equation representing the stochastic resonance system, which is expressed as follows:

[0055]

[0056] Since the potential energy function used in the present invention is relatively complex, it includes five system parameters: a, b, c, v, and r, which together determine the shape of the potential energy function; Figure 5 As shown, a and b determine the slope of the potential well wall, and the slope of the potential well wall increases with the increase of a and b. c determines the slope of the potential barrier, and the slope of the potential barrier gradually decreases with the increase of c. v determines the height of the potential barrier, and the larger the v, the higher the height of the potential barrier. Finally, r determines the width of the potential well. As r increases, the width of the potential well will decrease accordingly.

[0057] The complexity of the potential energy function leads to the difficulty of parameter adjustment. The potential energy function proposed in the present invention includes five parameters. Although the function properties controlled by each parameter can be qualitatively calculated, it is still time-consuming and laborious to manually adjust the parameters of the stochastic resonance system based on this alone, and it is not easy to achieve the most accurate detection effect, so it is difficult to apply to many scenes that need to realize real-time detection. To take advantage of the complex potential energy function to be more suitable for various usage scenarios, it is necessary to achieve accurate parameter adjustment. For this reason, the present invention combines optimization algorithms to achieve accurate parameter adjustment of the stochastic resonance system, such as optimization algorithms such as QPSO algorithm, PSO algorithm, and GA algorithm. This embodiment adopts the QPSO algorithm, which is inspired by quantum behavior and improves the movement mode of each particle in the particle swarm optimization algorithm (Particle Swarm Optimization, PSO) algorithm. At the same time, the QPSO algorithm simplifies the parameter setting, only needs to set a parameter of the innovation index, omits the setting of parameters such as individual learning factor, group learning factor, and speed inertia factor in the traditional PSO algorithm, and its method of updating the position is more random, not easy to fall into the local optimum, and can show a more stable effect under complex optimization problems by comparison.

[0058] Figure 6It is the flowchart of the QPSO algorithm. The first step is to set the initial state of each particle. In the present invention, the optimization of each particle includes five dimensions, namely a, b, c, r, and v. Their initial values ​​are randomly set within a certain range. Then, the second step is to evaluate the objective function of the position of each particle, set the first search result as the best historical position of the particle, and obtain the global optimal particle position p after comparison. gbest , and then calculate the average historical best position according to formula (6)

[0059]

[0060] Where M represents the specific number of particles. Represents the best historical position of the i-th particle.

[0061] In the third step, after obtaining the global optimal particle position and the best particle historical position, the position of each particle can be updated according to the quantum behavior, as shown in formula (7):

[0062]

[0063] Among them, φ and u are random parameters uniformly distributed between 0 and 1, α is the only parameter of the innovation parameter that needs to be set by yourself, and the ± in the formula means that the probability of the sign here being positive or negative is 0.5 respectively. After calculation by this formula, the particle position is updated.

[0064] The fourth step is to calculate the new objective function based on the new particle position, determine the latest historical optimal position, the global optimal position, and the average optimal position, so as to achieve iterative update. After the end condition is met, the loop can be ended, the optimal a, b, c, r, v can be found, and the stochastic resonance system that best suits the input noise can be determined.

[0065] The detection signal M(t) of the magnetic detector may contain the target magnetic anomaly signal S(t) and the geomagnetic background noise signal Γ(t), that is, M(t) = S(t) + Γ(t). In order to adapt to the geomagnetic background noise that changes over time, the present invention combines the local measured geomagnetic noise to construct a complete magnetic anomaly signal detection system, and for the first time proposes a training plus practical structure to achieve the precise construction of the stochastic resonance system.

[0066] Specifically, the present invention sets a stationary magnetic detector at the test site to monitor the local magnetic field information in real time. The sensor can reflect the local magnetic field level as effective reference information. The measured noise signal plus the simulated magnetic anomaly signal of different amplitudes can obtain simulated measurement signals with different signal-to-noise ratios. These signals are the training samples input into the training system. Afterwards, the measured pure noise signal and the training signal are input into the QPSO algorithm to find the optimal parameters, and the effective system parameters obtained by the input of the lowest signal-to-noise ratio signal training are selected as the determined system parameters, thereby determining the random resonance system put into use. In theory, the system is the optimal system between the input of pure noise signal without jump and the input of noisy magnetic anomaly signal with state jump. The random resonance system does not change state when there is no target signal input. When the measured signal contains a target magnetic anomaly signal with a certain signal-to-noise ratio, the state changes immediately to achieve detection. At the same time, it is also necessary to consider that since the noise level of a location may change over time, the measured signal input into the training system should be changed regularly, and the parameters of the random resonance system should be updated to ensure that the random resonance system that is most suitable for the magnetic anomaly signal at this time and local point is used during detection.

