De-noising method for infrasonic wave signal before bank collapse, feature recognition method and device

Abstract

The invention discloses a noise reduction processing method for infrasound wave signals before bank collapse, a feature recognition method and a device, wherein the noise reduction processing method for infrasound wave signals before bank collapse uses discrete wavelet transform to extract time-frequency domain features, and the discrete wavelet transform uses filters of different frequencies to analyze signals of different frequencies, mainly high-pass filters and low-pass filters, and determines the appropriate number of decomposition layers, and finally reorganizes the signal in combination with the wavelet coefficients of each layer and the calculated thresholds of each layer to achieve noise reduction and signal processing.

Description

Technical Field

[0001] The present invention relates to the field of infrasound monitoring, and in particular to a noise reduction processing method for an infrasound wave signal that is a precursor to bank collapse, and a feature recognition method and device. Background Art

[0002] After the infrasound signal is collected, it is difficult to judge the precursor infrasound signal of bank collapse, because the infrasound signal actually collected includes two types of component information: one is the characteristic signal, which directly or indirectly affects the research object to varying degrees, is the characteristic information part in signal processing, and is also the main part of the precursor infrasound of bank collapse; the other is the noise signal, which is the external interference irrelevant to the research object collected due to environmental influences, that is, the useless part. In order to better extract signal features, we should process the noise part. These "interferences" will cover up the characteristic signal, making it more hidden and difficult to be found. Therefore, in the actual key feature extraction process, it is often necessary to extract the infrasound signal for noise reduction processing in order to increase the reliability of signal recognition.

[0003] The working environment of bank collapse signal collection is complex and changeable. The infrasound signals extracted by sensors often contain a lot of interference noise, which will cover up the truly useful information. Therefore, before extracting the characteristic value of the bank collapse precursor infrasound signal, the original signal needs to be denoised.

[0004] The signal source carries useful information that we hope to transmit. However, during the signal change and transmission process, the superposition of noise and interference makes it difficult to identify the signal. To restore the useful signal carried, the superimposed noise and interference components in the signal must be removed. If the frequency of the noise is higher or lower than the effective signal, filtering is usually used to remove the noise. The burrs caused by interference can also be suppressed by smoothing the signal. The filtering method is a frequency domain processing method. When analyzing the frequency characteristics of the signal, the part with a small change rate of the signal corresponds to the low-frequency component, and the part with a large change rate corresponds to the high-frequency component. Using filtering to filter out the high-frequency part can remove the noise and make the signal smooth.

[0005] Filtering is the operation of filtering out specific frequency bands in a signal, and is an important measure to suppress and prevent interference. It is a probability theory and method for estimating another related random process based on the results of observing a random process.

[0006] Currently, the commonly used traditional noise reduction methods include Wiener filtering, Kalman filtering, wavelet filtering, low-pass filtering and other methods. These traditional methods will increase the entropy of the signal and make it difficult to characterize the non-stationary characteristics of the signal such as peak points and breakpoints. Summary of the invention

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] According to a first aspect of the present invention, a method for noise reduction processing of a precursor infrasound wave signal of bank collapse is provided, characterized in that the method comprises:

[0009] Obtain infrasound signals that are precursors to bank collapse;

[0010] Decomposing the bank collapse precursor infrasound wave signal into a plurality of wavelets according to a preset number of decomposition layers;

[0011] The corresponding wavelets are reorganized according to the wavelet coefficients and thresholds of each decomposition layer.

[0012] Furthermore, the precursor infrasonic wave signal of bank collapse is obtained by the following method:

[0013] The bank collapse precursor infrasound wave signal is obtained by testing the bank collapse infrasound physical model;

[0014] The bank collapse infrasound physical model includes a non-sliding body, a sliding body, a rigid wall, a steel pipe, a jack, an infrasound sensor and a pressure sensor. The sliding body is arranged above the non-sliding body, the sliding body has a slope surface, the steel pipe is inserted into the sliding body through the slope surface, the pressure sensor is arranged in the steel pipe, the rigid wall is arranged on one side of the non-sliding body, the jack is arranged on the rigid wall, the pressure sensor is arranged on one side of the sliding body, the top extension end of the jack is arranged corresponding to the pressure sensor, and when the jack is pushed out, the sliding body is pushed to slide on the non-sliding body through the pressure sensor.

[0015] Furthermore, decomposing the precursor infrasonic wave signal of bank collapse into a plurality of wavelets according to a preset number of decomposition layers specifically includes: decomposing the precursor infrasonic wave signal of bank collapse as an original signal into a first low-frequency signal and a second high-frequency signal, and continuing to decompose the first low-frequency signal and / or the second high-frequency signal as the original signal until a preset number of decomposition layers is reached.

[0016] Further, decomposing the bank collapse precursor infrasonic wave signal into a plurality of wavelets according to a preset number of decomposition layers specifically includes: converting the bank collapse precursor infrasonic wave signal into a digital signal f(m); based on the digital signal f(m), determining the wavelet corresponding to the decomposition of the bank collapse precursor infrasonic wave signal by the following formula:

[0017] c j+1 (n) = ∑ k∈Z h(k-2n)c j (k)

[0018] d j+1 (n) = ∑ k∈Z g(k-2n)c j (k)

[0019] In the formula, c j+1 (n) is the scale transformation coefficient of the j+1th layer, d j+1 (n) is the wavelet transform coefficient of the j+1th layer, k is a positive integer, j is the scale, h(n) and g(n) are two columns of conjugate filter coefficients determined by the wavelet function ψ(x);

[0020] The wavelet function ψ(x) is determined by the following formula:

[0021]

[0022] In the formula, g(k)=(-1) k h(1-k), is the scale function.

[0023] Furthermore, the threshold of each decomposition layer is determined by the following method: the absolute value of each number in the precursor infrasonic wave signal of bank collapse is arranged from large to small, and each absolute value is squared to obtain a new sequence f(k); the square root of the kth element with the threshold value f(k) is taken, and the risk Rish(k) generated by the threshold value is:

[0024]

[0025] Where j is the scale of wavelet transform, and N is the number of wavelet coefficients of the signal f(m) at the corresponding scale after wavelet transform;

[0026] Let the value corresponding to the minimum risk point be k min , determine the threshold

[0027] or

[0028] The threshold λ of each decomposition layer is determined by the following formula:

[0029]

[0030] Where N is the number of wavelet coefficients of the signal f(m) at the corresponding scale after wavelet transformation.

[0031] or

[0032] The threshold λ of each decomposition layer is determined by the following formula:

[0033]

[0034] Where N is the number of wavelet coefficients of the signal f(m) at the corresponding scale after wavelet transform.

