A method, system, electronic device, and storage medium for detecting fiber optic vibration events.

By extracting and stitching time-domain and frequency-domain features from optical fiber vibration signals, and combining them with an event recognition model, the problem of low accuracy in optical fiber vibration event recognition in existing technologies has been solved, achieving higher detection accuracy.

CN123111448APending Publication Date: 2026-11-17CHINA RAILWAY ELECTRIFICATION ENGINEERING GROUP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611456997.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-18
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing distributed fiber optic vibration sensing technology lacks deep fusion and structured dimensionality reduction processing of time and frequency features, resulting in low accuracy in vibration event identification and detection.

Method used

By acquiring the time-domain signal of the preset sampling points of the optical fiber, dividing it into frequency sub-signals, extracting time-domain and frequency-domain features, concatenating them into a time-frequency feature vector, and processing it in conjunction with a pre-trained event recognition model, feature redundancy is reduced and noise interference is removed.

Benefits of technology

It significantly improves the accuracy of fiber optic vibration event identification and detection, overcoming the problems of information loss and information fragmentation caused by single-dimensional feature analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123111448A_ABST
    Figure CN123111448A_ABST
Patent Text Reader

Abstract

This application relates to the field of fiber optic operation and maintenance, specifically to a method, system, electronic device, and storage medium for detecting fiber optic vibration events. The method includes: acquiring time-domain signals from preset sampling points on the fiber under test; dividing the time-domain signals into several frequency sub-signals; extracting time-domain features from the frequency sub-signals to obtain time-domain feature vectors, and extracting frequency-domain features to obtain frequency-domain feature vectors; concatenating the time-domain feature vectors and frequency-domain feature vectors to obtain time-frequency feature vectors for the frequency sub-signals; obtaining a set of main features based on the time-frequency feature vectors of all frequency sub-signals; and processing the set of main features using a pre-trained event recognition model to obtain the target vibration event type. Compared with existing technologies, this application effectively improves the accuracy of fiber optic vibration event identification and detection by concatenating time-domain and frequency-domain features and performing identification based on the concatenated features.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of fiber optic maintenance, and more specifically, to a method, system, electronic device, and storage medium for detecting fiber optic vibration events. Background Technology

[0002] Distributed Vibration Sensing (DVS) is based on phase-sensitive optical time-domain reflectometry and uses ordinary single-mode optical fiber as the sensing medium to achieve distributed detection of vibration signals along the fiber optic line. This system offers advantages such as high sensitivity, resistance to electromagnetic interference, passive detection, and long-distance continuous monitoring, and is widely used in perimeter security, oil and gas pipeline safety early warning, and railway line monitoring.

[0003] In existing distributed fiber optic vibration sensing technologies, vibration events along the fiber optic line are typically identified and predicted directly based on the acquired fiber optic vibration signals. However, most existing distributed fiber optic vibration sensing technologies analyze vibration signals only from a single dimension, either the time domain or the frequency domain. Even when some methods attempt to combine time-domain and frequency-domain features, they often treat the two types of features as independent information sources and simply splice them together or make separate judgments. This lack of deep fusion and structured dimensionality reduction of time-frequency features leads to high feature redundancy and significant information conflicts, thereby reducing the accuracy of identifying and detecting fiber optic vibration events. Summary of the Invention

[0004] This invention provides a method, system, electronic device, and storage medium for detecting fiber optic vibration events, which can effectively improve the accuracy of fiber optic vibration event identification and detection.

[0005] According to a first aspect of this application, a method for detecting fiber optic vibration events is provided, the method comprising: Acquire the time-domain signal of a preset sampling point on the optical fiber under test; the optical fiber under test has a number of preset sampling points set according to a preset interval. The time-domain signal at the preset sampling points is divided into several frequency sub-signals; Time-domain feature extraction is performed on the frequency sub-signal to obtain the time-domain feature vector of the frequency sub-signal; Frequency domain features are extracted from the frequency sub-signal to obtain the frequency domain feature vector of the frequency sub-signal; The time-domain feature vector of the frequency sub-signal is concatenated with the frequency-domain feature vector to obtain the time-frequency feature vector of the frequency sub-signal. Based on the time-frequency feature vectors of all frequency sub-signals at the preset sampling points, the main feature set of the optical fiber under test is obtained; The main feature set is processed by a pre-trained event recognition model to obtain the target vibration event type in the optical fiber under test.

[0006] By acquiring the time-domain signal of a preset sampling point on the optical fiber under test and dividing it into several frequency sub-signals, time-domain feature vectors and frequency-domain feature vectors are extracted from each frequency sub-signal. This achieves comprehensive extraction of the time-domain and frequency-domain features of the optical fiber vibration signal, thus overcoming the information loss problem caused by relying on only a single-dimensional feature. Furthermore, by concatenating the time-domain feature vector and the frequency-domain feature vector into a time-frequency feature vector, the correlation information between the time-domain and frequency-domain features is preserved, avoiding information fragmentation caused by separate judgments. Then, by obtaining the main feature set based on the time-frequency feature vectors of all frequency sub-signals, feature redundancy is effectively reduced and noise interference is removed. Finally, the main feature set is processed by a pre-trained event recognition model, which significantly improves the model's ability to discriminate vibration events, thereby effectively improving the recognition and detection accuracy of optical fiber vibration events.

[0007] Optionally, dividing the time-domain signal of the preset sampling points into several frequency sub-signals includes: Extract the effective frequency range of the optical fiber under test, and set the start sampling frequency point, end sampling frequency point, reference window size of the dynamic sampling window, and sliding step size for the time domain signal based on the effective frequency range; Starting from the initial sampling frequency point and ending from the final sampling frequency point, several frequency sub-signals are slidably extracted from the time-domain signal through the dynamic sampling window and the sliding step size; wherein, before each sliding, the actual window size of the current sliding extraction of the dynamic sampling window is determined according to the current number of sliding extractions and the reference window size, and the current sliding extraction is performed according to the actual size.

[0008] By extracting the effective frequency range of the optical fiber under test and setting the starting sampling frequency point, ending sampling frequency point, reference window size of the dynamic sampling window, and sliding step size for the time-domain signal, the division parameters of frequency sub-signals can be flexibly configured according to different application scenarios, improving the scenario adaptability of the method. By using the starting sampling frequency point as the starting point and the ending sampling frequency point as the ending point, several frequency sub-signals are truncate using the dynamic sampling window and sliding step size. Before each truncation, the actual window size of the current truncation is determined based on the number of truncations and the reference window size, which enables frequency overlap between adjacent frequency sub-signals. This effectively avoids signal energy leakage and boundary information loss caused by strict frequency band division, ensuring the integrity and accuracy of subsequent feature extraction.

