Signal-to-noise ratio improving method for long-distance optical fiber vibration detection

Abstract

The invention relates to a signal-to-noise ratio improvement method for long-distance optical fiber vibration detection, and relates to the technical field of long-distance optical fiber vibration signal-to-noise ratio improvement, and the method comprises the steps: obtaining differential data corresponding to each position and each time point of an optical fiber through distributed sound wave sensing equipment; performing integral conversion on the differential data of each position according to a time sequence to obtain vibration data, removing a direct current component through a Butterworth filter, and finally obtaining the vibration data of all positions of the optical fiber; forming a two-dimensional vibration waterfall image pixel matrix; utilizing two-dimensional variational mode decomposition to obtain mode components of different center frequencies; performing preliminary screening on all the obtained modal components according to the center frequency, and removing high-frequency modal components to obtain low-frequency modal components; and performing fast non-local mean denoising processing on each low-frequency mode component based on a logarithm-Cauchy adaptive weight kernel function, and reconstructing the processed mode components to finally obtain a long-distance optical fiber vibration signal with a remarkably improved signal-to-noise ratio.

Description

TECHNICAL FIELD

[0001] The present application is designed in the field of signal-to-noise ratio improvement of long-distance optical fiber vibration detection, and particularly relates to a vibration data collected by optical fiber sensing technology, which utilizes the space-time characteristics of the data, removes the high-frequency noise modal component through two-dimensional variational modal decomposition, and further denoises the remaining low-frequency modal component by using fast non-local mean, and then reconstructs, so as to realize the signal-to-noise ratio improvement of long-distance optical fiber vibration detection. BACKGROUND

[0002] In recent years, long-distance optical fiber vibration detection technology has been widely applied in the fields of submarine cable monitoring, earthquake warning and infrastructure health management. By using distributed optical fiber acoustic sensing equipment, vibration signals are collected along the line through optical fiber transmission, which can realize continuous monitoring of a large area. However, due to the influence of optical fiber tail end optical power attenuation, along-the-way loss and environmental noise, the signal-to-noise ratio is low in long-distance detection, which has been an important factor restricting the detection performance. In the prior art, integral transform, traditional filtering and simple frequency domain separation method are often used to process the collected vibration data, but these methods often cannot meet the requirements of denoising and retaining effective signals under complex space-time conditions, especially when high-frequency noise and low-frequency effective signals overlap.

[0003] In addition, the existing two-dimensional signal processing method (such as two-dimensional Fourier transform or wavelet transform) is difficult to accurately extract and denoise each frequency component in the vibration data when separating the modal of each frequency band due to the limitation of resolution and algorithm convergence. Therefore, it is urgent to develop a new algorithm to effectively suppress high-frequency noise modal and finely denoise low-frequency modal by fully utilizing the space-time characteristics of vibration data, so as to significantly improve the signal-to-noise ratio of long-distance optical fiber vibration detection. SUMMARY

[0004] The main purpose of the present application is to provide a long-distance optical fiber vibration detection signal-to-noise ratio improvement method, which proposes a comprehensive processing scheme based on two-dimensional signal decomposition and non-local mean denoising to solve the problems of signal attenuation at the tail of the optical fiber and insufficient noise suppression in the traditional method.

