Phi-OTDR vibration event classification method and system based on feature fusion

Abstract

The invention belongs to the technical field of optical fiber signal identification, and provides a phi-OTDR vibration event classification method and system based on feature fusion, and the technical scheme is that a preprocessed multi-class disturbance event spatio-temporal data matrix is divided into single-channel time domain data of multi-class disturbance events; training the constructed vibration event classification model based on the single-channel time domain data of the multi-class disturbance events to obtain a trained vibration event classification model; the construction process of the vibration event classification model comprises the following steps: extracting a first feature map and a second feature map of each type of disturbance events based on single-channel time domain data; fusing the first feature map and the second feature map of each event to obtain fusion features of each type of disturbance events, and performing classification according to the fusion features; and classifying the to-be-detected data based on the trained vibration event classification model to obtain a classification result. And the intelligent level of optical cable safety monitoring is obviously improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical fiber signal recognition, and particularly relates to a vibration event classification method and system for a Φ-OTDR system based on feature fusion. Background Technique

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] Nowadays, the information communication method relying on optical cables is becoming increasingly popular. Due to its many advantages such as large communication capacity, low transmission loss, and strong anti-electromagnetic interference performance, it has become the mainstream way of information transmission in modern society. However, in practical applications, the line is often damaged without warning due to construction of engineering machinery along the line, human external force damage, and other reasons. Therefore, it is crucial to monitor the state of the optical cable.

[0004] Distributed Optical Fiber Sensor (DOFS) realizes long-distance distributed monitoring along the optical fiber by measuring and demodulating the changes of various parameters during the propagation of light. DOFS can simultaneously obtain sensing information of tens of thousands of points on a single optical fiber. The Phase-Sensitive Optical Time Domain Reflectometry (Φ-OTDR) technology is a technical means to realize the function of DOFS. By using the Phase-Sensitive Optical Time Domain Reflectometry (Φ-OTDR) technology, the ground vibration near the underground optical fiber can be detected, the types of threat activities around the optical cable can be identified, and real-time early warning can be provided. They have the advantages of good real-time performance, anti-electromagnetic interference, and long-distance distributed detection.

[0005] Currently, the methods for monitoring the state of optical cables using Φ-OTDR technology can be divided into machine learning algorithms and deep learning algorithms. Machine learning algorithms such as Support Vector Machine (SVM), Random Forest (RF), Decision Tree, Nearest Neighbor Classification Network, etc. have all achieved certain classification effects. However, traditional machine learning algorithms require manual extraction of some features, which is time-consuming and laborious. After the emergence of deep learning, relevant experts at home and abroad have combined deep learning with distributed optical fiber sensors to perform pattern recognition on the occurrence of various events and obtained good recognition effects. Deep learning methods such as CNN can automatically extract data features and solve the pattern recognition problem end-to-end.

[0006] However, most neural networks use two-dimensional (2D) convolutional kernels to construct convolutional layers. Therefore, scholars always transform spatio-temporal data matrices into two-dimensional images. However, this will lead to a relatively high computational cost. To improve the algorithm efficiency, researchers have adopted a real-time distributed deep learning network model based on one-dimensional convolutional neural network (1DCNN), which can capture more features of one-dimensional time series and has better real-time processing ability and higher computational efficiency than traditional two-dimensional convolutional neural networks. However, some existing methods still cannot effectively capture long-sequence time signals, and there is a certain false positive rate (FPR). Summary of the Invention

[0007] To solve at least one of the technical problems existing in the above background art, the present invention provides a vibration event classification method and system for a Φ-OTDR system based on feature fusion, which improves the feature extraction ability of time series signals to enhance event recognition accuracy and reduce the false positive rate.