[0009] Optionally, obtaining the main feature set of the optical fiber under test based on the time-frequency feature vectors of all the frequency sub-signals at the preset sampling points includes: The time-frequency feature vector of the frequency sub-signal is standardized to obtain the time-frequency feature standard vector; Stack the time-frequency feature standard vectors of all the preset sampling points and their frequency sub-signals to obtain the global time-frequency feature matrix; Calculate the covariance matrix of the global time-frequency feature matrix; The covariance matrix is ​​subjected to eigenvalue decomposition to obtain several eigenvectors and the corresponding eigenvalues ​​of each eigenvector; wherein the magnitude of each eigenvalue represents the data variance in the direction of its corresponding eigenvector. Select the first few largest feature values ​​as candidate feature values. If the ratio of the sum of the values ​​of the candidate feature values ​​to the sum of the values ​​of the feature values ​​exceeds a preset ratio threshold, then the candidate feature values ​​are used as target feature values. Construct a projection matrix based on the eigenvectors corresponding to the target eigenvalues; The global time-frequency feature matrix is ​​projected according to the projection matrix to obtain the main feature set as the main feature set.

[0010] By standardizing the time-frequency feature vectors of the frequency sub-signals, a standard time-frequency feature vector is obtained, eliminating the influence of dimensional differences between different feature dimensions on the analysis results. A global time-frequency feature matrix is ​​obtained by stacking the standard time-frequency feature vectors of all preset sampling points and their frequency sub-signals, and the covariance matrix of this matrix is ​​calculated, laying the foundation for subsequent feature structure analysis. Eigenvalue decomposition of the covariance matrix yields several eigenvectors and their corresponding eigenvalues. The magnitude of each eigenvalue represents the data variance in the direction of its corresponding eigenvector, thus achieving a quantitative assessment of the importance of each principal component direction in the original feature space. By selecting the top few largest eigenvalues ​​as candidate eigenvalues, and constructing a projection matrix using the eigenvectors corresponding to the candidate eigenvalues ​​as target eigenvectors when the ratio of the sum of candidate eigenvalues ​​to the sum of all eigenvalues ​​exceeds a preset threshold, the number of principal components is adaptively selected, avoiding the subjectivity of manually setting the number of principal components. Finally, the global time-frequency feature matrix is ​​projected using the projection matrix to obtain the principal feature set. This effectively reduces the feature dimension while retaining the main information of the original features, reducing data redundancy and computational overhead, and removing noise and irrelevant features from interfering with subsequent recognition results.

[0011] Optionally, the event recognition model includes several parallel event recognition networks, each corresponding to a vibration event; the event recognition network is used to recognize its input features and output the confidence level of the vibration event corresponding to the event recognition network. The process of processing the main feature set using a pre-trained event recognition model to obtain the target vibration event type in the optical fiber under test includes: For each event recognition network in the event recognition model, the main feature set is used as input features to the event recognition network for processing, and the confidence level of the vibration event output by the event recognition network is obtained. Based on the confidence level of each vibration event, the type of target vibration event in the optical fiber under test is obtained.

[0012] By setting up several parallel event recognition networks, each corresponding to a vibration event and outputting the confidence level of that event, the discrimination task of different event types is undertaken by their respective independent network branches, avoiding the problem of feature competition and mutual interference between different categories in a single multi-classification network. By inputting the main feature set into each event recognition network to obtain the confidence level of each vibration event, and determining the target vibration event type based on each confidence level, parallel discrimination and comprehensive decision-making for multiple vibration events are realized, improving the accuracy and reliability of event type recognition.

[0013] Optionally, the event recognition network includes several channel processing subnetworks and differentiable fusion subnetworks; The step of inputting the main feature set as input features into the event recognition network for processing, and obtaining the confidence level of the vibration event output by the event recognition network, includes: Extract the frequency range of each frequency sub-signal corresponding to each of the preset sampling points; From the main feature set, extract the feature components corresponding to the frequency range of each sub-signal; Each of the aforementioned feature components is input into each of the aforementioned channel processing sub-networks for processing, thereby obtaining the channel feature vector of each of the aforementioned feature components; The channel feature vectors of all the aforementioned feature components are concatenated to obtain the fused feature vector; The confidence level of the vibration event is obtained by identifying the fused feature vector through the differentiable fusion subnetwork.

[0014] A hierarchical feature processing architecture was constructed by setting up several channel processing sub-networks and differentiable fusion sub-networks in the event recognition network. By extracting the frequency range of each frequency sub-signal and extracting the feature components corresponding to each frequency range from the main feature set, the precise separation and targeted extraction of features of different frequency sub-signals in the main feature set were achieved. By inputting each feature component into each channel processing sub-network for processing to obtain the channel feature vector of each feature component, the independent parallel processing of features of different frequency sub-signals was realized, and the discriminative information in each frequency sub-signal was fully explored. By concatenating all channel feature vectors into a fused feature vector and inputting it into the differentiable fusion sub-network for recognition, the comprehensive utilization and deep fusion of features of each frequency sub-signal were realized, effectively improving the accuracy of the confidence assessment of vibration events.

[0015] Optionally, the step of inputting each of the feature components into each of the channel processing sub-networks for processing to obtain the channel feature vector of each feature component includes: The feature components are processed by a differentiable support vector machine to obtain the first confidence vector; The second confidence vector is obtained by calculating the weighted probability of the decision path for the feature components using a decision tree. The feature components are processed by logistic regression to obtain the third confidence vector; The channel feature vector is obtained based on the first confidence vector, the second confidence vector, and the third confidence vector.

[0016] By applying differentiable support vector machine (SVM), decision tree weighted probability calculation, and logistic regression to the feature components through a channel processing subnetwork, first, second, and third confidence vectors are obtained, respectively. This enables multi-angle discrimination of the same feature component from three different technical paths. Differentiable SVM discriminates features through a maximum margin hyperplane, decision tree discriminates features through decision path weighted probability calculation, and logistic regression discriminates features through linear weighted mapping. These three basic operators have different decision mechanisms and discrimination preferences, each extracting discrimination information from the feature components from different perspectives. By obtaining the channel feature vector based on the first, second, and third confidence vectors, the discrimination results of the three basic operators are fused, effectively combining the advantages of each basic operator and suppressing the discrimination bias that may exist in a single basic operator, thereby improving the expressive power and robustness of the channel feature vector.