[0005] To achieve the above purpose, the present application provides a long-distance optical fiber vibration detection signal-to-noise ratio improvement method, which is characterized in that the method comprises the following steps: Step 1: using a distributed optical fiber acoustic sensing device to obtain differential data corresponding to each position and each time point of the optical fiber; Step 2: establishing a differential data set , defining as the differential data of the optical fiber at the position point at the time, and establishing a vibration data set ,definition For optical fiber Position point The vibration data at each moment is converted into vibration data by integrating the differential data of each position in time sequence, that is, , The DC component is removed by a Butterworth filter. The passband of the Butterworth filter is set to 20Hz, the stopband is set to 2Hz, the passband attenuation is less than 2dB, and the stopband attenuation is greater than 25dB. It is suitable for extracting high-frequency components above 20Hz while effectively suppressing low-frequency components below 2Hz. Finally, the vibration data set of all positions of the optical fiber is obtained. ; Step 3: Create a two-dimensional vibration waterfall pixel matrix ,definition , For optical fiber Position point The pixel value at the moment, normalize the vibration data to the range of 0 to 255: , in, Vibration data set The minimum value of Vibration data set The maximum value of The two-dimensional vibration waterfall graph pixel matrix is ​​constructed by taking the time point as the ordinate and the spatial position as the abscissa. ; Step 4: Define the modal component set ,in It is The modal components of the center frequency are obtained by using the two-dimensional vibration waterfall pixel matrix as input and using two-dimensional variational mode decomposition (2D-VMD) to obtain the modal component set of different center frequencies. :Sub-step 1. Set the initialization mode number , define the penalty factor and convergence threshold , using the time-space characteristics of optical fiber vibration data, the time point As the vertical axis, spatial position As the horizontal axis, a two-dimensional vibration waterfall chart matrix is ​​formed As the input of the two-dimensional variational mode decomposition; Sub-step 2. Perform fast Fourier transform (2D-FFT) on the input two-dimensional vibration signal matrix to obtain the frequency domain representation ,in, is the time frequency, is the spatial frequency, for each mode , its expression is updated in the frequency domain by the alternating direction multiplier method (ADMM), i.e. , wherein, is the frequency domain modal component, is the regularization term of the Lagrange multiplier, balancing the constraint condition, is the time center frequency, is the space center frequency; the center frequency of each mode is updated , i.e. , the center frequency of each mode is updated ; sub-step 3. updating the global Lagrange multiplier to enhance the convergence, i.e. , wherein is the update step size, here takes 0.25; sub-step 4. calculating the residual energy , i.e. , if the condition is met or the maximum number of iterations is reached, the iteration is terminated; otherwise, return to step 2; sub-step 5. performing inverse Fourier transform on each frequency domain modal to obtain the time-space domain modal component ; Step 5 defines a low-frequency modal component set , wherein is the modal component of the th center frequency, all obtained modal components are preliminarily screened according to the center frequency, according to experience, the real vibration signal generally does not exceed 200 Hz, so the modal components of the center frequency higher than 200 Hz are removed to obtain the low-frequency modal component set ; Step 6 defines a low-frequency modal component set , wherein is the modal component of the th center frequency after denoising based on the fast non-local mean (LC-FNLM) based on the log-Cauchy adaptive weight kernel function; each low-frequency modal component is respectively denoised by LC-FNLM: sub-step 1. setting the similar block radius , the block size is , setting the search window radius , indicating that the search range is the neighborhood of , the smoothing parameter is , the integral graph is selected as the acceleration method, and the low-frequency modal component As input; Sub-step 2. The integral map is used to quickly calculate the pixel sum or square sum of any rectangular area in the image, avoiding repeated traversal of pixels and pre-calculating the grayscale integral map Peace Law Scoreboard ,Right now , , Sub-step 3. For the target block and , use the integral graph to calculate the square of the Euclidean distance , target block The coordinates of the four corners of the matrix region are , candidate blocks The definition of is similar, the square of the Euclidean distance is: , in, , , , is the position of the four corner coordinates of the target block in the square integral graph, , , , and , , , is the position of the target block and the candidate block in the grayscale integral image; Sub-step 4. Calculate weights based on inter-block distances , where the weight kernel function , a new exponential-Log-Cauchy weight kernel function is proposed by combining the exponential weight kernel function with the Log-Cauchy weight kernel function. , in, , Controls the decay speed of the function tail; For each target pixel , normalize the weights of all its candidate blocks, that is, , in is the set of neighborhood pixels within the search window; Sub-step 4. For the target pixel The estimated values ​​of are weighted averaged, that is, , Traverse and process each mode to generate the denoised modal components ; Step 7: Define the final signal after reconstruction as The processed modal component is reconstructed, and the low-frequency modal component after noise reduction The long-distance fiber vibration signals with significantly improved signal-to-noise ratio are finally obtained by adding one by one , i.e. .