[0008] To achieve the above object, the present invention adopts the following technical solutions: The first aspect of the present invention provides a vibration event classification method for a Φ-OTDR system based on feature fusion, including the following steps: Obtain spatio-temporal data of multiple types of disturbance events in a phase-sensitive optical time domain reflectometry (Φ-OTDR) distributed optical fiber sensing system; Divide the preprocessed spatio-temporal data matrix of multiple types of disturbance events into single-channel time domain data of multiple types of disturbance events; Train the constructed vibration event classification model based on the single-channel time domain data of multiple types of disturbance events to obtain a trained vibration event classification model; wherein, the construction process of the vibration event classification model includes: Extract a first feature map and a second feature map of each type of disturbance event based on the single-channel time domain data; Fuse the first feature map and the second feature map of each event to obtain a fused feature of each type of disturbance event, and classify according to the fused feature; Classify the data to be detected based on the trained vibration event classification model to obtain a classification result.

[0009] Further, extracting a first feature map of each type of disturbance event based on the single-channel time domain data includes: Extract multi-scale features based on the single-channel time domain data, and fuse them to obtain multi-scale fused features; Input the multi-scale fused features into the BatchNorm layer, and calculate the mean and variance of each feature dimension in the BatchNorm layer to obtain normalized features; Perform global pooling on the normalized features to obtain a C-dimensional vector; Then, a one-dimensional convolutional layer with a size of k processes the C-dimensional vector, and the obtained result is applied with the Sigmoid activation function to obtain the channel weights; The channel weights are multiplied by the original feature map channel by channel to obtain the enhanced feature map.

[0010] Furthermore, multi-scale features of single-channel time-domain data are adopted by the multi-scale TCB module, including: The multi-scale TCB module includes parallel first causal convolution branch, second causal convolution branch, and third causal convolution branch; Among them, the input of the first causal convolution branch is single-channel time-domain data, and the output high-frequency transient event feature is denoted as The input of the second causal convolution branch is single-channel time-domain data, and the output medium-period event feature is denoted as The input of the third causal convolution branch is single-channel time-domain data, and the output long-period event feature is denoted as ; The features output by the first causal convolution branch, the second causal convolution branch, and the third causal convolution branch , and respectively pass through the corresponding global pooling GAP layer and then output the corresponding scale features.

[0011] Furthermore, a second feature map of each type of disturbance event is extracted based on the single-channel time-domain data, including: The single-channel time-domain data is processed in segments; The segmented single-channel time-domain data is respectively input into the corresponding forward LSTM layer and backward LSTM layer. Each forward LSTM layer processes the input sequence in the forward order, and each backward LSTM layer processes the input sequence in the reverse order. Finally, the vector formed by concatenating the two output vectors is used as the final feature.

[0012] Furthermore, when classifying according to the fusion features, the LogSoftmax layer is used to normalize the classification features to generate the classification probability distribution of vibration events.

[0013] Furthermore, the preprocessing of multi-class disturbance event spatio-temporal data includes normalization processing and first-order difference processing.

[0014] The second aspect of the present invention provides a vibration event classification system for a Φ-OTDR system based on feature fusion, including: A data acquisition module, which is used to acquire multi-class disturbance event spatio-temporal data in a phase-sensitive optical time-domain reflectometry Φ-OTDR distributed optical fiber sensing system; A data preprocessing module, which is used to divide the preprocessed multi-class disturbance event spatio-temporal data matrix into single-channel time-domain data of multi-class disturbance events; A vibration event classification model training module, which is used to train the constructed vibration event classification model based on the single-channel time-domain data of multi-class disturbance events to obtain a trained vibration event classification model; wherein, the construction process of the vibration event classification model includes: extracting a first feature map and a second feature map of each class of disturbance event based on the single-channel time-domain data; fusing the first feature map and the second feature map of each event to obtain a fused feature of each class of disturbance event, and classifying according to the fused feature; A classification module, which is used to classify the data to be detected based on the trained vibration event classification model to obtain a classification result.