[0017] Optionally, the step of performing differentiable support vector machine processing on the feature components to obtain the first confidence vector includes: The feature components are mapped to a high-dimensional space using a differentiable support vector machine to obtain a high-dimensional feature representation. The original score is obtained by calculating the high-dimensional feature representation using the differentiable decision boundary; The original score is calculated using a smoothed hinge loss function to obtain feature probability values ​​as the first confidence vector; and / or; The decision tree includes several node layers, each node layer including at least one node; wherein, the top node layer has one node, which is the root node of the decision tree; the bottom node layer has several nodes, which are the leaf nodes of the decision tree; each node in the node layer is connected to at least two nodes in the next node layer through directed edges, and the nodes in the next node layer are child nodes of the node layer nodes that have directed edge connections with them. The step of calculating the weighted probability of decision paths for the feature components using a decision tree to obtain the second confidence vector includes: The feature components are input into the root node of the decision tree; In the decision tree, each node calculates the radial probability of flowing to each child node in its next node layer based on the node input features and the learnable node parameters of the node; based on the radial probabilities of each node, the arrival probability of the feature component reaching each leaf node is calculated; based on the learnable node parameters of the leaf node and the arrival probability of reaching the leaf node, the path probability of the leaf node is calculated; the path probabilities of the leaf nodes are weighted to obtain the feature path probability, which is used as the second confidence vector.

[0018] When processing the feature components using a differentiable support vector machine (SVM), the SVM maps the feature components to a high-dimensional space to obtain a high-dimensional feature representation, making the originally linearly inseparable features linearly separable in the high-dimensional space. The original score is obtained by calculating the high-dimensional feature representation through the differentiable decision boundary, and the feature probability value is obtained by calculating the original score through the smooth hinge loss function, which serves as the first confidence vector. This ensures that the output of the SVM is a continuously differentiable probability value, guaranteeing the gradient differentiability of the entire processing process and providing support for end-to-end joint training.

[0019] And / or, When calculating the weighted probability of decision paths for feature components using a decision tree, the radial probability of each node flowing to its child nodes is calculated based on the node's input features and learnable node parameters, thus achieving soft and differentiable node splitting. The arrival probability of feature components reaching each leaf node is calculated based on the radial probability of each node, and then the path probability of each leaf node is calculated based on the learnable node parameters and arrival probability of the leaf node. The path probability of each leaf node is then weighted to obtain the feature path probability as the second confidence vector. This makes the output of the decision tree a continuously differentiable probability value, breaking through the non-differentiability limitation caused by hard splitting in traditional decision trees. This allows the decision tree to participate in gradient backpropagation and joint optimization as a layer of a neural network, while retaining the good interpretability of the decision tree.

[0020] According to a second aspect of this application, a fiber optic vibration event detection system is provided, the system comprising: The signal acquisition module is used to acquire the time-domain signal of a preset sampling point on the optical fiber under test; the optical fiber under test has a number of preset sampling points set according to a preset interval. The frequency division module is used to divide the time-domain signal of the preset sampling points into several frequency sub-signals; The time-domain feature extraction module is used to extract time-domain features from the frequency sub-signal to obtain the time-domain feature vector of the frequency sub-signal; The frequency domain feature extraction module is used to extract frequency domain features from the frequency sub-signal to obtain the frequency domain feature vector of the frequency sub-signal; The time-frequency splicing module is used to splice the time-domain feature vector of the frequency sub-signal with the frequency-domain feature vector to obtain the time-frequency feature vector of the frequency sub-signal; The main feature extraction module is used to obtain the main feature set of the optical fiber under test based on the time-frequency feature vectors of all the frequency sub-signals at the preset sampling points; The event recognition module is used to process the main feature set through a pre-trained event recognition model to obtain the target vibration event type in the optical fiber under test.

[0021] According to a third aspect of this application, an electronic device is provided, comprising: Memory, used to store one or more computer programs; A processor, when the one or more computer programs are executed by the processor, implements the fiber optic vibration event detection method described in the first aspect above.

[0022] According to a fourth aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the fiber optic vibration event detection method described in the first aspect above.

[0023] Based on any of the above aspects, the fiber optic vibration event detection method, system, electronic device, and computer storage medium provided in this application acquire the time-domain signal of a preset sampling point on the fiber under test and divide it into several frequency sub-signals. Time-domain feature vectors are obtained by extracting time-domain features from each frequency sub-signal, and frequency-domain feature vectors are obtained by extracting frequency-domain features from each frequency sub-signal. This achieves comprehensive extraction of both time-domain and frequency-domain features of the fiber optic vibration signal, thus overcoming the information loss problem caused by relying solely on single-dimensional features. Furthermore, by concatenating the time-domain feature vector and the frequency-domain feature vector into a time-frequency feature vector, the correlation information between the time-domain and frequency-domain features is preserved, avoiding information fragmentation caused by separate judgments. Then, by obtaining the main feature set based on the time-frequency feature vectors of all frequency sub-signals, feature redundancy is effectively reduced and noise interference is removed. Finally, the main feature set is processed by a pre-trained event recognition model, significantly improving the model's ability to discriminate vibration events, thereby effectively improving the accuracy of fiber optic vibration event recognition and detection. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of the steps of the detection method provided in this embodiment.

[0026] Figure 2 This is a schematic diagram of the steps for extracting frequency domain feature vectors provided in this embodiment.

[0027] Figure 3 This is a flowchart illustrating the steps for obtaining the main feature set provided in this embodiment.

[0028] Figure 4 This is a flowchart illustrating the steps for obtaining the confidence level of a vibration event in this embodiment.

[0029] Figure 5 This is a schematic diagram of the steps for obtaining the channel feature vector provided in this embodiment.

[0030] Figure 6 This is a schematic diagram of the functional modules of the detection system provided in this embodiment.

[0031] Figure 7 This is a schematic diagram of the device structure of the electronic device provided in this embodiment. Detailed Implementation

[0032] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this application. To better illustrate the following embodiments, some components in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product; it is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0033] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0034] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0035] Distributed Vibration Sensing (DVS) is based on phase-sensitive optical time-domain reflectometry and uses ordinary single-mode optical fiber as the sensing medium to achieve distributed detection of vibration signals along the fiber optic line. This system offers advantages such as high sensitivity, resistance to electromagnetic interference, passive detection, and long-distance continuous monitoring, and is widely used in perimeter security, oil and gas pipeline safety early warning, and railway line monitoring.