[0006] The present application uses a series of processing methods such as differential data integration, two-dimensional waterfall diagram construction, two-dimensional variational modal decomposition and LC-FNLM noise reduction to effectively separate and suppress high-frequency noise and fully retain low-frequency effective signals, thereby significantly improving the signal-to-noise ratio of long-distance fiber vibration detection. This method not only has a rigorous mathematical basis in theory, but also can overcome the problems of fiber tail signal attenuation and noise interference in practical application, and realizes high-sensitivity and low-false-alarm-rate vibration monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0007] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0008] Figure 1 is a flowchart of a signal-to-noise ratio improvement method for long-distance fiber vibration detection according to an embodiment of the present application; Figure 2 is an example interval waterfall diagram, wherein the red dashed line marks the vibration position, and the blue dashed line marks the noise interval; Figure 3 is a modal component obtained by two-dimensional variational modal decomposition; Figure 4 is a comparison of the low-frequency modal component obtained by screening and the respective LC-FNLM denoising waterfall diagram; Figure 5 is the original signal, 2D-VMD reconstructed signal and LC-FNLM denoising reconstructed waterfall diagram; Figure 6 is the original signal, 2D-VMD reconstructed signal and LC-FNLM denoising reconstructed 3D view. DETAILED DESCRIPTION

[0009] The present application will be described in detail below in combination with the drawings, tables and specific embodiments.

[0010] Embodiment: The present application proposes a signal-to-noise ratio improvement method for long-distance fiber vibration detection to solve the problems mentioned in the background art, which comprises the following steps: Step 1: Obtain the differential data corresponding to each position and each time point of the optical fiber using a distributed fiber optic acoustic sensing device; Step 2: Select audio 1 as the signal source to apply vibration at the tail of the light, and establish a differential data set , define as the differential data of the optical fiber at the position point at the time, and establish a vibration data set , define as the vibration data of the optical fiber at the position point at the time, and integrate the differential data at each position in time sequence to convert it into vibration data, i.e. , Pass through a Butterworth filter to remove the DC component, where the Butterworth filter passband is set to 20 Hz and the stopband is set to 2 Hz, the passband attenuation is less than 2 dB, and the stopband attenuation is greater than 25 dB, which is suitable for extracting high-frequency components above 20 Hz while effectively suppressing low-frequency components below 2 Hz; finally, the vibration data set of all positions of the optical fiber is obtained ; Step 3: Establish a two-dimensional vibration waterfall pixel matrix , define , as the pixel value of the optical fiber at the position point at the time, and normalize the vibration data to the interval 0~255:

