An Enhancement Method for Fiber Optic Disturbance Behavior Signals Based on Noise Reduction and Endpoint Detection

Abstract

This invention provides a method for enhancing fiber optic disturbance behavior signals based on noise reduction and endpoint detection. The enhancement steps include: spatiotemporal value filtering, EMD noise reduction, and endpoint detection. The input fiber optic signal is first subjected to spatiotemporal value filtering to eliminate pulse signal interference. Then, IMF components are obtained through EMD decomposition. Selective reconstruction is performed based on the correlation coefficient between the original signal and each IMF component to remove environmental noise. Finally, endpoint detection is used to extract the effective disturbance segment of the original signal over a long range. By removing a large number of normal behavior signals other than destructive behavior from the original signal, and using endpoint detection technology to locate the signals generated by destructive behavior, the extracted signal is purer and of higher quality, effectively improving the accuracy of security systems in identifying destructive behavior, and the time consumed is relatively small, effectively reducing time costs.

Description

Technical Field

[0001] This invention belongs to the field of optical fiber signal enhancement technology, and more specifically, it is a method for enhancing optical fiber disturbance behavior signals based on noise reduction and endpoint detection. Background Technology

[0002] Currently, many areas such as communities, airports, railways, and highways are facing threats from various acts of vandalism. Relying solely on personnel patrols is insufficient to meet security needs. Distributed fiber optic sensors, which exhibit signal changes when disturbed, are suitable for identifying various acts of vandalism in security systems. The advantages of this method are low cost, wide deployment range, and fast response speed.

[0003] However, noise in the environment can also cause changes in fiber optic signals. Therefore, high quality requirements are placed on the fiber optic disturbance signals. The key to accurately identifying various destructive behaviors lies in whether noise information can be removed from the existing fiber optic signals and the signals generated by disturbance behavior can be retained. When the quality of the fiber optic signal is low, noise information will be misidentified as destructive behavior, resulting in a large increase in the number of identified destructive behaviors. This requires users to check each destructive behavior one by one, increasing the user's workload.

[0004] Currently, there are few studies and methods for enhancing optical fiber disturbance behavior signals. Optical fibers are not only affected by various environmental noise information, but also by channel attenuation signals caused by long-distance propagation, which will affect the accuracy of subsequent damage behavior identification. In addition, damage behavior is usually not continuous, but ends instantaneously or after a period of time. If the identification device identifies all the signals of the optical fiber in real time every time, it will cause a lot of waste of hardware resources and the identification time will also increase significantly.

[0005] Therefore, this invention provides a method for enhancing fiber optic disturbance behavior signals based on noise reduction and endpoint detection, which improves the accuracy of subsequent destructive behavior identification while reducing hardware resources and time waste. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for enhancing fiber optic disturbance behavior signals based on noise reduction and endpoint detection, thereby solving the problems of low recognition rate and long recognition time of existing technologies for fiber optic damage behavior.

[0007] The purpose and effectiveness of the fiber optic perturbation behavior signal enhancement method based on noise reduction and endpoint detection in this invention are achieved by the following specific technical means:

[0008] A method for enhancing fiber optic perturbation behavior signals based on noise reduction and endpoint detection, comprising the following steps:

[0009] S1: The original fiber optic signal is preprocessed by a filter, and median filtering is performed on the spatiotemporal data to remove channel attenuation signals, thereby improving the recognition accuracy of subsequent steps. The technical challenge to be solved in this step is that the waveform, frequency and occurrence of channel attenuation signals are irregular. How to avoid affecting other signals and the signals generated by destructive behavior during the removal of channel attenuation signals?

[0010] S2: Then, EMD is used as a time-frequency domain processing method to directly decompose the signal without the need for pre-analysis domain research, and multiple intrinsic mode functions (IMFs) are obtained. Based on the correlation coefficient between the original signal and each IMF component, selective reconstruction is performed to obtain the enhanced fiber optic disturbance signal.