[0015] Further, the Φ-OTDR distributed optical fiber sensing system includes an ultra-narrow linewidth laser, a signal generator, an acousto-optic modulator, an erbium-doped fiber amplifier, a circulator, a sensing optical fiber, a photodetector, a data acquisition card and a signal processing module; Wherein, the output ends of the ultra-narrow linewidth laser and the signal generator are connected to the input end of the acousto-optic modulator, the output end of the acousto-optic modulator is connected to the input end of the erbium-doped fiber amplifier, the output end of the erbium-doped fiber amplifier is connected to one end of the circulator, the other end of the circulator is respectively connected to the input ends of the sensing optical fiber and the photodetector, the output end of the photodetector is connected to the input end of the data acquisition card, and the output end of the data acquisition card is connected to the signal processing module.

[0016] The third aspect of the present invention provides a computer-readable storage medium.

[0017] A computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it realizes the steps in the above-mentioned Φ-OTDR system vibration event classification method based on feature fusion.

[0018] The fourth aspect of the present invention provides a computer device.

[0019] A computer device, including a memory, a processor and a computer program stored on the memory and executable on the processor, and when the processor executes the program, it realizes the steps in the above-mentioned Φ-OTDR system vibration event classification method based on feature fusion.

[0020] Compared with the prior art, the beneficial effects of the present invention are: The present invention introduces the MTCB-ECA module, extracts multi-scale features based on single-channel time-domain data, dynamically adjusts the contribution of each channel to the final classification result by calculating the importance of the feature channels, and through multi-level feature fusion, passes the fused features through the BiLSTM layer and then combines with the context features for fusion to obtain the final fusion, giving full play to their advantages in time-series feature extraction, channel attention mechanism, and context modeling, and effectively improving the model's ability to extract Φ-OTDR signal features.

[0021] Advantages of additional aspects of the present invention will be partly given in the following description, partly will become obvious from the following description, or will be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings forming a part of this specification are used to provide a further understanding of the present invention, and the schematic embodiments and descriptions thereof of the present invention are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0023] Figure 1 is a flowchart of a method for classifying vibration events of a Φ-OTDR system based on feature fusion provided by an embodiment of the present invention; Figure 2 is a Φ-OTDR distributed optical fiber sensing system provided by an embodiment of the present invention; Figure 3 is a time-domain diagram of a watering event provided by an embodiment of the present invention; Figure 4 is a time-domain diagram of a mining event provided by an embodiment of the present invention; Figure 5 is a time-domain diagram of a knocking event provided by an embodiment of the present invention; Figure 6 is a time-domain diagram of a walking event provided by an embodiment of the present invention; Figure 7 is a time-domain diagram of a shaking event provided by an embodiment of the present invention; Figure 8 is a feature fusion process provided by an embodiment of the present invention; Figure 9 is a diagram of the MTCB-ECA module provided by an embodiment of the present invention; Figure 10 is a BiLSTM layer provided by an embodiment of the present invention; Figure 11 is a gated spatio-temporal feature fusion mechanism provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0025] It should be noted that the following detailed description is illustrative and aims to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0026] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0027] Aiming at the problem that the existing model has insufficient feature extraction ability for long-sequence time-series signals, the present invention provides a vibration event classification method recognition algorithm for a Φ-OTDR system based on feature fusion, which improves the event recognition accuracy and reduces the false alarm rate by enhancing the feature extraction ability of time-series signals, thereby improving the protection of optical cables. First, through multi-level feature fusion, feature extraction and fusion are carried out respectively in time-series feature extraction, channel attention mechanism and context modeling, effectively improving the model's ability to extract Φ-OTDR signal features. In addition, the MTCB-ECA module is introduced to dynamically adjust the contribution of each channel to the final classification result by calculating the importance of feature channels, enabling the model to focus more on key signal features and effectively suppressing the influence of redundant information, thus improving the recognition accuracy of the model. It effectively solves the problems of low recognition accuracy and high false alarm rate in the prior art, and significantly improves the intelligent level of optical cable safety monitoring. Embodiment 1 As Figure 1 shown, this embodiment provides a vibration event classification method for a Φ-OTDR system based on feature fusion, including the following steps: Step 1: Obtain the original data of multiple types of disturbance events in a phase-sensitive optical time domain reflectometry (Φ-OTDR) distributed optical fiber sensing system; As Figure 2 shown, the Φ-OTDR distributed optical fiber sensing system includes an ultra-narrow linewidth laser, a signal generator, an acousto-optic modulator, an erbium-doped fiber amplifier, a circulator, a sensing optical fiber, a photodetector, a data acquisition card, and a signal processing module; Among them, the output ends of the ultra-narrow linewidth laser and the signal generator are connected to the input end of the acousto-optic modulator, the output end of the acousto-optic modulator is connected to the input end of the erbium-doped fiber amplifier, the output end of the erbium-doped fiber amplifier is connected to one end of the circulator, the other end of the circulator is respectively connected to the input ends of the sensing fiber and the photodetector, the output end of the photodetector is connected to the input end of the data acquisition card, and the output end of the data acquisition card is connected to the signal processing module; The specific data acquisition principle is as follows: The continuous optical signal emitted by the ultra-narrow linewidth laser is input into the acousto-optic modulator, and the pulse signal generated by the signal generator is input into the acousto-optic modulator; the acousto-optic modulator modulates the continuous optical signal into a pulsed optical signal; the erbium-doped fiber amplifier amplifies the pulsed optical signal, and the amplified pulsed optical signal is connected to the sensing fiber through the circulator; the backward Rayleigh scattered optical signal returned from the sensing fiber is transmitted to the photodetector through the circulator, the photodetector converts the received optical signal into an electrical signal, and transmits the electrical signal to the data acquisition card; the data acquisition card samples the electrical signal and transmits the obtained sampled signal to the signal processing module; the signal processing module identifies and marks the sampled signal to obtain the original data corresponding to multiple types of disturbance events.