[0035] Furthermore, the reorganization of the corresponding wavelet according to the wavelet coefficients and thresholds of each decomposition layer specifically includes: comparing the numerical value of the wavelet coefficient of each decomposition layer with the numerical value of the corresponding preset threshold; if the numerical value of the wavelet coefficient is less than the corresponding preset threshold, the numerical value of the wavelet coefficient is determined to be 0; if the numerical value of the wavelet coefficient is greater than or equal to the corresponding preset threshold, the numerical value of the wavelet coefficient is determined to remain unchanged or the numerical value of the wavelet coefficient is determined to be the difference between the numerical value of the wavelet coefficient and the corresponding preset threshold.

[0036] Furthermore, the step of reorganizing the corresponding wavelets according to the wavelet coefficients and thresholds of each decomposition layer also includes:

[0037] The wavelet is reorganized by the following formula:

[0038] c j (k) = H * c j+1 +G * d j+1 .

[0039] In the formula, c j (k) is the reconstructed signal of the jth layer, H * , G * is the rewrite of h(n) and g(n), two columns of conjugate filter coefficients determined by the wavelet function ψ(x), c j+1 is the scale transformation coefficient of the j+1th layer, d j+1 is the wavelet transform coefficient of the j+1th layer.

[0040] According to a second aspect of the present invention, a method for identifying features of infrasonic wave signals that are precursors to bank collapse is provided, the method comprising: extracting features of the infrasonic wave signals that are precursors to bank collapse after noise reduction processing according to any of the above methods in the time domain and the frequency domain, respectively, to obtain time domain feature parameters and frequency domain feature parameters; the time domain feature parameters at least include dimensioned parameters, the dimensioned parameters at least include one of the following: average value, effective value, variance, mean square value, peak value, mean square deviation, mean square error, root mean square error, kurtosis factor, peak factor, margin factor, waveform factor, and pulse factor, and combinations thereof; the frequency domain parameters at least include one of the following: center frequency, average frequency, frequency root mean square, and frequency standard deviation, and combinations thereof;

[0041] Constructing a time domain feature vector matrix and a frequency domain feature vector matrix respectively according to the time domain feature parameters and the frequency domain feature parameters;

[0042] The average value is calculated by the following formula Effective value x rms , variance σ 2 , mean square value Peak x max, mean square error σ, mean square error MSE, root mean square error RMSE, kurtosis factor x kurtosis , peak factor x peak , margin factor x margin , waveform factor x wave and the pulse factor x pulse :

[0043]

[0044]

[0045]

[0046]

[0047] x max =max|x(t)|

[0048]

[0049]

[0050]

[0051]

[0052]

[0053]

[0054]

[0055]

[0056] In the above formula, x i is the amount of bank collapse signals in the time series, and N is the total number of infrasound signals;

[0057] The center frequency F is calculated by the following formula center , average frequency F center , frequency RMS, and frequency standard deviation:

[0058]

[0059]

[0060]

[0061]

[0062] In the above formula, f(j) and X(j) are the frequency value and amplitude value of each point of the bank collapse signal after Fourier transform.

[0063] According to a third aspect of the present invention, a device for denoising a precursor infrasound wave signal of a bank collapse is provided, the device comprising: an acquisition unit configured to acquire a precursor infrasound wave signal of a bank collapse; a decomposition unit configured to decompose the precursor infrasound wave signal of a bank collapse into a plurality of wavelets according to a preset number of decomposition layers; and a recombination unit configured to recombine the corresponding wavelets according to the wavelet coefficients and thresholds of each decomposition layer.

[0064] Furthermore, the decomposition unit is further configured to decompose the precursor infrasound wave signal of bank collapse as an original signal into a first low-frequency signal and a second high-frequency signal, and continue to decompose the first low-frequency signal and / or the second high-frequency signal as the original signal until a preset number of decomposition layers is reached.

[0065] Furthermore, the decomposition unit is further configured to convert the bank collapse precursor infrasound wave signal into a digital signal f(m); based on the digital signal f(m), determine the wavelet corresponding to the decomposition of the bank collapse precursor infrasound wave signal by the following formula:

[0066] c j+1 (n) = ∑ k∈Z h(k-2n)c j (k)

[0067] d j+1 (n) = ∑ k∈Z g(k-2n)c j (k)

[0068] In the formula, c j+1 (n) is the scale transformation coefficient of the j+1th layer, d j+1 (n) is the wavelet transform coefficient of the j+1th layer, k is a positive integer, j is the scale, h(n) and g(n) are two columns of conjugate filter coefficients determined by the wavelet function ψ(x);

[0069] The wavelet function ψ(x) is determined by the following formula:

[0070]

[0071] In the formula, g(k)=(-1) k h(1-k), is the scale function.

[0072] Furthermore, the threshold of each decomposition layer is determined by the following method: the absolute value of each number in the precursor infrasonic wave signal of bank collapse is arranged from large to small, and each absolute value is squared to obtain a new sequence f(k); the square root of the kth element with the threshold value f(k) is taken, and the risk Rish(k) generated by the threshold value is:

[0073]

[0074] Where j is the scale of wavelet transform, and N is the number of wavelet coefficients of the signal f(m) at the corresponding scale after wavelet transform;

[0075] Let the value corresponding to the minimum risk point be k min , determine the threshold

[0076] or

[0077] The threshold λ of each decomposition layer is determined by the following formula:

[0078]

[0079] Where N is the number of wavelet coefficients of the signal f(m) at the corresponding scale after wavelet transformation.

[0080] or

[0081] The threshold λ of each decomposition layer is determined by the following formula:

[0082]

[0083] Where N is the number of wavelet coefficients of the signal f(m) at the corresponding scale after wavelet transform.

[0084] Furthermore, the reorganization unit is further configured to: compare the numerical value of the wavelet coefficient of each decomposition layer with the numerical value of the corresponding preset threshold; if the numerical value of the wavelet coefficient is less than the corresponding preset threshold, the numerical value of the wavelet coefficient is determined to be 0; if the numerical value of the wavelet coefficient is greater than or equal to the corresponding preset threshold, the numerical value of the wavelet coefficient is determined to remain unchanged or the numerical value of the wavelet coefficient is determined to be the difference between the numerical value of the wavelet coefficient and the corresponding preset threshold.

[0085] Further, the recombination unit is further configured as:

[0086] The wavelet is reorganized by the following formula:

[0087] c j (k) = H * c j+1 +G * d j+1 .

[0088] In the formula, c j (k) is the reconstructed signal of the jth layer, H * , G * is the rewrite of h(n) and g(n), two columns of conjugate filter coefficients determined by the wavelet function ψ(x), c j+1 is the scale transformation coefficient of the j+1th layer, d j+1 is the wavelet transform coefficient of the j+1th layer.