[0011] where is the minimum value of the vibration data set , is the maximum value of the vibration data set ; According to the time point as the vertical coordinate and the spatial position as the horizontal coordinate, a two-dimensional vibration waterfall pixel matrix is formed ; Step 4: Define the modal component set , where is the modal component of the center frequency, and the two-dimensional vibration waterfall pixel matrix is used as input to obtain the modal component set of different center frequencies using two-dimensional variational modal decomposition (2D-VMD) : Sub-step 1. Set the initial number of modes , define the penalty factor and the convergence threshold , and use the time-space characteristics of the fiber vibration data to set the time point Spatial position as longitudinal axis Constituting a two-dimensional vibration waterfall matrix as transversal axis Input of two-dimensional variational modal decomposition; Sub-step 2. Perform a 2D-FFT on the input two-dimensional vibration signal matrix to obtain a frequency domain representation where, is the time frequency, is the spatial frequency, update each modal , in the frequency domain by alternating direction multiplier method (ADMM) to update its expression, i.e. , where, is the frequency domain modal component, is the regularization term of Lagrange multiplier, balancing the constraint condition, is the time center frequency, is the spatial center frequency; update the center frequency of each modal , i.e. , update in the same way; Sub-step 3. Update the global Lagrange multiplier to enhance the convergence, i.e. , where is the update step, here take 0.25; Sub-step 4. Calculate the residual energy , i.e. , if it meets or reaches the maximum number of iterations, terminate the iteration; otherwise return to step 2; Sub-step 5. Perform inverse Fourier transform on each frequency domain modal to obtain the modal component in the time-space domain ; Step 5 defines a low-frequency modal component set , where is the modal component of the center frequency, all obtained modal components are preliminarily screened according to the center frequency, according to experience, the real vibration signal generally does not exceed 200Hz, so the modal component of the center frequency higher than 200Hz is removed, and a low-frequency modal component set is obtained; Step 6 defines a low-frequency modal component set , where is the modal component of the a modal component of a center frequency after fast non-local means (LC-FNLM) denoising based on a log-Cauchy adaptive weight kernel function; each low-frequency modal component is respectively subjected to LC-FNLM denoising processing: substep 1. setting a similar block radius , a block size , setting a search window radius , indicating a neighborhood with a search range of , a smoothing parameter , an acceleration method selection integral graph, and a low-frequency modal component as input; substep 2. The integral graph is used to quickly calculate the sum or square sum of pixels in any rectangular region of the image, avoiding repeated traversal of pixels, and precomputing the gray integral graph and the flat integral graph , that is , , substep 3. The Euclidean distance square of the target block and is calculated using the integral graph , the matrix region four corner coordinates of the target block are , the definition of the candidate block is the same, and the Euclidean distance square is: , wherein , , , is the position of the four corner coordinates of the target block in the square integral graph , , , and , , , is the position of the target block and the candidate block in the gray integral graph; substep 4. Calculate the weight according to the distance between the blocks, wherein the weight kernel function , an exponential weight kernel function is combined with a Log-Cauchy weight kernel function to propose a new exponential-Log-Cauchy weight kernel function, that is , wherein , controls the decay speed of the function tail; for each target pixel , normalize the weight of all candidate blocks, that is , wherein is a set of neighborhood pixels within the search window; sub-step 4. a weighted average of the estimated value of the target pixel is performed, i.e. , each modality is processed to generate a denoised modality component ; Step 7 defines the final reconstructed signal as , the processed modality components are reconstructed, and the denoised low-frequency modality components are added one by one, and finally the long-distance fiber vibration signal with significantly improved signal-to-noise ratio is obtained , i.e. ; Step 8 can be analyzed from the accompanying drawings that six modality components with different center frequencies are obtained by two-dimensional variational modal decomposition, after screening, the fourth and fifth modality components are left, the fourth and fifth modality components are denoised by LC-FNLM, then reconstructed, and finally the signal-to-noise ratio is improved by 9.65dB.

[0012] Although the content of the present application has been described in detail through the above preferred embodiments, it should be recognized that the above description should not be considered as limiting the present application. After reading the above content, various modifications and replacements of the present application will be apparent to those skilled in the art. Therefore, the protection scope of the present application should be defined by the appended claims.