[0011] S3: Then, the spectral entropy method is used for endpoint detection. The fiber optic disturbance signal is filtered in the space-time of step S1 and EMD in step S2 to suppress channel attenuation and noise generated by the environment, respectively. The short-time energy and short-time spectral entropy of the fiber optic disturbance signal are used for endpoint detection. From the perspective of energy and information entropy, the signal range generated by the destructive behavior is extracted more accurately, and the influence of local interference is avoided as much as possible.

[0012] S4: Finally, by calculating the spectral entropy energy of each frame of the fiber optic perturbation behavior signal, the effective perturbation segment of the original signal over a longer range is extracted.

[0013] Preferably, in step S1, the spatiotemporal median filtering includes temporal median filtering and spatial median filtering.

[0014] Preferably, the energy change of the channel attenuation signal relative to the normal signal is large and the change rate is fast. Median filtering in the time dimension can remove the part of the channel attenuation signal with large energy. The median filtering in the time dimension is for the time domain data of the optical fiber in the same channel. A sliding window of length L1 is generated on the optical fiber data of each channel. In this task, L1 is set to 1s. The displacement of the window each time is D1. In this task, D1 is set to 0.5s. That is, the initial window covers the data from 0s to 1s. After sliding once, the window covers the data from 0.5 to 1s. While the sliding window moves, the median M1 of the signal energy covered by each window is calculated.

[0015] By setting the threshold T1 for median filtering in the time dimension (T1 is set to 5 in this task), the signal energy of each sliding window that exceeds T1×M1 is modified to the signal value corresponding to the median energy.

[0016] Preferably, when the fiber optic cable detects destructive behavior, it is not a single-point signal change, but rather a similar change occurs in the fiber optic signals of multiple channels over a distance. Channel attenuation signals are only generated on a single channel, and it is rare for adjacent channels to experience channel attenuation simultaneously. Therefore, after median filtering in time, median filtering is performed on adjacent channels in the spatial dimension to remove channel attenuation signals that differ significantly from those of adjacent channels.

[0017] The spatial dimension median filtering targets data from different channels of the optical fiber at the same time. A sliding window of length L2 is generated on the optical fiber data at each time. In this task, L2 is set to 5, covering 5 adjacent channels. The displacement of the window each time is D2. In this task, D2 is 2 channels. That is, the initial window covers data from channels 1 to 5, and after sliding once, the window covers data from channels 3 to 7. While the sliding window moves, the median M2 of the signal energy covered by each window is calculated.

[0018] By setting the threshold T2 for spatial median filtering (T2 is set to 2 in this task), the signal values ​​of each sliding window whose signal energy exceeds T2×M2 are modified to the signal values ​​corresponding to the median energy.

[0019] Preferably, in step S2, since there is a lot of environmental interference noise in the perturbation behavior signal detected by the optical fiber, this noise is random and non-stationary, which will seriously affect the feature extraction process, thus making the subsequent recognition model less robust; while traditional noise reduction methods (such as spectral subtraction) have great limitations and cannot effectively remove noise in the optical fiber; while EMD, as a time-frequency domain processing method, does not require pre-analysis of the signal domain and can directly decompose it to obtain multiple intrinsic mode functions (IMFs), each IMF component containing local feature signals of different time scales of the original signal;

[0020] The specific steps of the EMD time-frequency domain processing method are as follows:

[0021] C1: First, calculate all local extrema of the fiber optic perturbation signal x(t), and then fit the upper and lower envelopes of x(t) using a cubic spline interpolation function, denoted as xt and xt respectively. up (t) and x low (t); then calculate x. up (t) and x low The average envelope m(t) of x(t) is obtained by subtracting m(t) from x(t); the new data h1(t) is then obtained by subtracting m(t) from x(t).