[0036] In existing distributed fiber optic vibration sensing technologies, vibration events along the fiber optic line are typically identified and predicted directly based on the acquired fiber optic vibration signals. However, most existing distributed fiber optic vibration sensing technologies analyze vibration signals only from a single dimension, either the time domain or the frequency domain. Even when some methods attempt to combine time-domain and frequency-domain features, they often treat the two types of features as independent information sources and simply splice them together or make separate judgments. This lack of deep fusion and structured dimensionality reduction of time-frequency features leads to high feature redundancy and significant information conflicts, thereby reducing the accuracy of identifying and detecting fiber optic vibration events.

[0037] This embodiment provides a technical solution that can solve the above problems. The specific implementation of this application will be described in detail below with reference to the accompanying drawings.

[0038] like Figure 1 As shown, this embodiment provides a method for detecting fiber optic vibration events, which may include the following steps: S1: Acquire the time-domain signal of a preset sampling point on the optical fiber under test; the optical fiber under test has a number of preset sampling points set according to a preset interval. In this embodiment, the optical fiber under test is pre-set with several preset sampling points at preset intervals. The optical fiber under test can be connected to a distributed optical fiber vibration sensing system, and the distributed sensing system can be used to collect the time-domain signal of the previous preset sampling point.

[0039] Specifically, after the optical fiber under test is connected to the distributed optical fiber vibration sensing system, the system injects probe pulses into the fiber based on the principle of phase-sensitive optical time-domain reflectometry (OTDR) and receives the backscattered Rayleigh light returning along the fiber. After photoelectric conversion, the light is sampled by a high-speed ADC. According to the OTDR principle, different sampling times of the ADC correspond to different positions on the optical fiber. The signal amplitude at each sampling time is mapped to a corresponding preset sampling point, thus obtaining the instantaneous vibration amplitude at each preset sampling point. By continuously transmitting multiple probe pulses and accumulating the sampled values ​​of each preset sampling point over multiple pulse cycles, the time-domain signal of that preset sampling point is obtained.

[0040] S2: Divide the time-domain signal of the preset sampling points into several frequency sub-signals; In this embodiment, the time-domain signal of each preset sampling point can be subjected to polyphase filtering according to a preset frequency range, and divided into multiple frequency sub-signals.

[0041] In one implementation, step S2 may include the following steps: First, the effective frequency range of the optical fiber under test is extracted. Based on the effective frequency range, the starting sampling frequency point, the ending sampling frequency point, the reference window size of the dynamic sampling window, and the sliding step size are set for the time-domain signal.

[0042] Then, starting from the initial sampling frequency point and ending at the final sampling frequency point, several frequency sub-signals are slidably extracted from the time-domain signal using the dynamic sampling window and the sliding step size. Before each slide, the actual window size of the current sliding extraction is determined based on the current number of extractions and the reference window size, and the current sliding extraction is performed according to the actual size.

[0043] Understandably, the effective frequency range is predetermined based on the deployment scenario of the optical fiber under test and the preset vibration event type. The reference window size corresponds to the minimum bandwidth unit of the frequency division. The sliding step size is less than or equal to the reference window size, so that the frequency sub-signals obtained by adjacent sliding intercepts have partial overlap in the frequency domain, thereby avoiding signal energy leakage caused by strict frequency band division.

[0044] S3: Extract time-domain features from the frequency sub-signal to obtain the time-domain feature vector of the frequency sub-signal; In this embodiment, after acquiring the frequency sub-signals, time-domain features can be extracted from each frequency sub-signal. For any frequency sub-signal, its peak value, peak-to-peak value, root mean square (RMS), variance, skewness, and kurtosis can be calculated as the time-domain feature vector. Specifically, the peak value is the maximum absolute value of the frequency sub-signal, representing the extreme value of vibration intensity; the peak-to-peak value is the difference between the maximum and minimum values ​​of the frequency sub-signal, representing the range of vibration amplitude variation; the RMS is the square root of the squared mean of the frequency sub-signal, representing the effective energy of the vibration; the variance is the average of the squared differences between each sampled value of the frequency sub-signal and the mean, representing the degree of signal fluctuation; the skewness represents the asymmetry of the signal distribution; and the kurtosis represents the impact characteristics of the signal. Arranging these time-domain statistical features in a preset order yields the time-domain feature vector of the frequency sub-signal.

[0045] S4: Extract frequency domain features from the frequency sub-signal to obtain the frequency domain feature vector of the frequency sub-signal; In this embodiment, as Figure 2 As shown, step S4 may include the following sub-steps: S41: Extract the time-frequency distribution corresponding to the frequency sub-signal, and the sub-signal frequency range of the frequency sub-signal; In this embodiment, the frequency sub-signal can be subjected to Gabor transform to obtain the time-frequency distribution corresponding to the frequency sub-signal and the sub-signal frequency range of the frequency sub-signal. The Gabor transform, by windowing in both the time and frequency domains, can convert a one-dimensional time-domain signal into a two-dimensional time-frequency distribution. It can be understood that the time-frequency distribution of the frequency sub-signal reflects the distribution of signal energy in both time and frequency dimensions.

[0046] S42: Based on the preset frequency range and the frequency range of the sub-signal, the time-frequency distribution is divided into several sub-frequency bands; In this embodiment, the preset frequency range is used to determine the frequency width of each sub-band, and the sub-signal frequency range is used to define the overall boundary of the division, so that each sub-band after division completely covers the effective frequency band range of the frequency sub-signal.

[0047] S43: Extract frequency domain features from the sub-frequency band to obtain the frequency domain feature sub-vector of the sub-frequency band; In this embodiment, the frequency domain features include at least one of the following: sub-band energy proportion, sub-band energy entropy, sub-band center frequency, sub-band bandwidth, spectral centroid, spectral spread, spectral flux, and spectral flatness. Specifically, the sub-band energy proportion is the ratio of energy within a sub-band to the total energy of the sub-signal at that frequency; the sub-band energy entropy is the entropy value of the energy distribution within a sub-band, used to characterize the degree of energy dispersion within the sub-band; the sub-band center frequency is the weighted average frequency of energy within the sub-band; the sub-band bandwidth is the degree of dispersion of the frequency distribution within the sub-band; the spectral centroid is the weighted average frequency of the global spectral energy; the spectral spread is the degree of dispersion of spectral energy relative to the spectral centroid; the spectral flux is the change in spectral energy between adjacent time frames; and the spectral flatness is the ratio of the geometric mean to the arithmetic mean of the spectral energy, used to characterize the flatness of the spectrum.