[0089] According to a fourth aspect of the present invention, there is provided a device for identifying characteristics of infrasound wave signals that are precursors to bank collapse, the device comprising a processor, the processor being configured to perform feature extraction in the time domain and the frequency domain on the infrasound wave signals that are precursors to bank collapse after noise reduction processing according to any of the above methods, respectively, to obtain time domain feature parameters and frequency domain feature parameters; the time domain feature parameters at least include dimensioned parameters, the dimensioned parameters at least include one of the following: average value, effective value, variance, mean square value, peak value, mean square deviation, mean square error, root mean square error, kurtosis factor, peak factor, margin factor, waveform factor and pulse factor, and combinations thereof; the frequency domain parameters at least include one of the following: center frequency, average frequency, frequency root mean square and frequency standard deviation, and combinations thereof;

[0090] Constructing a time domain feature vector matrix and a frequency domain feature vector matrix respectively according to the time domain feature parameters and the frequency domain feature parameters;

[0091] The average value is calculated by the following formula Effective value x rms , variance σ 2 , mean square value Peak x max , mean square error σ, mean square error MSE, root mean square error RMSE, kurtosis factor x kurtosis , peak factor x peak , margin factor x marqin , waveform factor x wave and the pulse factor x pulse :

[0092]

[0093]

[0094]

[0095]

[0096] x max =max|x(t)|

[0097]

[0098]

[0099]

[0100]

[0101]

[0102]

[0103]

[0104]

[0105] In the above formula, x i is the amount of bank collapse signals in the time series, and N is the total number of infrasound signals;

[0106] The center frequency F is calculated by the following formula center , average frequency F center , frequency RMS, and frequency standard deviation:

[0107]

[0108]

[0109]

[0110]

[0111] In the above formula, f(j) and X(j) are the frequency value and amplitude value of each point of the bank collapse signal after Fourier transform.

[0112] Compared with the prior art, the present invention has the following beneficial effects:

[0113] According to the method for denoising the precursor infrasound wave signal of bank collapse of the present invention, wavelet transform is performed on the noisy signal; the wavelet coefficients obtained by the transform are processed to remove the noise contained therein; and the processed wavelet coefficients are inversely transformed to obtain a denoised signal.

[0114] The essence of the present invention in the denoising problem is a function approximation problem, that is, how to find the best approximation of the original signal in the function space developed by the stretched and translated version of the wavelet mother function according to the proposed criterion to complete the distinction between the original signal and the noise signal. In other words, find the best mapping from the actual signal space to the wavelet function space to obtain the best restoration of the original signal.

[0115] From the perspective of signal science, wavelet denoising is a problem of signal filtering. The present invention is superior to the traditional low-pass filter in this respect because it can successfully retain the signal characteristics after denoising. BRIEF DESCRIPTION OF THE DRAWINGS

[0116] In the drawings, which are not necessarily drawn to scale, the same reference numerals may describe similar parts in different views. The same reference numerals with letter suffixes or different letter suffixes may represent different instances of similar parts. The accompanying drawings generally illustrate various embodiments by way of example and not limitation, and together with the specification and claims, are used to illustrate the disclosed embodiments. When appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the present apparatus or method.

[0117] Figure 1 It is a flow chart of a method for noise reduction processing of a precursory infrasonic wave signal of bank collapse in an embodiment of the present invention;

[0118] Figure 2 is a structural diagram of a bank collapse infrasound physical model in an embodiment of the present invention;

[0119] Figure 3 A schematic diagram of signal decomposition in an embodiment of the present invention;

[0120] Figure 4 A schematic diagram of signal decomposition in an embodiment of the present invention;

[0121] Figure 5 This is a frequency distribution diagram of infrasound events in a bank collapse model test in an embodiment of the present invention;

[0122] Figure 6a This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200805-1-3 from the south bank of Yuweizhou, Shishou City, Hubei Province;

[0123] Figure 6b This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200805-2-3 at the south bank of Yuweizhou, Shishou City, Hubei Province;

[0124] Figure 6c This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200805-3-3 from the south bank of Yuweizhou, Shishou City, Hubei Province;

[0125] Figure 6d This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200806-1-3 from the south bank of Yuweizhou, Shishou City, Hubei Province;

[0126] Figure 6eThis is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200806-2-3 from the south bank of Yuweizhou, Shishou City, Hubei Province;

[0127] Figure 6f This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200806-3-1 from the north bank of Ershengzhou, Jiangling County, Hubei Province;

[0128] Figure 6g This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200806-4-1 from the north bank of Ershengzhou, Jiangling County, Hubei Province;

[0129] Figure 6h This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200807-1-1 from the north bank of Ershengzhou, Jiangling County, Hubei Province;

[0130] Figure 6i This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200807-2-1 from the north bank of Ershengzhou, Jiangling County, Hubei Province;

[0131] Figure 6j This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200807-3-1 from the north bank of Ershengzhou, Jiangling County, Hubei Province;

[0132] Figure 6k This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200807-4-2 at the south bank of Ershengzhou, Gong'an County, Hubei Province;

[0133] Figure 6l This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200807-5-2 at the south bank of Ershengzhou, Gong'an County, Hubei Province;

[0134] Figure 6m This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200808-1-2 at the south bank of Ershengzhou, Gong'an County, Hubei Province;

[0135] Figure 6n This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200808-2-2 from the south bank of Ershengzhou, Gong'an County, Hubei Province;

[0136] Figure 6o This is the experimental acoustic-mechanical diagram in the embodiment of the present invention. The experimental soil sample was collected on 20200808-3-2 at the south bank of Ershengzhou, Gong'an County, Hubei Province;

[0137] Figure 7 It is an overall effect diagram of the embodiment of the present invention. DETAILED DESCRIPTION

[0138] The following examples are only for better explanation of the present invention, but the content of the present invention is not limited to the application in the examples. Therefore, those skilled in the art can make non-essential improvements and adjustments to the implementation scheme according to the above content of the invention and apply them to other examples, which are still within the protection scope of the present invention.

[0139] The inventors found in previous studies that the energy released during the deformation and rupture of rock and soil will inevitably generate elastic waves and sound waves propagating in the air. The "infrasonic wave signal that is a precursor to bank collapse" described in this article is the infrasonic frequency band of the sound wave. Infrasonic waves have the properties of low frequency and low attenuation. Therefore, infrasonic monitoring can be used as a technical means to determine bank collapse. However, since the collected infrasonic signals are mixed with noise signals, it is necessary to perform noise reduction processing on the infrasonic signals. Based on this, an embodiment of the present invention provides a noise reduction processing method for infrasonic wave signals that are a precursor to bank collapse, such as Figure 1 As shown, the method starts at step S100, obtaining an infrasonic wave signal that is a precursor to bank collapse.

[0140] In some embodiments, the bank collapse precursor infrasound wave signal is obtained by testing the bank collapse infrasound physical model 100 .