[0022] C2: Then, determine whether h1(t) meets the IMF conditions:

[0023] First, in the entire dataset, the number of extreme values ​​and the number of zero crossings must be equal or no greater than 1;

[0024] Secondly, at any point, the mean of the envelope defined by the local maxima and local minima is zero;

[0025] C3: If the IMF condition is not met, then based on h1(t), repeat steps C1 and C2 until the condition is met. If h1(t) meets the condition, then it is the first IMF component, and subtract h1(t) from x(t) to obtain the remaining signal r1(t).

[0026] C4: Based on r1(t), repeat steps C1-C3 above until r n (t) is a monotonic function, and the mathematical expression for extracting the IMF component of the entire fiber optic disturbance signal is:

[0027] C5: Extract n IMF components, each containing the frequency components present in the original fiber optic disturbance signal. To remove noise and separate frequencies, calculate the similarity between each component and the original signal x(t). The original EMD algorithm usually uses the correlation coefficient r as the judgment criterion, but the large number of sampling points in the fiber optic signal will cause r to be too small. Therefore, Euclidean distance d is introduced, and the components with r greater than 0.5 and d less than 1 are taken as the reconstructed components. All the reconstructed components that meet the conditions are superimposed to obtain the enhanced fiber optic disturbance signal y(t).

[0028] Preferably, in step S3, since the fiber optic disturbance signal is suppressed by spatiotemporal mid-range filtering and EMD to suppress channel attenuation and environmental noise respectively, but the external destructive behavior usually occurs in a short period of time, most of the signal data detected by the fiber optic is under normal conditions. If the continuous signal is further processed or identified, it will not only occupy more storage space, but also increase the amount of computation, making real-time monitoring impossible. Therefore, endpoint detection is used to preprocess the signal data of each channel, extracting only the signal within the time period of the abnormality, and combining the short-time energy and short-time spectral entropy of the fiber optic disturbance signal for endpoint detection. From the perspective of energy and information entropy, the signal range generated by the destructive behavior can be extracted more accurately, which can avoid the influence of local interference as much as possible, thereby effectively obtaining the signal segment of the fiber optic disturbance behavior.

[0029] The specific steps for endpoint detection are as follows:

[0030] F1: Because the fiber optic disturbance signal exhibits short-time stationarity, the enhanced signal y(t) is framed using a Hamming window ω(n), with a window length of 512 and a window shift of 256. Then, the average frame energy of each frame is calculated using the following formula:

[0031] F2: When calculating the short-time frame spectral entropy, the enhanced signal y(t) is first subjected to a short-time Fourier transform (STFT), still using a Hamming window, with a window length of 512, a window shift of 256, and an FFT of 512. Here, FFT refers to Fast Fourier Transform, a general term for efficient and fast computation methods that utilize computers to calculate the Discrete Fourier Transform (DFT). Each frame of signal yields a 256-dimensional spectrum Y(ω), and then the spectral energy |Y(ω)| of each frame is calculated. 2 Then, the probability of each spectral component accounting for a portion of the total energy in each frame is calculated: Then, according to the information entropy formula: H(i) is the spectral entropy of the signal in each frame;

[0032] F3: Calculate the spectral entropy energy based on the average frame energy and spectral entropy of each frame of signal. Then calculate the HE(i) value of each frame of signal and compare it with the set threshold HE0, where HE0 is the spectral entropy energy measured under interference-free conditions. If it is greater than HE0, the frame data is retained; otherwise, the frame data is discarded.

[0033] F4: Finally, the frame data is reconstructed based on the following criteria: a. The duration of consecutive spliced ​​frames is greater than 2 seconds; b. If the interval between adjacent spliced ​​signals is less than 1 second, they are merged into a signal segment generated by the same action.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. This invention employs three parts: spatiotemporal mid-range filtering, EMD noise reduction, and endpoint detection. Compared to the original fiber optic signal, the enhanced fiber optic disturbance behavior signal removes a large number of normal behavior signals other than destructive behavior from the original signal. Endpoint detection technology is used to locate the signals generated by destructive behavior, resulting in a purer and higher-quality extracted signal. This effectively improves the accuracy of security systems in identifying destructive behavior, and the time required is relatively small, effectively reducing time costs. Attached Figure Description