Claims

1. A method for improving the signal-to-noise ratio of long-distance optical fiber vibration detection, characterized in that: The steps include: Step 1: Use distributed fiber optic acoustic wave sensing equipment to obtain differential data corresponding to each position and time point of the optical fiber; Step 2: Create a differential data set ,definition For optical fiber Position point Differential data at each moment, creating a vibration data set ,definition For optical fiber Position point The vibration data at each moment is converted into vibration data by integrating the differential data of each position in time sequence, and then the DC component is removed by Butterworth filter to obtain the vibration data set of all positions of the optical fiber. ; Step 3: Create a two-dimensional vibration waterfall pixel matrix ,definition , For optical fiber Position point The pixel value at the moment, normalize the vibration data to the range of 0~255, and use the time point as the vertical coordinate and the spatial position as the horizontal coordinate to form a two-dimensional vibration waterfall graph pixel matrix ; Step 4: Define the modal component set ,in It is The modal components of the center frequency are obtained by using the two-dimensional vibration waterfall pixel matrix as input and using two-dimensional variational mode decomposition (2D-VMD) to obtain the modal component set of different center frequencies. ; Step 5: Define the low-frequency modal component set ,in It is The modal components of the center frequency are preliminarily screened according to the center frequency, and the high-frequency modal components are removed to obtain the low-frequency modal component set. ; Step 6: Define the low-frequency modal component set ,in It is The modal components of the center frequency are denoised by the fast non-local mean (LC-FNLM) based on the logarithmic-Cauchy adaptive weight kernel function; each low-frequency modal component is denoised by LC-FNLM to obtain a set of low-frequency modal components denoised by LC-FNLM. ; Step 7: Define the final signal after reconstruction as , reconstruct the processed modal components, and finally obtain a long-distance optical fiber vibration signal with significantly improved signal-to-noise ratio .

2. The method for improving the signal-to-noise ratio of long-distance optical fiber vibration detection according to claim 1, characterized in that: In step 4, two-dimensional variational modal decomposition is used to find the modal components of different center frequencies. The steps are as follows: Sub-step 1. Set the initialization mode number , define the penalty factor and convergence threshold , using the time-space characteristics of optical fiber vibration data, the time point As the vertical axis, spatial position As the horizontal axis, a two-dimensional vibration waterfall chart matrix is ​​formed As input to the 2D variational mode decomposition; Sub-step 2. Perform a fast Fourier transform (2D-FFT) on the input two-dimensional vibration signal matrix to obtain the frequency domain representation , for each mode , and update its expression in the frequency domain by the alternating direction method of multipliers (ADMM), namely ， Update the center frequency of each mode ,Right now ， Similar update ; Sub-step 3. Update the global Lagrange multiplier To enhance convergence, ， in is the update step size, here Take 0.25; Sub-step 4. Calculate the residual energy ,Right now ， If satisfied Or the maximum number of iterations is reached, the iteration is terminated; otherwise, return to step 2; sub-step 5. For each frequency domain mode Perform inverse Fourier transform to obtain the modal components in the time-space domain .

3. The method for improving the signal-to-noise ratio of long-distance optical fiber vibration detection according to claim 1, characterized in that: In step 6, the fast non-local means (LC-FNLM) based on the logarithmic-Cauchy adaptive weight kernel function is used to denoise the low-frequency modal components. The steps are as follows: Sub-step 1. Set the similarity block radius , the block size is , set the search window radius , indicating that the search range is Neighborhood, smoothing parameter , the acceleration method selects the integral diagram and converts the low frequency modal components As input; Sub-step 2. The integral map is used to quickly calculate the pixel sum or square sum of any rectangular area in the image, avoiding repeated traversal of pixels and pre-calculating the grayscale integral map Peace Law Scoreboard ,Right now ， ， Sub-step 3. For the target block and , use the integral graph to calculate the square of the Euclidean distance , target block The coordinates of the four corners of the matrix region are , candidate blocks The definition of is similar, the square of the Euclidean distance is: ， in, , , , is the position of the four corner coordinates of the target block in the square integral graph, , , , and , , , is the position of the target block and the candidate block in the grayscale integral image; Sub-step 4. Calculate weights based on inter-block distances , where the weight kernel function , a new exponential-Log-Cauchy weight kernel function is proposed by combining the exponential weight kernel function with the Log-Cauchy weight kernel function. ， in, , Controls the decay speed of the function tail; For each target pixel , normalize the weights of all its candidate blocks, that is, ， in is the set of neighborhood pixels within the search window; Sub-step 4. For the target pixel The estimated values ​​of are weighted averaged, that is, ， Traverse and process each mode to generate the denoised modal components .

Images