[0141] Soil samples are taken at different depths and locations on the embankment as raw materials for indoor experiments to identify infrasound characteristics that are precursors to bank collapse. Before a bank collapse occurs, there is a tensile cracking effect in the soil inside it, that is, before macroscopic deformation occurs, there is a cumulative effect of tensile strain inside it, which will cause the force chain between particles inside the soil to break, and then generate infrasound waves. Based on the principle of infrasound generation in this process, an embodiment of the present invention designs a bank collapse infrasound physical model 100, and intends to design an indoor experiment to simulate the tensile cracking process of the force chain of soil particles inside the soil. At the same time, a corresponding infrasound monitoring method is designed to collect the secondary signals generated by the tensile cracking process. Figure 2As shown, the bank collapse infrasound physical model 100 includes a non-sliding body 101, a sliding body 102, a rigid wall 103, a steel pipe 104, a jack 105, an infrasound sensor 106 and a pressure sensor 107. The sliding body 102 is arranged above the non-sliding body 101. The sliding body 102 has a slope surface 108. The steel pipe 104 is inserted into the sliding body 102 through the slope surface 108. The pressure sensor 107 is arranged in the steel pipe 104. The rigid wall 103 is arranged on one side of the non-sliding body 101, the jack 105 is arranged on the rigid wall 103, the pressure sensor 107 is arranged on one side of the sliding body 102, and the extending end of the jack 105 is arranged corresponding to the pressure sensor 107. When the jack 105 is pushed out, the sliding body 102 is pushed to slide on the non-sliding body 101 through the pressure sensor 107, and the contact surface between the non-sliding body 101 and the sliding body 102 is a slippery surface.

[0142] The experimental design is guided by the formation mechanism of bank collapse and the friction between soil particles at the crack surface as the root cause of infrasound. From a microscopic perspective, the soil near the crack surface is subject to the tensile force in the air. Before the collapse, local damage has already occurred at the crack surface, and local mutual extrusion, destruction and friction have occurred between particles. When the macroscopic erosion effect increases and the tensile force in the air reaches the maximum bonding force at the crack surface, the damage at the crack surface has been penetrated. Therefore, after the bank was exposed to the air, the crack surface inside it had already begun to be destroyed. With the increase of the tensile force in the air, the local shear damage at the crack surface has gradually increased, and the area of ​​particles involved in extrusion, crushing and friction will also increase, and the intensity of the precursory infrasound of bank collapse will also increase accordingly.

[0143] In the specific test, the jack 105 fixed on the rigid wall 103 is used to simulate the thrust, the pressure sensor 107 set between one side of the sliding body 102 and the jack 105 is used to monitor the change of the sliding force applied to the sliding body 102, and the infrasound sensor 106 placed inside the steel pipe 104 inserted into the sliding body 102 is used to monitor the infrasound signal generated when the bank collapses and slides. Although it is inconsistent with the effect of air cracking, the infrasound physical model 100 of bank collapse provided by the embodiment of the present invention can still detect the infrasound signal of particle crushing inside the slope, because its essence is the same: both are particle crushing at the sliding surface under shearing, and the corresponding emitted infrasound signal is also essentially the same.

[0144] As an example only, the embodiment of the present invention provides a test method for collecting the infrasound wave signal of the bank collapse precursor by the bank collapse infrasound physical model 100, and the method includes the following steps:

[0145] (1) Place the soil samples collected from the riverside and evenly distribute them on the test bench, and construct the following Figure 2 The non-sliding body 101 and the sliding body 102 are shown;

[0146] (2) Inserting an infrasound sensor into a regular soil sample on a test bench;

[0147] (3) Place the pressure sensor between the jack and the regular soil sample;

[0148] (4) Connecting the infrasound sensor and the pressure sensor to the computer;

[0149] (5) Record the start time of the test and open the data acquisition software to start the test;

[0150] (6) First, record the background noise for 30 seconds, and then start pressing the jack to push the upper regular soil sample until the soil sample is sheared and broken.

[0151] In step S200, the precursor infrasonic wave signal of bank collapse is decomposed into several wavelets according to the preset decomposition layer number. Wavelet is a special waveform with limited energy and length and an average value equal to 0. It is called wavelet because of its two characteristics of "small" and "wave". "Small" is because it has strong attenuation, is locally non-zero but will quickly decay to zero outside the center; "wave" is because its amplitude oscillates between positive and negative. The biggest difference between wavelet transform and Fourier transform is that it can not only see the details of the signal but also the overall picture of the signal. The theoretical system of wavelet transform is established using scaling and translation invariance.

[0152] The meaning of wavelet is to scale the most basic wavelet or mother wavelet ψ(t) by a and translate b to form a wavelet basis function ψ a,b (t).

[0153] The function f(t) has a clear practical meaning in R and its exponential can be integrated, so the set formed by this function is L 2 (R), that is: exists and makes sense, where f(t)∈L 2 (R). The continuous wavelet transform expression is as follows:

[0154]

[0155] Where α is the expansion factor, which is not zero; b is the translation factor. It can ensure that the wavelet can maintain energy normalization during the expansion and contraction process. The purpose of normalization is to make the sample data referential. Different parameter indicators often have different dimensions and dimensional units, which will affect the results of data analysis. In order to eliminate the dimensional influence between indicators and make the relative size of the data not very different, the comparability problem between parameter indicators can be solved. The normalization process can be achieved through Excel processing and computer programming.

[0156] Performing continuous wavelet transform on f(t) yields the following result:

[0157]

[0158] The “<>” in the formula represents the inner product, and the basic wavelet ψ(t) must satisfy the following condition:

[0159]

[0160] Among them C ψ Bounded.

[0161] According to the AdmissibleCondition condition, the mother wavelet ψ(t) will have oscillation characteristics and attenuation characteristics in the time domain, that is:

[0162]

[0163] From the above explanation, we can know that wavelet transform converts a one-dimensional signal in the time domain into a two-dimensional signal, because a two-dimensional signal is easier to decompose the time domain characteristics of the signal, that is, the time and frequency domains of a one-dimensional signal are mixed together, while a two-dimensional signal can more easily decompose the time and frequency characteristics of the signal.

[0164] In the continuous wavelet transform, the scale a, time t, and the translation factor b related to t are all continuous. However, under actual working conditions, the vibration signals collected are mostly in discrete form, so the discrete wavelet transform is more widely used. The discrete wavelet transform is not a discretization in time, but a discretization of the scale a and the translation factor b, which converts the scale a and the translation factor b into a power series form. Usually, it is assumed that a0>1, that is:

[0165]

[0166] Substituting the above formula into the continuous wavelet expression, we can get the expression of discrete wavelet transform:

[0167]

[0168] From the above expression, we can see that converting the scale a and the translation factor b into a power series is a very effective method, because the characteristics of the power series will cause a large change in the scale a when the exponent m changes slightly. The most common method currently used is to perform binary discretization on the scale and translation factors, that is, taking a0 = 2, b0 = 1, and the wavelet after binary discretization is:

[0169]

[0170] The binary wavelet can adjust the focal length when analyzing the signal. If you want to see the rougher content of the signal, you only need to increase m. If you want to observe more detailed content of the signal, you need to reduce m. All the information of the signal can be viewed by only changing a single value.