[0036] Figure 1 This is a flowchart of the fiber optic perturbation behavior signal enhancement method of the present invention;

[0037] Figure 2 This is a comparison diagram of the time-domain signals of the mid-space filtering of this invention;

[0038] Figure 3 This is a diagram of each IMF component corresponding to the EMD decomposition of this invention;

[0039] Figure 4 This is a comparison diagram of signals before and after EMD noise reduction according to the present invention;

[0040] Figure 5 This is a schematic diagram of the endpoint detection of the knocking perturbation behavior signal based on spectral entropy energy according to the present invention. Detailed Implementation

[0041] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0042] This invention provides a method for enhancing fiber optic disturbance behavior signals based on noise reduction and endpoint detection. Taking the destructive behavior of continuous impact on an elevator car wall as an example, the sampling frequency of the impact signal detected by the fiber optic cable is 2500Hz. According to... Figure 1 The implementation flow of the enhancement method shown is as follows: The specific implementation method of the enhancement method for fiber optic disturbance behavior signals is as follows:

[0043] Step 1: First, perform spatiotemporal median filtering, which includes median filtering in the time dimension and median filtering in the spatial dimension;

[0044] (1) Time dimension median filtering is used for the time domain data of optical fibers in the same channel. Time median filtering uses a sliding window with a duration of 1s, which slides backward with an overlap of 0.5s each time, and calculates the median in each window, and then corrects it according to the threshold T1 (set to 5).

[0045] The specific process is as follows: A sliding window of length L1 is generated on the fiber data of each channel. In this task, L1 is set to 1s, and the displacement of the window each time is D1. In this task, D1 is set to 0.5s. That is, the initial window covers the data from 0s to 1s, and after sliding once, the window covers the data from 0.5 to 1s. While the sliding window is moving, the median M1 of the signal energy covered by each window is calculated.

[0046] Then, the threshold T1 of the median filtering in the time dimension is set for correction. In this task, T1 is set to 5. The signal energy of each sliding window that exceeds the value of T1×M1 is modified to the signal value corresponding to the median energy, thereby obtaining the median in each window.

[0047] (2) Spatial median filtering targets data from different channels of the optical fiber at the same time. By using spatial median filtering, a sliding window with a length of 5 sampling points and a width of 5 channels is used, which still overlaps and slides, and the median of the values ​​within each coverage area is calculated and corrected according to the threshold T2 (set to 2).

[0048] The specific process is as follows: A sliding window of length L2 is generated on the fiber data at each time step. In this task, L2 is set to 5, covering 5 adjacent channels. The displacement of the window each time is D2. In this task, D2 is 2 channels. That is, the initial window covers the data of channels 1-5, and after sliding once, the window covers the data of channels 3-7. While the sliding window moves, the median M2 of the signal energy covered by each window is calculated.

[0049] By setting the threshold T2 for spatial median filtering (T2 is set to 2 in this task), the signal values ​​of each sliding window whose signal energy exceeds T2×M2 are modified to the signal values ​​corresponding to the median energy.

[0050] Through the above steps, as Figure 2 As shown in the figure, the time-domain signal comparison diagram of spatiotemporal mid-value filtering shows that after spatiotemporal mid-value filtering, the channel attenuation signal is effectively removed. This solves the problem of how to remove the channel attenuation signal while minimizing its impact on other signals and the signals generated by the destructive behavior when the waveform, frequency and occurrence of the channel attenuation signal lack regularity. This provides a basic guarantee for the subsequent enhancement of fiber optic disturbance signals and improves the recognition accuracy of fiber optic disturbance signals.

[0051] The second step is to perform Empirical Mode Decomposition (EMD) denoising. Since there is a lot of environmental interference noise in the perturbation behavior signal of fiber detection, this noise is random and non-stationary, which will seriously affect the feature extraction process and make the subsequent recognition model less robust. Traditional denoising methods (such as spectral subtraction) have great limitations and cannot effectively remove noise in the fiber.