[0048] S44: Concatenate the frequency domain feature vectors of all sub-bands of the frequency sub-signal to obtain the frequency domain feature vector of the frequency sub-signal.

[0049] Understandably, by dividing the time-frequency distribution into sub-bands and extracting features from each sub-band, we can more precisely characterize the energy distribution of the signal in different frequency ranges and avoid the loss of local frequency domain information that may occur when extracting features from the entire frequency band as a whole.

[0050] S5: Concatenate the time-domain feature vector of the frequency sub-signal with the frequency-domain feature vector to obtain the time-frequency feature vector of the frequency sub-signal; In this embodiment, for each frequency sub-signal, the time-domain feature vector and the frequency-domain feature vector of the frequency sub-signal can be concatenated end-to-end to form the time-frequency feature vector of the frequency sub-signal. The time-frequency feature vector contains both time-domain statistical information and frequency-domain distribution information, providing a comprehensive feature representation for subsequent feature reduction and event recognition.

[0051] S6: Obtain the main feature set of the optical fiber under test based on the time-frequency feature vectors of all the frequency sub-signals at the preset sampling points; In this embodiment, as Figure 3 As shown, step S6 may include the following sub-steps: S61: Standardize the time-frequency feature vector of the frequency sub-signal to obtain a time-frequency feature standard vector; In one embodiment of this example, the standardization process can employ the Z-score standardization method, which ensures that each dimension of each time-frequency feature vector has zero mean and unit variance, thereby eliminating the influence of dimensional differences between different feature dimensions on subsequent analysis results.

[0052] S62: Stack the time-frequency feature standard vectors of all the preset sampling points and their frequency sub-signals to obtain the global time-frequency feature matrix; In this embodiment, the time-frequency feature standard vectors of all frequency sub-signals under each preset sampling point can be arranged in the order of frequency sub-signals, and then the arrangement results of all preset sampling points can be stacked in the order of sampling points to form the global time-frequency feature matrix.

[0053] S63: Calculate the covariance matrix of the global time-frequency feature matrix; S64: Perform eigenvalue decomposition on the covariance matrix to obtain several eigenvectors and the eigenvalues ​​corresponding to each eigenvector; In this embodiment, the covariance matrix can be eigenvalued to obtain several eigenvectors and their corresponding eigenvalues. The magnitude of each eigenvalue represents the data variance along the direction of its corresponding eigenvector; a larger eigenvalue indicates a more dispersed data distribution along that eigenvector direction and contains a greater amount of information.

[0054] S65: Select the first few largest feature values ​​as candidate feature values. If the ratio of the sum of the values ​​of the candidate feature values ​​to the sum of the values ​​of the feature values ​​exceeds a preset ratio threshold, then the candidate feature values ​​are used as target feature values. S66: Construct a projection matrix based on the eigenvectors corresponding to the target eigenvalues; S67: Project the global time-frequency feature matrix according to the projection matrix to obtain the main feature set as the main feature set.

[0055] In this embodiment, the top few largest feature values ​​can be selected as candidate feature values. The ratio of the sum of the candidate feature values ​​to the sum of all feature values ​​is calculated. If this ratio exceeds a preset threshold, the feature vector corresponding to the candidate feature value is taken as the target feature vector. Then, a projection matrix is ​​constructed based on the target feature vector, and the projection matrix is ​​used to project the global time-frequency feature matrix to obtain the main feature set.

[0056] It is understandable that the main feature set retains the main information in the original global time-frequency feature matrix, while effectively reducing the feature dimension and reducing data redundancy and computational overhead.

[0057] S7: The main feature set is processed by a pre-trained event recognition model to obtain the target vibration event type in the optical fiber under test.

[0058] In this embodiment, the event recognition model includes several parallel event recognition networks, each corresponding to the recognition of a vibration event. Each event recognition network is used to recognize its input features and output the confidence level of the vibration event corresponding to that event recognition network.

[0059] In this embodiment, step S7 may include the following steps: For each event recognition network in the event recognition model, the main feature set is input as input features into the event recognition network for processing to obtain the confidence level of the vibration event output by the event recognition network; based on the confidence level of each vibration event, the target vibration event type in the optical fiber under test is obtained.

[0060] In one embodiment, the event recognition network includes several channel processing subnetworks and differentiable fusion subnetworks; in this embodiment, such as Figure 4 As shown, the step of inputting the main feature set as input features into the event recognition network for processing, and obtaining the confidence level of the vibration event output by the event recognition network, may include the following steps: A1: Extract the frequency range of each frequency sub-signal corresponding to the preset sampling point; A2: Extract the feature components corresponding to the frequency range of each sub-signal from the main feature set; In this embodiment, the feature component refers to the part of the feature data belonging to the same frequency sub-signal in the main feature set.

[0061] A3: Input each of the aforementioned feature components into each of the aforementioned channel processing sub-networks for processing to obtain the channel feature vector of each of the aforementioned feature components; In this embodiment, each of the channel processing sub-networks operates in parallel, with each channel processing sub-network corresponding to a frequency sub-signal, thus realizing independent parallel processing of the characteristics of different frequency sub-signals.

[0062] Furthermore, such as Figure 5 As shown, step A3 may include the following sub-steps: A31: Perform differentiable support vector machine processing on the feature components to obtain the first confidence vector; In this embodiment, step A31 may include the following steps: The feature components are mapped to a high-dimensional space using a differentiable support vector machine to obtain a high-dimensional feature representation. The high-dimensional space mapping uses a radial basis kernel function, which maps the original features to a high-dimensional space, making the originally linearly inseparable features linearly separable in the high-dimensional space. The high-dimensional feature representation is calculated using a differentiable decision boundary to obtain the original score; the original score is then calculated using a smooth hinge loss function to obtain the feature probability value as the first confidence vector. In this embodiment, the smoothing hinge loss function can be expressed as follows: In the formula, the Let represent the feature probability value, and s represent the original score obtained by the differentiable support vector machine after mapping the feature components to a high-dimensional space and calculating through the differentiable decision boundary. This represents the true class label corresponding to the feature component, with a value of +1 or -1. During the training phase, the true class label... Determined by the labeled information in the training data; during the inference phase, The sign of is used to determine the confidence direction corresponding to the feature probability value. The larger the value of , the higher the consistency between the original score and the true category label; the smaller the value of the smooth hinge loss function, the higher the confidence of the classification result.