[0171] The embodiment of the present invention designs a method for performing discrete wavelet transform on a signal. The decomposition steps of the method are as follows:

[0172] Assume that the discrete sampling sequence of the digital signal f(m) is f(n), n = 1, 2..., N, and let f(n) be the approximate value of the signal when the scale j = 0, which is expressed as:

[0173] c0(n)=f(n)

[0174] Then the discrete wavelet transform of f(m) is:

[0175] c j+1 (n) = ∑ k∈Z h(k-2n)c j (k)

[0176] d j+1 (n) = ∑ k∈Z g(k-2n)c j (k)

[0177] Where: h(n) and g(n) are two columns of conjugate filter coefficients determined by the wavelet function ψ(x). At this time, the scaling function Determined by the crest factor:

[0178]

[0179] Where:

[0180] Fourier transform of h(k) satisfy

[0181]

[0182] Scaling function The Fourier transform of

[0183]

[0184] Correspondingly, the wavelet function ψ(x) is determined by

[0185]

[0186] Where: g(k) = (-1) k h(1-k).

[0187] The above method can be implemented as Figure 3 The decomposition process is shown. Figure 3 It can be clearly seen that the discrete wavelet only splits the low-frequency part of the signal at each layer of decomposition, but not the high-frequency part of the signal. However, in some cases, when judging a fault, not only the low-frequency part of the signal but also the high-frequency part of the signal is needed. In this case, the discrete wavelet cannot meet the requirements.

[0188] In the actual implementation process, according to the needs, you can also use Figure 4 The decomposition process shown in the figure can continue to decompose the undecomposed high-frequency part of the signal during the decomposition process, so that the information of the high-frequency part of the signal can be observed and studied, and the resolution of the target signal is comprehensively improved. Specifically, the original signal S is first decomposed into a low-frequency signal A1 and a high-frequency signal D1 by the first layer of wavelet packet decomposition; in the second layer of decomposition, the low-frequency signal A1 and the high-frequency signal D1 are decomposed into low-frequency signals AA2, DA2, AD2 and DD2 respectively according to the frequency domain width. The subscript represents the scale of decomposition, A represents the low-frequency segment, D represents the high-frequency segment, and the number in front indicates the number of decomposition layers, or the decomposition scale. When the decomposition scale is larger, the spatial resolution of the decomposed wavelet packet coefficients is lower.

[0189] After the signal is decomposed by wavelet packet, appropriate features are extracted from it. Commonly used wavelet packet decomposition features include the root mean square amplitude of the sub-band signal component after wavelet packet decomposition, wavelet packet energy, wavelet packet energy coefficient, and wavelet packet energy entropy.

[0190] Finally, in step S300, the corresponding wavelets are reorganized according to the wavelet coefficients and thresholds of each decomposition layer.

[0191] The embodiment of the present invention facilitates threshold denoising to quickly implement denoising processing on the signal. The embodiment of the present invention provides two threshold function selection methods, including soft threshold denoising and hard threshold denoising. The principle of soft threshold denoising is: when the absolute value of the wavelet coefficient is smaller than the set threshold, it is taken as 0; when it is larger than the given threshold, the difference between it and the threshold is taken, that is:

[0192]

[0193] The principle of hard threshold denoising is: when the absolute value of the wavelet coefficient is smaller than the set threshold, it is set to 0; when it is greater than the threshold, it is kept unchanged.

[0194]

[0195] The advantages of soft and hard threshold functions are compared as follows: The hard threshold function can better retain the end feature information of the signal. The wavelet coefficients obtained by the soft threshold function have good smoothness and avoid signal oscillation and jump. According to different signal characteristics, any of the above threshold functions can be selected for reorganization.

[0196] The following embodiments of the present invention will specifically describe how to select the threshold.

[0197] In the wavelet domain, the real vibration signal corresponds to a larger coefficient, and the noise part corresponds to a smaller coefficient. The coefficient corresponding to the noise in the wavelet domain still conforms to the Gaussian white noise distribution. The threshold selection rule is based on the model y=f(t)+e, where e is Gaussian white noise. Therefore, in order to select a threshold that can eliminate noise, it can be evaluated by wavelet coefficients or original signals. The present invention adopts three threshold selection methods, namely unbiased risk estimation, extreme value threshold estimation and fixed threshold estimation.

[0198] i) Unbiased risk estimation

[0199] Take the absolute value of each number in the signal S(i), rearrange them from small to large, and then square each absolute value to obtain a new sequence.

[0200] f(k)=(sort(|s|)) 2 (k=0,1…,N-1)#(4.2.33)

[0201] If the threshold is taken as the square root of the kth element of f(k), that is,

[0202]

[0203] The risk generated by this threshold is:

[0204]

[0205] According to the above risk curve Rish(k), let the value corresponding to the minimum risk point be k min , so rigrsure is defined as:

[0206]

[0207] ii) Fixed threshold

[0208]

[0209] iii) Maximum and minimum thresholds

[0210]

[0211] Generally speaking, extreme threshold and unbiased risk are relatively conservative. They can only better distinguish noise signals when there are not many noise components in the high-frequency band. Fixed threshold and heuristic threshold denoising are relatively complete and have significant denoising effects, but sometimes they may confuse useful information and noise and remove them together.

[0212] The following embodiment of the present invention will describe in detail how to reconstruct the signal of wavelet analysis.

[0213] In formula c j+1 (n) = ∑ k∈Z h(k-2n)c j (k) and formula d j+1 (n) = ∑ k∈Z g(k-2n)c j In (k), let H n,k =h k-2n , G n,k =g k-2n , then the matrix H=(H n,k ),G=(G n,k ). The discrete wavelet transform of f(m) can be expressed as

[0214]

[0215] Therefore, there is

[0216] c j (k) = H * c j+1 +G * d j+1

[0217] The above formula is the reorganization method of wavelet decomposition, and the reorganization method of wavelet packet is similar. The sequence after wavelet or wavelet packet decomposition is reconstructed using the above algorithm, and the original signal can be gradually restored. Each time the signal is decomposed, the amount of data is halved, that is, the time resolution of the signal is halved. However, each reconstruction will increase the amount of signal data exponentially, which can improve the time resolution of the signal. The process of decomposition and reconstruction can separate the signal from the noise, which is of great value in actual signal processing and analysis.