[0052] By using EMD as a time-frequency domain processing method, there is no need to conduct pre-analysis domain research on the signal. The modified continuous knocking behavior signal on the elevator car wall can be directly decomposed into EMD, generating a total of 11 different intrinsic mode function (IMF) components. Each IMF component contains local feature signals of different time scales of the original signal.

[0053] like Figure 3 As shown, this is a diagram of the IMF components corresponding to EMD decomposition. It can be seen that the first few components mainly contain a large number of high-frequency signals, which are primarily caused by environmental noise. Therefore, the correlation coefficient is calculated for each component, components with a correlation coefficient less than 0.5 are filtered out, and the remaining components are summed to generate a noise-reduced and enhanced fiber optic striking behavior signal, as shown below. Figure 4 The image shows a time-domain comparison of the original and denoised tapping signals.

[0054] The specific method for calculating the correlation coefficient between the IMF component and the original signal is as follows:

[0055] (1) First, calculate all local extreme points of the fiber optic disturbance signal x(t), and use the cubic spline interpolation function to fit the upper and lower envelopes of x(t), denoted as xup(t) and xlow(t) respectively; then calculate the average envelope m(t) of xup(t) and xlow(t); then subtract m(t) from x(t) to obtain the new data h1(t);

[0056] (2) Then, determine whether h1(t) meets the IMF conditions:

[0057] a. In the entire dataset, the number of extreme values ​​and the number of zero crossings must be equal or not greater than 1;

[0058] b. At any point, the mean of the envelope defined by the local maxima and local minima is zero;

[0059] (3) If the IMF condition is not met, then based on h1(t), repeat steps C1 and C2 until the condition is met. If h1(t) meets the condition, then it is the first IMF component, and subtract h1(t) from x(t) to obtain the remaining signal r1(t).

[0060] (4) Based on r1(t), repeat steps C1-C3 above until r n (t) is a monotonic function, and the mathematical expression for extracting the IMF component of the entire fiber optic disturbance signal is:

[0061] (5) Extract n corresponding IMF components, each of which contains the frequency components present in the original fiber optic disturbance signal. In order to remove noise frequency separation, calculate the similarity between each component and the original signal x(t). The original EMD algorithm usually uses the correlation coefficient r as the judgment basis. However, there are many sampling points of the fiber optic signal, which will cause r to be too small. Therefore, the Euclidean distance d is introduced, and the components with r greater than 0.5 and d less than 1 are taken as the reconstruction components. All the reconstruction components that meet the conditions are added together to obtain the noise-reduced and enhanced fiber optic disturbance signal y(t).

[0062] The third step is to use the spectral entropy method for endpoint detection to extract the signal range generated by the destructive behavior more accurately and to avoid the influence of local interference as much as possible.

[0063] After spatiotemporal mid-range filtering and EMD decomposition, the knocking disturbance signal has suppressed most of the interference signals. However, since external destructive behavior usually occurs within a short period of time, most of the signal data detected by the fiber optic cable is signal data under normal conditions. If this continuous signal is further processed or identified, it will not only occupy more storage space but also increase the amount of computation, making real-time monitoring impossible. Therefore, endpoint detection is used to preprocess the signal data of each channel, extracting only the signal within the time period of the anomaly. By combining the short-time energy and short-time spectral entropy of the fiber optic disturbance signal for endpoint detection, the signal range generated by the destructive behavior can be extracted more accurately from the perspective of energy and information entropy, minimizing the impact of local interference.

[0064] First, a Hamming window with a window length of 512 and a window shift of 256 is used to perform frame processing on the enhanced tapping behavior signal. Then, the average frame energy E is calculated for each frame. n (i) and spectral entropy H(i) (i=1,2,...,N / 2-1), then calculate the spectral entropy energy HE(i) of each frame of signal, and at the same time calculate the spectral entropy energy HE0 under the condition of no disturbance in the same way. Using HE0 as the threshold, each frame of HE(i) is judged, and finally spliced ​​according to the rules to select the effective disturbance signal segment.