[0028] The finally collected data includes 6 types of events: background noise, excavation, knocking, shaking, and walking. Non-background events include excavation, knocking, shaking, and walking. Each event consists of 10,000 points in the time domain and 10 adjacent spatial nodes. The number of samples in the collected original data is shown in Table 1: Table 1 Number of samples of multiple events collected

[0029] Step 2: Preprocess the obtained original data of multiple types of disturbance events, and divide the preprocessed data into a training set and a test set according to a certain ratio to form an independent single-channel time-domain data set; In this embodiment, preprocessing the original data includes performing normalization processing and first-order difference processing; Using Min-Max normalization processing to obtain the normalized data, the calculation formula is: , Among them, x represents the original data, represents the minimum value of the data, represents the maximum value of the data, represents the normalized data; Perform first-order difference processing on the normalized data with an original time series signal length of 10,000, and the calculation formula is: , Among them, represents the original data, represents the data after the first-order difference, represents the index of the data point.

[0030] After that, the sample sequence is labeled, shuffled, and divided into a dataset according to the ratio of 7:2:1 to obtain single-channel time-domain data .

[0031] In this embodiment, the preprocessed spatio-temporal data matrix is divided into independent single-channel time-domain data, Figure 3 is the time-domain diagram of the watering event, Figure 4 is the time-domain diagram of the excavation event, Figure 5 is the time-domain diagram of the knocking event, Figure 6 is the time-domain diagram of the walking event, Figure 7 is the time-domain diagram of the shaking event.