[0067] Compared with the traditional stochastic resonance method, the present invention collects local measured noise as auxiliary information, and incorporates noise information by constructing an objective function in the training system. The local noise information can play a key role in the accurate construction of the stochastic resonance system.

[0068] The purpose of selecting the parameters of the stochastic resonance system is to achieve a high detection rate of the target measurement signal under the premise of a low false alarm rate. For the stochastic resonance system, successful detection means a state transition, so "low false alarm rate" means that when a pure noise signal is input into the stochastic resonance system, the system state will not change. At the same time, "high detection rate" theoretically means that under the same noise level, if the noise contains a certain intensity of target magnetic anomaly signal, the stochastic resonance system can identify that the signal power is greater than the pure noise signal, thereby achieving a state transition, that is, successful detection. A sufficiently high detection rate can only be achieved when the parameters of the stochastic resonance system are set to fit the noise. Figure 7It can explain how the noise situation determines the training of the random resonance system. In the figure, the Brownian particle vibrates randomly in the potential energy function. The noise and signal of the input system determine the vibration of the Brownian particle in the potential energy function. Taking the training of the system under the noise situation in the figure as an example, the yellow threshold is the movement range of the Brownian particle after the input of pure noise, and the red threshold is the movement range of the Brownian particle after the input of the magnetic anomaly signal (including the target signal) under the same noise level. If the potential barrier after our parameter adjustment is lower than the yellow threshold, then regardless of whether the magnetic anomaly signal exists, a state transition will occur, and effective detection cannot be achieved; if the potential barrier after parameter adjustment is higher than the red threshold, regardless of whether the pure noise signal or the noisy magnetic anomaly signal is input, the Brownian particle can only vibrate in the potential well, without state transition, and detection cannot be achieved; so the potential barrier height of the ideal system we need to find should be between the yellow threshold and the red threshold, and the closer to the yellow threshold, the better. In theory, the system constructed in this way will not only not undergo state transition under this noise condition, but also can achieve detection as long as the amplitude of the noisy magnetic anomaly signal is slightly larger than the pure noise signal. Therefore, this paper uses two signals, noisy magnetic anomaly signals and pure noise signals, to gradually approximate the ideal system and realize the training of the stochastic resonance system under this noise condition.

[0069] Specifically, this paper uses the signal-to-noise ratio indicator to determine whether the stochastic resonance system has a state transition. Figure 8 It is the typical output of the stochastic resonance system after successful detection by inputting pure noise signal and noisy magnetic anomaly signal respectively. Figure 8 The blue signal in the middle is the stochastic resonance output of the pure noise input signal. The Brownian particle can only vibrate in the potential well and no state transition occurs. Therefore, the fluctuation of the motion trajectory x of the Brownian particle is very small. The red signal is the system output after the noisy magnetic anomaly signal is input. It can be seen that the output state transition has obviously occurred. By calculating the standard deviation of a sliding window, we can get Fig. 9 The standard deviation detection output of the input noisy magnetic anomaly signal is on the order of 1, which has a high signal-to-noise ratio, but the output of the input pure noise signal is on the order of 10. -3 , the signal-to-noise ratio is very low. By calculating the signal-to-noise ratio of the signal after standard deviation detection, the evaluation index can be obtained. By calculation, the signal-to-noise ratio of the standard deviation detection result of the noisy magnetic anomaly signal is much greater than that of the pure noise signal. Based on the above idea, the present invention uses two signals to approximate the parameters of the ideal system, that is, to achieve the jump of the magnetic anomaly signal while the noise signal does not jump, so the objective function of this paper is designed as:

[0070]

[0071] Among them, N ref Refers to the measured reference noise, S MADRefers to the noisy magnetic anomaly signal, EWSSR() refers to the output of the EWSSR system constructed according to the a, b, c, v, and r parameters when the signal is input, and SNR() is a function for calculating the signal-to-noise ratio.