[0065] The specific calculation steps for endpoint detection using the spectral entropy method are as follows:

[0066] (1) First, based on the short-time stationary characteristic of the fiber optic disturbance signal, the enhanced signal y(t) is framed with a Hamming window ω(n), where the window length is 512 and the window shift is 256; then, the average frame energy of each frame is calculated using the following formula:

[0067] In digital signal processing, each FFT transform can only transform a finite length of time-domain data. Therefore, it is necessary to truncate the time-domain signal. Even for periodic signals, if the truncated time length is not an integer multiple of the period (periodic truncation), the truncated signal will have leakage. To minimize this leakage error, we need to use a weighting function, also called a window function. Windowing is mainly to make the time-domain signal seem to better meet the periodicity requirements of FFT processing and reduce leakage. However, since directly truncating the signal (using a rectangular window) will cause frequency leakage, in order to improve the frequency leakage situation, a non-rectangular window is usually used, generally a Hamming window, because the amplitude-frequency characteristic of the Hamming window is that the sidelobe attenuation is large, and the attenuation between the main lobe peak and the first sidelobe peak can reach 40dB.

[0068] The principle formula of the Hanming window is:

[0069]

[0070] When a0 = 0.53836, it is called a Hamming window; when a0 = 0.5, it is called a Hann window.

[0071] The Hann window, also known as the raised cosine window, can be viewed as the sum of the spectra of three rectangular time windows, or the sum of three sinc(t) type functions. The two terms in parentheses are shifted to the left and right by π / T relative to the first spectral window, thereby canceling out the side lobes and eliminating high-frequency interference and energy leakage.

[0072] From the perspective of reducing leakage, the Hann window is superior to the rectangular window, but the main lobe of the Hann window is widened, which is equivalent to widening the analysis bandwidth and reducing the frequency resolution.

[0073] If we set a0 to a value close to 0.53836, or more precisely 25 / 46, we will get the Hamming window. The purpose of setting this value is to generate a zero intersection at a frequency of 5π / (N-1), so that the first side lobe of the original Hann window can be greatly eliminated, producing a side lobe with only 1 / 5 the height of the Hann window.

[0074] (2) When calculating the short-time frame spectral entropy, the enhanced signal y(t) is first subjected to a short-time Fourier transform (STFT), still using a Hamming window, with a window length of 512, a window shift of 256, and an FFT of 512; each frame of the signal yields a spectrum Y(ω) with a dimension of 256, and then the spectral energy |Y(w)| of each frame of the signal is calculated. 2 Then, the probability of each spectral component accounting for a portion of the total energy in each frame is calculated: Then, according to the information entropy formula: H(i) is the spectral entropy of the signal in each frame;

[0075] (3) Then, based on the average frame energy and spectral entropy of each frame obtained in (1) and (2), the spectral entropy energy is calculated. Next, the HE(i) value of each frame of signal is calculated and compared with the set threshold HE0, where HE0 is the spectral entropy energy measured under interference-free conditions. If it is greater than HE0, the frame data is retained; otherwise, the frame data is discarded.

[0076] (4) Finally, the frame data is reconstructed by judgment. The judgment criteria are: a. The duration of consecutive frames after splicing is greater than 2 seconds; b. If the interval between adjacent spliced ​​signals is less than 1 second, they are merged into a signal segment generated by the same behavior.

[0077] Step 4: By calculating the spectral entropy energy of each frame of the fiber optic perturbation behavior signal, the effective perturbation segment is extracted from the original signal over a longer range, thereby improving the recognition rate of the fiber optic perturbation signal and reducing the time required for system recognition; for example... Figure 5As shown, this is a schematic diagram of endpoint detection of the knocking disturbance signal based on spectral entropy energy. It can be seen that the spectral entropy method for endpoint detection can extract the signal range generated by the destructive behavior more accurately, effectively reducing the storage space occupied and the related computational load.