[0063] Understandably, the original score is calculated using a smoothed hinge loss function to obtain the feature probability value, which serves as the first confidence vector. The smoothed hinge loss function replaces the non-differentiable portion of the standard hinge loss function at the inflection point with a squared term, making the first confidence vector globally continuous and differentiable, thus supporting gradient backpropagation. This allows a differentiable support vector machine to participate as a layer in end-to-end joint training of a neural network, achieving joint optimization of feature extraction and decision-making while retaining the maximum margin classification advantage of support vector machines.

[0064] A32: The second confidence vector is obtained by calculating the weighted probability of the decision path for the feature components using a decision tree; In this embodiment, the decision tree includes several node layers, each node layer including at least one node; wherein, the top node layer has one node, which is the root node of the decision tree; the bottom node layer has several nodes, which are the leaf nodes of the decision tree; each node in the node layer is connected to at least two nodes in the next node layer through directed edges, and the nodes in the next node layer are child nodes of the node layer nodes that have directed edge connections with them; In this embodiment, step A32 may include the following steps: The feature components are input into the root node of the decision tree; In the decision tree, each node calculates the radial probability of flowing to each child node in its next node layer based on the node's input features and the node's learnable node parameters. Each node is mapped to the radial probability of flowing to each child node by taking the inner product of the node's learnable weight vector and the input feature vector, adding a bias, and then mapping it to each child node using the Sigmoid function.

[0065] Based on the radial probabilities of each node, the arrival probability of the feature component reaching each leaf node is calculated, and the arrival probability is the product of the radial probabilities of all nodes on the path from the root node to the leaf node. The path probability of a leaf node is calculated based on its learnable node parameters and the probability of reaching that leaf node. The path probability of a leaf node can be obtained by multiplying the learnable node parameters of each leaf node by the probability of reaching that leaf node. The path probabilities of the leaf nodes are weighted to obtain the feature path probabilities, which are used as the second confidence vector.

[0066] Understandably, the decision tree adopts a soft splitting mechanism based on probability distribution. Since each node splits to all child nodes simultaneously in a probabilistic manner, the final output is the probability weighted sum of the outputs of all leaf nodes. Therefore, the entire calculation process of the decision tree is continuously differentiable and supports gradient backpropagation, enabling the decision tree to participate in end-to-end joint training as a layer of a neural network, while retaining the good interpretability of the decision tree.

[0067] A33: The feature components are processed by logistic regression to obtain the third confidence vector; In this embodiment, the logistic regression can be implemented using an L2-regularized logistic regression model. The logistic regression model performs a linear weighted summation of the input feature components to obtain a linear score, and then uses a Sigmoid function to map the linear score to probability values ​​between 0 and 1, which serve as the third confidence vector. The Sigmoid function compresses the linear score from an unbounded real-number space to a bounded probability space, giving the logistic regression output a clear probabilistic meaning. Furthermore, the Sigmoid function itself is continuously differentiable, allowing logistic regression to participate as a differentiable layer in end-to-end joint training of a neural network without requiring additional differentiability modifications.

[0068] Understandably, the third confidence vector is obtained by processing the feature components through logistic regression. The continuous differentiability of the Sigmoid function itself enables logistic regression to naturally support gradient backpropagation, allowing it to be trained collaboratively with other differentiable layers in the cascaded network. L2 regularization, by introducing a L2 norm penalty term into the weight vector in the loss function, suppresses overfitting of the model to the training data and enhances the generalization ability of logistic regression under small sample conditions. At the same time, as a linear classifier, logistic regression has a clear linear separability interpretation for its decision boundary, which can complement the nonlinear kernel mapping of differentiable support vector machines and the probabilistic splitting of soft decision trees. The three basis operators comprehensively evaluate the same feature component from three different perspectives: linear discrimination, nonlinear kernel discrimination, and tree-based probabilistic discrimination. Through subsequent splicing and fusion, the expressive power and robustness of the channel feature vector are effectively improved.

[0069] A34: Obtain the channel feature vector based on the first confidence vector, the second confidence vector, and the third confidence vector.

[0070] Understandably, differentiable support vector machines, decision trees, and logistic regression employ different technical approaches to discriminate the same feature component, and these three basic operators have different decision-making mechanisms and discrimination preferences. By concatenating the outputs of the three into a channel feature vector, the fusion of the three discrimination results is achieved, effectively combining the advantages of each basic operator.

[0071] A4: Concatenate the channel feature vectors of all the aforementioned feature components to obtain the fused feature vector; A5: The confidence level of the vibration event is obtained by identifying the fused feature vector through the differentiable fusion subnetwork.

[0072] Through the above steps, the confidence level of the vibration event corresponding to each event recognition network is obtained. The event type with the highest confidence level is selected as the final target vibration event type. Alternatively, it can be understood that when the confidence level of all event types is lower than a preset threshold, it is determined to be a no-vibration event.

[0073] For example, suppose the preset vibration event types include three types: "climbing", "knocking", and "vehicle passage". Three event recognition networks are trained for these three vibration event types respectively. Through the above processing, the confidence level output by the event recognition network for the "climbing" vibration event type is 0.82, the confidence level output by the event recognition network for the "knocking" vibration event type is 0.15, and the confidence level output by the event recognition network for the "vehicle passage" vibration event type is 0.21. Since "climbing" has the highest confidence level, it is determined to be the target vibration event type.

[0074] As described above, each of the event recognition networks corresponds to a specific vibration event. It is understood that the event recognition network is pre-trained using samples corresponding to the respective vibration event, enabling it to specifically identify the corresponding vibration event and output the confidence level of that vibration event.

[0075] In one implementation, training the event recognition network may include: For any event recognition network, the fiber optic vibration signal collected for the corresponding event type is used as a positive sample, and the fiber optic vibration signal collected for the non-corresponding event type and the environmental noise signal in the absence of vibration are used as negative samples. The event recognition network is then trained in a supervised manner.

[0076] Understandably, through training, each event recognition network acquires the ability to determine whether an input feature belongs to its corresponding event type. That is, it outputs a higher confidence score when the input feature matches the feature pattern of that event type, and a lower confidence score otherwise. It is also understandable that the training processes of different event recognition networks are independent of each other, and their network parameters are optimized only for their corresponding event types. Therefore, there is no issue of feature competition or mutual interference between the event recognition networks.