[0218] The embodiment of the present invention has explained the necessity of removing noise components from infrasound signals before feature extraction, and analyzed the principles and advantages of the wavelet transform method in signal denoising. Subsequently, the advantages and disadvantages of several threshold selection methods are compared and analyzed. The discrete wavelet transform used in the embodiment of the present invention performs time-frequency domain feature extraction. The discrete wavelet transform uses filters of different frequencies to analyze signals of different frequencies, mainly high-pass filters and low-pass filters, and determines the appropriate number of decomposition layers (the number of wavelet decomposition layers set in this experiment is 4 layers), and finally recombines the signal by combining the wavelet coefficients of each layer and the calculated thresholds of each layer. The low-pass filter used in this experiment is used to process low-frequency signals to achieve noise reduction and signal processing.

[0219] An embodiment of the present invention further provides a method for identifying features of infrasonic wave signals that are precursors to bank collapse. The method extracts features of the infrasonic wave signals that are precursors to bank collapse after noise reduction processing by the method described in any of the above embodiments in the time domain and frequency domain, respectively, to obtain time domain feature parameters and frequency domain feature parameters.

[0220] The extraction of time domain characteristic values ​​for the precursory infrasound signal of bank collapse is relatively simple. The commonly used time domain characteristic parameters are mainly divided into dimensional parameters and dimensionless parameters, among which dimensional parameters include:

[0221] (1) Average value: reflects the average amplitude of the infrasound signal before bank collapse. The calculation formula is as follows:

[0222]

[0223] (2) Effective value: Also known as the root mean square value, the root mean square is the arithmetic square root of the mean square value, reflecting the total energy of the signal. The calculation formula is as follows:

[0224]

[0225] (3) Variance: The average of the average of the differences between each sample and the average of all sample values. It represents the dynamic component of the signal and reflects the magnitude of the signal fluctuation and the degree of discreteness between signal data. The calculation formula is as follows:

[0226]

[0227] (4) Mean square value: The average of the signal squares, which represents the average energy of the signal and also indicates the strength of the signal. It reflects the vibration level of the equipment. The calculation formula is as follows:

[0228]

[0229] (5) Peak value: The maximum value of a signal within a certain period of time. The formula is as follows:

[0230] x max=max|x(t)|

[0231] (6) Mean square error: The arithmetic square root of the variance. The mean square error can reflect the discreteness of the signal data. The calculation formula is as follows:

[0232]

[0233] (7) Mean square error: It is a kind of variance. The mean square error refers to the mathematical expectation of the square of the difference between the parameter estimate and the true value of the parameter. If we regard the mathematical expectation E of the random variable as the parameter estimate (future) and the random variable itself as the true value of the parameter, then the root mean square is the ordinary error. The calculation formula is as follows:

[0234]

[0235] (8) Root mean square error: The root mean square error (RMSE) is the square root of the mean square error:

[0236]

[0237] In the process of time domain analysis of signals, many dimensionless indicators will appear, such as kurtosis factor, margin factor, peak factor, pulse factor, waveform factor, etc. They also play a great role in the analysis of infrasound signals.

[0238] (9) Kurtosis factor: Indicates the flatness of the waveform and is used to describe the distribution characteristics of the signal. The kurtosis of the normal distribution is equal to 3. When the kurtosis is less than 3, the distribution curve is relatively "flat", and when it is greater than 3, the distribution curve is relatively "steep". The calculation formula is as follows:

[0239]

[0240] (10) Crest factor: It is the ratio of the signal peak value to the effective value (RMS). It is a statistical indicator used to detect whether there is an impact in the signal. The calculation formula is as follows:

[0241]

[0242] (11) Margin factor: It is the ratio of the signal peak value to the square root amplitude. It can be used to detect the wear of mechanical equipment. The calculation formula is as follows:

[0243]

[0244] (12) Form factor: The ratio of the effective value to the rectified average value. The physical meaning can be understood as the ratio of DC current to AC current of equal power. Form factor = pulse factor / peak factor. The calculation formula is as follows:

[0245]

[0246] (13) Pulse factor: It is the ratio of the signal peak value to the rectified average value (average of the absolute value), and can also be used to detect whether there is an impact in the signal. The calculation formula is as follows:

[0247]

[0248] In the above formula, x i is the amount of bank collapse signal in the time series, N is the total number of infrasound signals, and according to the above calculation formula, the characteristic amount of the time domain angle can be calculated, thereby constructing the time domain characteristic vector matrix.

[0249] Generally speaking, the analysis of signal data from the time domain perspective is more intuitive, while the analysis from the frequency domain perspective is more concise, and observing the signal in the frequency domain makes the analysis of the problem more profound and convenient. The feature information that can be reflected by the time domain eigenvalue is relatively limited and cannot meet the needs of signal recognition research, so it is particularly important to extract feature quantities from the frequency domain perspective. Common feature parameters in the frequency domain include center frequency, average frequency, frequency root mean square, and frequency standard deviation, and the expressions are as follows:

[0250] (1) Center frequency:

[0251]

[0252] (2) Average frequency:

[0253]

[0254] (3) Frequency root mean square:

[0255]

[0256] (4) Frequency standard deviation:

[0257]

[0258] In the above formula, f(j) and X(j) are the frequency value and amplitude of each point of the bank collapse signal after Fourier transformation. Based on the above calculation formula, the relevant characteristic quantities of the frequency domain angle can be calculated, and the frequency domain characteristic vector matrix can be constructed.

[0259] The following embodiments of the present invention will be combined with specific experiments to further illustrate the feasibility and innovation of the present invention.

[0260] Based on the above-mentioned infrasound monitoring test for bank collapse precursors, a large number of infrasound signals generated during the bank collapse process were collected. Using the above-mentioned method for identifying the characteristics of the infrasound signals for bank collapse precursors, the time and frequency domain characteristics of the infrasound signals generated during the bank collapse process are shown in Tables 1 and 2. Based on these extracted time and frequency domain related characteristics of the infrasound signals for bank collapse precursors, sufficient data support can be provided for the accurate identification of the subsequent infrasound signals for bank collapse precursors.

[0261] Table 1 Analysis of time-frequency characteristics of infrasound signals before bank collapse

[0262] ID X-1 X-2 X-3 X-4 X-5 PZ YS SV FF QD 1 0.95 0.00 0.00 0.00 0.05 0.01 0.02 0.00 0.06 0.00 2 0.96 0.00 0.00 0.00 0.04 0.07 0.13 0.02 0.80 0.00 3 0.89 0.00 0.00 0.02 0.09 0.08 0.13 0.02 0.67 0.00 4 0.89 0.00 0.00 0.03 0.08 0.01 0.02 0.00 0.08 0.00 5 0.83 0.00 0.00 0.01 0.17 0.35 0.53 0.40 2.13 0.41 6 0.88 0.00 0.00 0.02 0.10 0.19 0.27 0.11 0.92 0.02 7 0.86 0.00 0.00 0.02 0.11 0.14 0.27 0.09 1.54 0.08 8 0.97 0.00 0.00 0.01 0.02 0.04 0.08 0.01 0.34 0.00 … … … … … … … … … … …

[0263] Table 2 Analysis of time-frequency characteristics of infrasound signals before bank collapse - continued