[0078] In summary, the fiber optic perturbation behavior signal enhancement process based on noise reduction and endpoint detection is as follows: Figure 1 As shown, it mainly consists of three parts: spatiotemporal mid-value filtering, EMD noise reduction, and endpoint detection. Spatiotemporal mid-value filtering and EMD noise reduction are noise reduction processes performed on the optical fiber signal, while endpoint detection is used to extract optical fiber disturbance behavior segments.

[0079] The input fiber optic signal is first subjected to spatiotemporal mid-range filtering to eliminate pulse interference caused by fiber channel attenuation. Then, IMF components are obtained through EMD decomposition. Selective reconstruction is performed based on the correlation coefficient between the original signal and each IMF component to remove environmental interference noise and obtain an enhanced fiber optic perturbation signal. Finally, by calculating the spectral entropy energy of each frame of the fiber optic perturbation behavior signal, effective perturbation segments of the original signal over a long range are extracted. Through the above method, other interference signals in the fiber optic perturbation signal are effectively removed, making the detection time required by this invention relatively shorter than that of existing technologies, thereby reducing the time cost. Furthermore, using EMD noise reduction for fiber optic perturbation signal enhancement is a method that has not yet been tried in this field, and the introduction of Euclidean distance in similarity makes the selection of reconstructed components more accurate.

[0080] The embodiments of the present invention are given for the purposes of illustration and description, and are not intended to be exhaustive or to limit the invention to the forms disclosed.

Claims

1. A method for enhancing fiber optic perturbation behavior signals based on noise reduction and endpoint detection, characterized in that, The steps of the method for enhancing the fiber optic perturbation behavior signal are as follows: S1: The original optical fiber signal is preprocessed by a filter, and median filtering is performed on the spatiotemporal data to remove channel attenuation signals, thereby improving the recognition accuracy of subsequent steps. S2: Next, EMD is used as a time-frequency domain processing method to directly decompose the signal without the need for pre-analysis domain research, and multiple intrinsic mode functions (IMFs) are obtained. Based on the correlation coefficient between the original signal and each IMF component, selective reconstruction is performed to obtain an enhanced fiber optic disturbance signal, which eliminates the influence of interference signals for subsequent endpoint detection using the spectral entropy method. S3: Then, the spectral entropy method is used for endpoint detection. The fiber optic disturbance signal is filtered in the space-time of step S1 and EMD in step S2 to suppress channel attenuation and noise generated by the environment, respectively. The short-time energy and short-time spectral entropy of the fiber optic disturbance signal are used for endpoint detection. From the perspective of energy and information entropy, the signal range generated by the destructive behavior is extracted more accurately, and the influence of local interference is avoided as much as possible. S4: Finally, by calculating the spectral entropy energy of each frame of the fiber optic perturbation behavior signal, the effective perturbation segment of the original signal over a longer range is extracted.

2. The method for enhancing fiber optic perturbation behavior signals based on noise reduction and endpoint detection as described in claim 1, characterized in that: In step S1, the spatiotemporal median filtering includes median filtering in the time dimension and median filtering in the spatial dimension.

3. The method for enhancing fiber optic perturbation behavior signals based on noise reduction and endpoint detection as described in claim 2, characterized in that: The time-dimensional median filtering is applied to the time-domain data of the same optical fiber channel. A sliding window of length L1 is generated on the optical fiber data of each channel. In this task, L1 is set to 1s, and the displacement of the window each time is D1. In this task, D1 is set to 0.5s, that is, the initial window covers the data from 0s to 1s, and after sliding once, the window covers the data from 0.5 to 1s. While the sliding window moves, the median M1 of the signal energy covered by each window is calculated. By setting the threshold T1 for median filtering in the time dimension (T1 is set to 5 in this task), the signal energy of each sliding window that exceeds T1×M1 is modified to the signal value corresponding to the median energy.