[0077] like Figure 6 As shown in the illustration, this application also provides an optical fiber vibration event detection system. Optionally, the system may include: Signal acquisition module 11 is used to acquire time-domain signals of preset sampling points on the optical fiber under test; the optical fiber under test has a number of preset sampling points set according to preset intervals. In this embodiment, the signal acquisition module 11 can be used to perform... Figure 1 For a detailed description of the signal acquisition module 11 shown in step S1, please refer to the description of step S1.

[0078] The frequency division module 12 is used to divide the time-domain signal of the preset sampling point into several frequency sub-signals; In this embodiment, the frequency division module 12 can be used to perform... Figure 1 For a detailed description of the frequency division module 12 shown in step S2, please refer to the description of step S2.

[0079] The time-domain feature extraction module 13 is used to extract time-domain features from the frequency sub-signal to obtain the time-domain feature vector of the frequency sub-signal; In this embodiment, the temporal feature extraction module 13 can be used to perform... Figure 1 For a detailed description of the temporal feature extraction module 13 shown in step S3, please refer to the description of step S3.

[0080] The frequency domain feature extraction module 14 is used to extract frequency domain features from the frequency sub-signal to obtain the frequency domain feature vector of the frequency sub-signal; In this embodiment, the frequency domain feature extraction module 14 can be used to perform... Figure 1 For a detailed description of the frequency domain feature extraction module 14 shown in step S4, please refer to the description of step S4.

[0081] The time-frequency splicing module 15 is used to splice the time-domain feature vector of the frequency sub-signal with the frequency-domain feature vector to obtain the time-frequency feature vector of the frequency sub-signal. In this embodiment, the time-frequency splicing module 15 can be used to perform... Figure 1 For a detailed description of the time-frequency splicing module 15 shown in step S5, please refer to the description of step S5.

[0082] The main feature extraction module 16 is used to obtain the main feature set of the optical fiber under test based on the time-frequency feature vectors of all the frequency sub-signals at the preset sampling points. In this embodiment, the main feature extraction module 16 can be used to perform... Figure 1 For a detailed description of the main feature extraction module 16 shown in step S6, please refer to the description of step S6.

[0083] The event recognition module 17 is used to process the main feature set through a pre-trained event recognition model to obtain the target vibration event type in the optical fiber under test.

[0084] In this embodiment, the event recognition module 17 can be used to perform... Figure 1 For a detailed description of the event recognition module 17 shown in step S7, please refer to the description of step S7.

[0085] This application provides an electronic device with the following structure: Figure 7 As shown.

[0086] The electronic device includes a memory 21, a processor 22, a communication module 23, and an input / output interface 24, etc. Optionally, the memory 21, the processor 22, the communication module 23, and the input / output interface 24 can be connected and communicate with each other through a bus 25.

[0087] The memory 21 is used to store one or more computer programs and to transfer the code of the computer programs to the processor 22; when the one or more computer programs are executed by the processor 22, the fiber optic vibration event detection method in this application embodiment is implemented.

[0088] Optionally, the electronic device can be connected to a network via communication module 23 to communicate with other devices, such as terminals or servers, to achieve data interaction. The electronic device can be various forms of digital computers, exemplarily such as desktop computers, servers, workbenches, mainframes, or other types of computers. The electronic device can also be various forms of mobile terminals, exemplarily such as smartphones, tablets, wearable devices (such as helmets, glasses, watches, etc.), and other similar mobile terminals.

[0089] Optionally, the electronic device can connect to required input / output devices, such as a keyboard or display device, via the input / output interface 24. The electronic device itself may have a display device, and other display devices can also be connected externally via the input / output interface 24. Optionally, a storage device, such as a hard disk, can also be connected via the input / output interface 24 to store data from the electronic device, read data from the storage device, or store data from the storage device in the memory 21. It is understood that the input / output interface 24 can be a wired interface or a wireless interface. Depending on the actual application scenario, the device connected to the input / output interface 24 can be a component of the electronic device or an external device connected to the electronic device when needed.

[0090] Optionally, the memory 21 may be a volatile memory and / or a non-volatile memory. The volatile memory may be a random access memory, etc., and the non-volatile memory may be a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, or a flash memory, etc.

[0091] Optionally, the computer program stored in the processor 22 can be divided into one or more modules, which are stored in the memory 21 and executed by the processor 22 to perform the method provided in this embodiment. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the electronic device.

[0092] Optionally, the processor 22 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 22 include, but are not limited to, a central processing unit, a graphics processing unit, a digital signal processor, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, and can also be any suitable controller, microcontroller, processor, etc. The processor 22 executes the various methods and processes of this embodiment, exemplarily, such as a fiber optic vibration event detection method according to an embodiment of this application.

[0093] Optionally, the bus 25 may include a path for transmitting information. Depending on its function, the bus 25 may be divided into an address bus, a data bus, a control bus, etc.

[0094] In an optional implementation, this application embodiment also provides a computer storage medium storing a computer program thereon, which, when executed by a computer, enables the computer to perform the methods described in the above-described method embodiments. Part or all of the computer program can be loaded and / or installed on the memory 21 of an electronic device. When the computer program is executed by the processor 22, one or more steps of a fiber optic vibration event detection method according to this application embodiment can be performed.

[0095] Optionally, the computer-readable storage medium may be a random access memory, a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, etc.

[0096] Obviously, the above embodiments of this application are merely examples for clearly illustrating the technical solution of this application, and are not intended to limit the specific implementation of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the claims of this application should be included within the protection scope of the claims of this application.

Claims

1. A method for detecting fiber optic vibration events, characterized in that, The method includes: Acquire the time-domain signal of a preset sampling point on the optical fiber under test; the optical fiber under test has a number of preset sampling points set according to a preset interval. The time-domain signal at the preset sampling points is divided into several frequency sub-signals; Time-domain feature extraction is performed on the frequency sub-signal to obtain the time-domain feature vector of the frequency sub-signal; Frequency domain features are extracted from the frequency sub-signal to obtain the frequency domain feature vector of the frequency sub-signal; The time-domain feature vector of the frequency sub-signal is concatenated with the frequency-domain feature vector to obtain the time-frequency feature vector of the frequency sub-signal. Based on the time-frequency feature vectors of all frequency sub-signals at the preset sampling points, the main feature set of the optical fiber under test is obtained; The main feature set is processed by a pre-trained event recognition model to obtain the target vibration event type in the optical fiber under test.