[0264] ID FZ YD PD MC BX FP CF PK PG EA 1 3.87 8.98 -1.18 5.99 1.55 1.40 2.42 3.59 4.32 0.06 2 6.00 16.96 -0.11 10.78 1.80 1.30 3.04 4.53 5.45 0.57 3 5.30 11.20 -0.65 8.44 1.59 1.70 2.90 4.56 5.41 0.51 4 5.13 12.88 -0.63 8.52 1.66 2.00 4.82 8.11 9.44 0.07 5 4.01 8.46 -1.28 6.10 1.52 1.50 2.98 3.80 4.83 2.40 6 3.48 5.99 -1.42 4.80 1.38 1.00 3.18 5.07 5.99 1.20 7 5.77 16.11 -1.57 10.80 1.87 1.80 2.96 3.98 4.96 1.07 8 4.26 21.94 -1.08 9.46 2.22 1.20 2.44 3.92 4.62 0.21 … … … … … … … … … … …

[0265] In Tables 1 and 2, X-1 = normalized wavelet energy coefficient (low frequency), X-2 = normalized wavelet energy coefficient (first layer high frequency); X-3 = normalized wavelet energy coefficient (second layer high frequency); X-4 = normalized wavelet energy coefficient (third layer high frequency); X-5 = normalized wavelet energy coefficient (fourth layer high frequency); PZ = average amplitude; YS = effective sound pressure; SV = amplitude variance; FF = peak-to-peak value; QD = kurtosis index; FZ = peak index; YD = margin index; PD = skewness index; MC = pulse index; BX = waveform index; FP = dominant frequency; CF = center frequency; PK = spectral line width; PG = frequency root mean square; EA = spectral energy average.

[0266] Table 3 Characteristics of infrasound signals before bank collapse

[0267]

[0268] Combination Figure 5 As shown in Table 3, the frequency distribution of soil particle crushing obeys the normal distribution law. The dominant frequency of soil particle crushing is mainly distributed between 0-2Hz, but the probability of the dominant frequency appearing between 0.5-1Hz is very high. From the perspective of discrete elements, the essence of particle cluster crushing is the same, so even if the soil is different, the dominant frequency is not much different; from the standard deviation value, the dominant frequency of soil particle cluster crushing infrasound has a low discreteness.

[0269] From the comprehensive results of acoustic mechanics (see Figure 6a -o), a peak of sound pressure will appear in the process of each group of in-situ soil destruction; before reaching the peak, the sound pressure signal is gradually enhanced; after reaching the peak, the sound pressure gradually weakens.

[0270] Overall (see Figure 7) Before the sliding force reaches its peak value (the anti-sliding force of the sliding body), the intensity of the infrasound signal increases synchronously with the sliding force; when the sliding force reaches its peak value, the sound pressure of the infrasound signal also reaches its peak value accordingly. From the simulation of the bank collapse scene, the gradual increase of the infrasound signal represents the gradual breakage of the particle cluster force chain on the bank slope, and the bank collapse disaster warning needs to issue a danger warning; the peak value of the infrasound signal means that the bank slope is in the stage of collapse, which represents the highest warning level of bank collapse.

[0271] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, a person of ordinary skill in the art can use other embodiments when reading the above description. In addition, in the above-mentioned specific embodiments, various features can be grouped together to simplify the present disclosure. This should not be interpreted as an intention that a disclosed feature that is not required to be protected is necessary for any claim. On the contrary, the subject matter of the present invention may be less than all the features of a specific disclosed embodiment. Thus, the following claims are incorporated into the specific embodiments as examples or embodiments, wherein each claim is independently used as a separate embodiment, and it is considered that these embodiments can be combined with each other in various combinations or arrangements. The scope of the present invention should be determined with reference to the attached claims and the full scope of equivalent forms granted by these claims.

Claims

1. A method for noise reduction of infrasonic wave signals before bank collapse, characterized in that: The method comprises: Obtain infrasound signals that are precursors to bank collapse; Decomposing the bank collapse precursor infrasound wave signal into a plurality of wavelets according to a preset number of decomposition layers; The corresponding wavelets are reorganized according to the wavelet coefficients and thresholds of each decomposition layer; Decomposing the bank collapse precursor infrasonic wave signal into a plurality of wavelets according to a preset number of decomposition layers specifically includes: Decomposing the precursor infrasound wave signal of bank collapse as an original signal into a first low-frequency signal and a second high-frequency signal, and continuing to decompose the first low-frequency signal and / or the second high-frequency signal as the original signal until a preset number of decomposition layers is reached; Decomposing the bank collapse precursor infrasonic wave signal into a plurality of wavelets according to a preset number of decomposition layers specifically includes: Converting the bank collapse precursor infrasound wave signal into a digital signal f(m); Based on the digital signal f(m), the wavelet corresponding to the decomposed infrasonic wave signal before bank collapse is determined by the following formula: c j+1 (n)=∑ k∈Z h(k-2n)c j (k) d j+1 (n)=∑ k∈Z g(k-2n)c j (k) In the formula, c j+1 (n) is the scaling coefficient of the j+1th scaling layer, d j+1 (n) is the wavelet transform coefficient of the j+1th scale layer, k is a positive integer, j is the scale, h(n) and g(n) are two columns of conjugate filter coefficients determined by the wavelet function ψ(x); The wavelet function ψ(x) is determined by the following formula: In the formula, g(k)=(-1) k h(1-k), is the scale function; The threshold of each decomposition layer is determined by the following method: Arrange the absolute values ​​of each number in the precursor infrasonic wave signal of bank collapse from large to small, and then square each absolute value to obtain a new sequence f(k); Taking the square root of the kth element of the threshold f(k), the risk Rish(k) generated by this threshold is: Where j is the scale of wavelet transform, and N is the number of wavelet coefficients of the signal f(m) at the corresponding scale after wavelet transform; Let the value corresponding to the minimum risk point be k min , determine the threshold or The threshold λ of each decomposition layer is determined by the following formula: Where N is the number of wavelet coefficients of the signal f(m) at the corresponding scale after wavelet transformation. or The threshold λ of each decomposition layer is determined by the following formula: Where N is the number of wavelet coefficients of the signal f(m) at the corresponding scale after wavelet transform.

2. The method according to claim 1, characterized in that The precursory infrasonic wave signal of bank collapse is obtained by the following method: The bank collapse precursor infrasound wave signal is obtained by testing the bank collapse infrasound physical model; The bank collapse infrasound physical model includes a non-sliding body, a sliding body, a rigid wall, a steel pipe, a jack, an infrasound sensor and a pressure sensor. The sliding body is arranged above the non-sliding body, the sliding body has a slope surface, the steel pipe is inserted into the sliding body through the slope surface, the pressure sensor is arranged in the steel pipe, the rigid wall is arranged on one side of the non-sliding body, the jack is arranged on the rigid wall, the pressure sensor is arranged on one side of the sliding body, the top extension end of the jack is arranged corresponding to the pressure sensor, and when the jack is pushed out, the sliding body is pushed to slide on the non-sliding body through the pressure sensor.