4. The method for enhancing fiber optic perturbation behavior signals based on noise reduction and endpoint detection as described in claim 2, characterized in that: The spatial dimension median filtering targets data from different channels of the optical fiber at the same time. A sliding window of length L2 is generated on the optical fiber data at each time. In this task, L2 is set to 5, covering 5 adjacent channels. The displacement of the window each time is D2. In this task, D2 is 2 channels. That is, the initial window covers data from channels 1 to 5, and after sliding once, the window covers data from channels 3 to 7. While the sliding window moves, the median M2 of the signal energy covered by each window is calculated. By setting the threshold T2 for spatial median filtering (T2 is set to 2 in this task), the signal values ​​of each sliding window whose signal energy exceeds T2×M2 are modified to the signal values ​​corresponding to the median energy.

5. The method for enhancing fiber optic perturbation behavior signals based on noise reduction and endpoint detection as described in claim 1, characterized in that: In step S2, the specific steps of the EMD time-frequency domain processing method are as follows: C1: First, calculate all local extrema of the fiber optic perturbation signal x(t), and then fit the upper and lower envelopes of x(t) using a cubic spline interpolation function, denoted as xt and xt respectively. up (t) and x low (t); then calculate x. up (t) and x low The average envelope m(t) of x(t) is obtained by subtracting m(t) from x(t); the new data h1(t) is then obtained by subtracting m(t) from x(t). C2: Then, determine whether h1(t) meets the IMF conditions: First, in the entire dataset, the number of extreme values ​​and the number of zero crossings must be equal or no greater than 1; Secondly, at any point, the mean of the envelope defined by the local maxima and local minima is zero; C3: If the IMF condition is not met, then based on h1(t), repeat steps C1 and C2 until the condition is met. If h1(t) meets the condition, then it is the first IMF component, and subtract h1(t) from x(t) to obtain the remaining signal r1(t). C4: Based on r1(t), repeat steps C1-C3 above until r n (t) is a monotonic function, and the mathematical expression for extracting the IMF component of the entire fiber optic disturbance signal is: C5: Extract n IMF components, each of which contains the frequency components present in the original fiber optic disturbance signal. In order to remove noise and separate frequencies, calculate the similarity between each component and the original signal x(t). Take the components with a correlation coefficient r greater than 0.5 and an Euclidean distance d less than 1 as the reconstructed components. Superimpose all the reconstructed components that meet the conditions to obtain the enhanced fiber optic disturbance signal y(t).

6. The method for enhancing fiber optic perturbation behavior signals based on noise reduction and endpoint detection as described in claim 5, characterized in that: In step S3, the specific steps for endpoint detection are as follows: F1: Because the fiber optic disturbance signal exhibits short-time stationarity, the enhanced fiber optic disturbance signal y(t) is framed using a Hamming window ω(n), with a window length of 512 and a window shift of 256. Then, the average frame energy of each frame is calculated using the following formula: F2: When calculating the short-time frame spectral entropy, the enhanced fiber disturbance signal y(t) is first subjected to a short-time Fourier transform, still using a Hamming window, with a window length of 512, a window shift of 256, and an FFT of 512; each frame of signal yields a spectrum Y(ω) with a dimension of 256, and then the spectral energy |Y(w|) of each frame of signal is calculated. 2 Then, the probability of each spectral component accounting for a portion of the total energy in each frame is calculated: Then, according to the information entropy formula: H(i) is the spectral entropy of the signal in each frame; F3: Calculate the spectral entropy energy based on the average frame energy and spectral entropy of each frame of signal. Then calculate the HE(i) value of each frame of signal and compare it with the set threshold HE0, where HE0 is the spectral entropy energy measured under interference-free conditions. If it is greater than HE0, the frame data is retained; otherwise, the frame data is discarded. F4: Finally, the frame data is reconstructed based on the judgment.

Images