[0072] After training with this objective function, it is possible to achieve a high signal-to-noise ratio for the output of noisy signals and a low signal-to-noise ratio for the output of pure noise signals. After multiple iterations of training, by substituting the trained parameters a, b, c, v, and r, an ideal stochastic resonance system can be obtained. Fig.10 It is a graph showing the objective function changing with the number of iterations during training. The objective function continues to increase with the number of iterations.

[0073] like Fig.11 As shown, under the same noise and signal conditions, the higher detection performance of the present invention can be verified by comparative experiments with the classical parallel stochastic resonance system and the three-stable stochastic resonance system. This is because the present invention uses a multi-parameter, more flexible potential energy function; in addition, the QPSO algorithm is used to solve the system parameter adjustment problem. Secondly, the traditional stochastic resonance method also needs to consider the initial value setting problem when detecting non-periodic signals. The potential energy function selected by the present invention can achieve the switching of the three stable states through design, so there is no need to consider the setting of the initial value, whether it is a peak or a trough, it can be successfully detected. Finally, the present invention can realize adaptive detection. By measuring the real-time noise at the detection location, the parameters of the stochastic resonance system can be adjusted, and the system parameters can be continuously updated, thereby realizing high-quality detection at any time and any place. The present invention can meet the detection requirements with strong real-time performance in complex scenes of noise environment.

[0074] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An adaptive magnetic anomaly signal detection method based on stochastic resonance, characterized in that: include: Step 1: Use dual magnetic detectors, one of which is stationary around the test site to measure the local magnetic field information in real time as the measured reference noise N ref Another magnetic detector moves on the test ground to measure the magnetic field information at different positions as the target detection signal S MAD ; Step 2, construct the stochastic resonance system EWSSR, whose differential equation is: Where M(t) is the detection signal of the magnetic detector, x is the particle displacement; U(x) is the potential energy function, U(x)=U E (x)-U WS (x), a, b, v, r, c are parameters; a and b determine the slope of the potential well wall, and the slope of the well wall increases with the increase of a and b; c determines the slope of the potential barrier, and the slope of the potential barrier decreases with the increase of c; v determines the height of the potential barrier, and the height of the potential barrier increases with the increase of v; r determines the width of the potential well, and the width of the potential well decreases with the increase of r; Step 3, synchronously input the detection signal of the dual magnetic detector in step 1 into the stochastic resonance system EWSSR, and use the optimization algorithm to optimize the potential energy function parameters a, b, v, r, c, wherein the optimization objective function is: Among them, SNR(EWSSR(S MAD ,a,b,c,v,r)) is M(t)=S MAD The signal-to-noise ratio of the stochastic resonance system EWSSR, SNR(EWSSR(N ref ,a,b,c,v,r)) is M(t)=N ref The signal-to-noise ratio of the stochastic resonance system EWSSR is: Step 4: Transform the target detection signal S MAD Input into the stochastic resonance system EWSSR optimized in step 3 for detection.

2. The method according to claim 1, characterized in that Regularly use the measured reference noise N ref The parameters of the potential energy function are optimized, the stochastic resonance system EWSSR is updated, and the updated stochastic resonance system EWSSR is used to detect magnetic anomaly signals.

3. The method according to claim 1, characterized in that In step 3, the optimization algorithm adopts QPSO, PSO or GA algorithm.

4. An adaptive magnetic anomaly signal detection system based on stochastic resonance, characterized in that: It includes a magnetic detector A, a magnetic detector B and a detection unit; The magnetic detector A is statically set around the test site to measure the local magnetic field information in real time, which is used as the measured reference noise N. ref ; The magnetic detector B moves on the test ground and measures the magnetic field information at different positions as the target detection signal S MAD ; The detection unit adopts the stochastic resonance system EWSSR in the method as described in claims 1 to 3, and based on the measured reference noise N of the magnetic detector A ref and the target detection signal S of the magnetic detector B MAD Parameter optimization is carried out, and magnetic anomaly signals are detected based on the optimized stochastic resonance system EWSSR.

Images