3. The method according to claim 1, characterized in that: The reorganization of the corresponding wavelet according to the wavelet coefficients and thresholds of each decomposition layer specifically includes: Compare the numerical value of the wavelet coefficient of each decomposition layer with the numerical value of the corresponding preset threshold. If the numerical value of the wavelet coefficient is less than the corresponding preset threshold, the numerical value of the wavelet coefficient is determined to be 0; if the numerical value of the wavelet coefficient is greater than or equal to the corresponding preset threshold, the numerical value of the wavelet coefficient is determined to remain unchanged or the numerical value of the wavelet coefficient is determined to be the difference between the numerical value of the wavelet coefficient and the corresponding preset threshold.

4. The method according to claim 3, characterized in that The reorganizing the corresponding wavelet according to the wavelet coefficients and thresholds of each decomposition layer also includes: The wavelet is reorganized by the following formula: c j (k)=H * c j+1 +G * d j+1 In the formula, c j (k) is the reconstructed signal of the jth layer, H * , G * is the rewrite of h(n) and g(n), two columns of conjugate filter coefficients determined by the wavelet function ψ(x), c j+1 is the scale transformation coefficient of the j+1th layer, d j+1 is the wavelet transform coefficient of the j+1th layer.

5. A method for identifying characteristics of infrasonic wave signals before bank collapse, characterized in that: The method comprises: The precursor infrasound wave signal of bank collapse after noise reduction processing according to the method described in any one of claims 1 to 4 is subjected to feature extraction in the time domain and the frequency domain respectively to obtain time domain feature parameters and frequency domain feature parameters; the time domain feature parameters at least include dimensioned parameters, and the dimensioned parameters at least include one of the mean value, effective value, variance, mean square value, peak value, mean square deviation, mean square error, root mean square error, kurtosis factor, peak factor, margin factor, waveform factor and pulse factor and their combinations; the frequency domain parameters at least include one of the center frequency, average frequency, frequency root mean square and frequency standard deviation and their combinations; Constructing a time domain feature vector matrix and a frequency domain feature vector matrix respectively according to the time domain feature parameters and the frequency domain feature parameters; The average value is calculated by the following formula Effective value x rms , variance σ 2 , mean square value Peak x max , mean square error σ, mean square error MSE, root mean square error RMSE, kurtosis factor x kurtosis , peak factor x peak , margin factor x margin , waveform factor x wave and the pulse factor x pulse : x max =max|x(t)| In the above formula, x i is the amount of bank collapse signals in the time series, and N is the total number of infrasound signals; The center frequency F is calculated by the following formula center , average frequency F center , frequency RMS, and frequency standard deviation: In the above formula, f(j) and X(j) are the frequency value and amplitude value of each point of the bank collapse signal after Fourier transform.

6. A device for noise reduction processing of infrasonic wave signals before bank collapse, characterized in that: The device comprises: An acquisition unit configured to acquire an infrasonic wave signal that is a precursor to bank collapse; A decomposition unit, configured to decompose the bank collapse precursor infrasound wave signal into a plurality of wavelets according to a preset number of decomposition layers; a reorganization unit configured to reorganize the corresponding wavelet according to the wavelet coefficients and thresholds of each decomposition layer; The decomposition unit is further configured as: Decomposing the precursor infrasound wave signal of bank collapse as an original signal into a first low-frequency signal and a second high-frequency signal, and continuing to decompose the first low-frequency signal and / or the second high-frequency signal as the original signal until a preset number of decomposition layers is reached; Converting the bank collapse precursor infrasound wave signal into a digital signal f(m); Based on the digital signal f(m), the wavelet corresponding to the decomposed infrasonic wave signal before bank collapse is determined by the following formula: c j+1 (n)=Σ k∈Z h(k-2n)c j (k) d j+1 (n)=Σ k∈Z g(k-2n)c j (k) In the formula, c j+1 (n) is the scaling coefficient of the j+1th scaling layer, d j+1 (n) is the wavelet transform coefficient of the j+1th scale layer, k is a positive integer, j is the scale, h(n) and g(n) are two columns of conjugate filter coefficients determined by the wavelet function ψ(x); The wavelet function ψ(x) is determined by the following formula: In the formula, g(k)=(-1) k h(1-k), is the scale function; The reorganization unit is further configured to determine the threshold of each decomposition layer by the following method: Arrange the absolute values ​​of each number in the precursor infrasonic wave signal of bank collapse from large to small, and then square each absolute value to obtain a new sequence f(k); Taking the square root of the kth element of the threshold f(k), the risk Rish(k) generated by this threshold is: Where j is the scale of wavelet transform, and N is the number of wavelet coefficients of the signal f(m) at the corresponding scale after wavelet transform; Let the value corresponding to the minimum risk point be k min , determine the threshold or The threshold λ of each decomposition layer is determined by the following formula: Where N is the number of wavelet coefficients of the signal f(m) at the corresponding scale after wavelet transformation. or The threshold λ of each decomposition layer is determined by the following formula: Where N is the number of wavelet coefficients of the signal f(m) at the corresponding scale after wavelet transform.

7. A device for identifying characteristics of infrasonic wave signals before bank collapse, characterized in that: The device includes a processor, and the processor is configured to perform feature extraction in the time domain and the frequency domain on the precursor infrasound wave signal of bank collapse after noise reduction processing according to the method of any one of claims 1 to 4, respectively, to obtain time domain feature parameters and frequency domain feature parameters; the time domain feature parameters at least include dimensioned parameters, and the dimensioned parameters at least include one of the following: average value, effective value, variance, mean square value, peak value, mean square deviation, mean square error, root mean square error, kurtosis factor, peak factor, margin factor, waveform factor and pulse factor, and a combination thereof; the frequency domain parameters at least include one of the following: center frequency, average frequency, frequency root mean square and frequency standard deviation, and a combination thereof; Constructing a time domain feature vector matrix and a frequency domain feature vector matrix respectively according to the time domain feature parameters and the frequency domain feature parameters; The average value is calculated by the following formula Effective value x rms , variance σ 2 , mean square value Peak x max , mean square error σ, mean square error MSE, root mean square error RMSE, kurtosis factor x kurtosis , peak factor x peak , margin factor x margin , waveform factor x wave and the pulse factor x pulse : x max =max|x(t)| In the above formula, x i is the amount of bank collapse signals in the time series, and N is the total number of infrasound signals; The center frequency F is calculated by the following formula center , average frequency F center , frequency RMS, and frequency standard deviation: In the above formula, f(j) and X(j) are the frequency value and amplitude value of each point of the bank collapse signal after Fourier transform.

Images