2. The fiber optic vibration event detection method according to claim 1, characterized in that, The step of dividing the time-domain signal of the preset sampling points into several frequency sub-signals includes: Extract the effective frequency range of the optical fiber under test, and set the start sampling frequency point, end sampling frequency point, reference window size of the dynamic sampling window, and sliding step size for the time domain signal based on the effective frequency range; Starting from the initial sampling frequency point and ending from the final sampling frequency point, several frequency sub-signals are slidably extracted from the time-domain signal through the dynamic sampling window and the sliding step size; wherein, before each sliding, the actual window size of the current sliding extraction of the dynamic sampling window is determined according to the current number of sliding extractions and the reference window size, and the current sliding extraction is performed according to the actual size.

3. The fiber optic vibration event detection method according to claim 1, characterized in that, The step of obtaining the main feature set of the optical fiber under test based on the time-frequency feature vectors of all the frequency sub-signals at the preset sampling points includes: The time-frequency feature vector of the frequency sub-signal is standardized to obtain the time-frequency feature standard vector; Stack the time-frequency feature standard vectors of all the preset sampling points and their frequency sub-signals to obtain the global time-frequency feature matrix; Calculate the covariance matrix of the global time-frequency feature matrix; The covariance matrix is ​​subjected to eigenvalue decomposition to obtain several eigenvectors and the corresponding eigenvalues ​​of each eigenvector; wherein the magnitude of each eigenvalue represents the data variance in the direction of its corresponding eigenvector. Select the first few largest feature values ​​as candidate feature values. If the ratio of the sum of the values ​​of the candidate feature values ​​to the sum of the values ​​of the feature values ​​exceeds a preset ratio threshold, then the candidate feature values ​​are used as target feature values. Construct a projection matrix based on the eigenvectors corresponding to the target eigenvalues; The global time-frequency feature matrix is ​​projected according to the projection matrix to obtain the main feature set as the main feature set.

4. The fiber optic vibration event detection method according to claim 1, characterized in that, The event recognition model includes several parallel event recognition networks, each corresponding to a vibration event; the event recognition network is used to recognize its input features and output the confidence level of the vibration event corresponding to the event recognition network. The process of processing the main feature set using a pre-trained event recognition model to obtain the target vibration event type in the optical fiber under test includes: For each event recognition network in the event recognition model, the main feature set is used as input features to the event recognition network for processing, and the confidence level of the vibration event output by the event recognition network is obtained. Based on the confidence level of each vibration event, the type of target vibration event in the optical fiber under test is obtained.

5. The fiber optic vibration event detection method according to claim 4, characterized in that, The event recognition network includes several channel processing sub-networks and differentiable fusion sub-networks; The step of inputting the main feature set as input features into the event recognition network for processing, and obtaining the confidence level of the vibration event output by the event recognition network, includes: Extract the frequency range of each frequency sub-signal corresponding to each of the preset sampling points; From the main feature set, extract the feature components corresponding to the frequency range of each sub-signal; Each of the aforementioned feature components is input into each of the aforementioned channel processing sub-networks for processing, thereby obtaining the channel feature vector of each of the aforementioned feature components; The channel feature vectors of all the aforementioned feature components are concatenated to obtain the fused feature vector; The confidence level of the vibration event is obtained by identifying the fused feature vector through the differentiable fusion subnetwork.

6. The fiber optic vibration event detection method according to claim 5, characterized in that, The step of inputting each of the aforementioned feature components into each of the aforementioned channel processing sub-networks for processing to obtain the channel feature vector of each of the aforementioned feature components includes: The feature components are processed by a differentiable support vector machine to obtain the first confidence vector; The second confidence vector is obtained by calculating the weighted probability of the decision path for the feature components using a decision tree. The feature components are processed by logistic regression to obtain the third confidence vector; The channel feature vector is obtained based on the first confidence vector, the second confidence vector, and the third confidence vector.

7. The fiber optic vibration event detection method according to claim 6, characterized in that, The step of performing differentiable support vector machine processing on the feature components to obtain the first confidence vector includes: The feature components are mapped to a high-dimensional space using a differentiable support vector machine to obtain a high-dimensional feature representation. The original score is obtained by calculating the high-dimensional feature representation using the differentiable decision boundary; The original score is calculated using a smoothed hinge loss function to obtain feature probability values ​​as the first confidence vector; and / or; The decision tree includes several node layers, each node layer including at least one node; wherein, the top node layer has one node, which is the root node of the decision tree; the bottom node layer has several nodes, which are the leaf nodes of the decision tree; each node in the node layer is connected to at least two nodes in the next node layer through directed edges, and the nodes in the next node layer are child nodes of the node layer nodes that have directed edge connections with them. The step of calculating the weighted probability of decision paths for the feature components using a decision tree to obtain the second confidence vector includes: The feature components are input into the root node of the decision tree; In the decision tree, each node calculates the radial probability of flowing to each child node in its next node layer based on the node input features and the learnable node parameters of the node; based on the radial probabilities of each node, the arrival probability of the feature component reaching each leaf node is calculated; based on the learnable node parameters of the leaf node and the arrival probability of reaching the leaf node, the path probability of the leaf node is calculated; the path probabilities of the leaf nodes are weighted to obtain the feature path probability, which is used as the second confidence vector.

8. A fiber optic vibration event detection system, characterized in that, The system includes: The signal acquisition module is used to acquire the time-domain signal of a preset sampling point on the optical fiber under test; the optical fiber under test has a number of preset sampling points set according to a preset interval. The frequency division module is used to divide the time-domain signal of the preset sampling points into several frequency sub-signals; The time-domain feature extraction module is used to extract time-domain features from the frequency sub-signal to obtain the time-domain feature vector of the frequency sub-signal; The frequency domain feature extraction module is used to extract frequency domain features from the frequency sub-signal to obtain the frequency domain feature vector of the frequency sub-signal; The time-frequency splicing module is used to splice the time-domain feature vector of the frequency sub-signal with the frequency-domain feature vector to obtain the time-frequency feature vector of the frequency sub-signal; The main feature extraction module is used to obtain the main feature set of the optical fiber under test based on the time-frequency feature vectors of all the frequency sub-signals at the preset sampling points; The event recognition module is used to process the main feature set through a pre-trained event recognition model to obtain the target vibration event type in the optical fiber under test.

9. An electronic device, characterized in that, include: Memory, used to store one or more computer programs; A processor, when the one or more computer programs are executed by the processor, implements the fiber optic vibration event detection method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the fiber optic vibration event detection method as described in any one of claims 1-7.