[0032] Step 3: Train the constructed vibration event classification model based on the single-channel time-domain data to obtain the trained vibration event classification model; As Figure 8 shown, the construction process of the vibration event classification model specifically includes the following steps: Step 301: Input the single-channel time-domain data into the MTCB-ECA module to extract the first feature map of each event; As Figure 9 shown, the MTCB-ECA module includes a multi-scale TCB module, a BatchNorm layer, an ECA layer, a weight normalization layer, and a Dropout layer; As Figure 9 shown, specifically includes: Step 3011: Combine the single-channel time-domain data and the multi-scale TCB module to extract multi-scale features; Among them, the multi-scale TCB module includes parallel first causal convolution branches, second causal convolution branches, and third causal convolution branches; among them, the input of the first causal convolution branch is the single-channel time-domain data, and the output feature is denoted as , the input of the second causal convolution branch is the single-channel time-domain data, and the output feature is denoted as , the input of the third causal convolution branch is the single-channel time-domain data, and the output feature is denoted as ; In this embodiment, the dilation rates of the first causal convolution branch, the second causal convolution branch, and the third causal convolution branch are 1, 2, and 4 respectively; Specifically, the first causal convolution branch is used to capture high-frequency transient events, such as excavation, etc.

[0033] Causal dilated convolution: , pad 2 zeros on the left , the output is: , where is the convolution kernel weight, the value of the input sequence at time step t-k .

[0034] The second causal convolution branch is used to capture medium-period events, such as walking, etc.

[0035] Causal dilated convolution: , pad 4 zeros on the left ( ) * 2 = 4), the output is: ; The third causal convolution branch is used to capture long-period events.

[0036] Causal dilated convolution: , pad 8 zeros on the left ( ) * 4 = 8), the output is: ; The features output by the first causal convolution branch, the second causal convolution branch, and the third causal convolution branch , and are respectively output after passing through the corresponding global average pooling (GAP) layer , , ; Specifically, the calculation of the global average pooling (GAP) layer is: , Then , , are input to the fully connected layer to dynamically generate the fusion weights : , where the input weight , satisfies , is the bias; The weighted sum outputs the multi-scale fusion feature: , Step 3012. Input the multi-scale fusion feature into the BatchNorm layer. In the BatchNorm layer, calculate the mean and variance on each feature dimension to obtain the normalized feature ; Step 3013. Perform global average pooling (GAP) on the normalized feature in Step 3012 to obtain a C-dimensional vector (C is the number of channels): , Step 3014. Then process this C-dimensional vector through a one-dimensional convolutional layer with a size of k . Here, k is the kernel size for cross-channel interaction. Apply the Sigmoid activation function to obtain the channel weights; Specifically, use a one-dimensional convolutional layer (kernel size k) to perform cross-channel interaction on the input first feature map, where k is determined by an adaptive function: , where C is the number of channels, and b and γ are hyperparameters, designed to ensure that models with different numbers of channels can maintain sufficient cross-channel interaction coverage.

[0037] Specifically, apply the Sigmoid activation function to obtain the channel weights as: , where are the weight parameters of the one-dimensional convolutional layer, which determine the importance of each input feature in cross-channel interaction, is the part related to the current channel in the C-dimensional vector obtained from the previous operations, used to calculate the channel weights; Finally, perform feature weighting. Multiply these weights channel by channel with the original feature map to obtain the enhanced feature map . Specifically, for the feature map in the original feature map c at the th channel, after weighting, the enhanced feature map is obtained, and its formula is: , where c = 1, 2,... C, and C is the number of channels of the feature map. Combine the weighted results of all channels to obtain the complete enhanced feature map.

[0038] Introduce the ECA module to perform weighted adjustment on the fused features. The ECA module calculates the importance of feature channels and dynamically adjusts the contribution of each channel to the final classification result, thereby further improving the recognition accuracy of the model. Through this mechanism, the model can focus more on key signal features and suppress the influence of redundant information, and finally output the classification result of vibration events.

[0039] Step 302. Input the single-channel time-domain data into the BiLSTM layer to extract the second feature map of each event; As Figure 10 shown, it includes the following steps: Step 3021. Input the single-channel time-domain data Segmented processing; Assume the segmented length T = 100: , where B is the batch processing dimension, T is the number of time steps, and 1 is the single-channel input feature dimension; Step 3022: Input the segmented single-channel time-domain data into the corresponding forward LSTM layer and backward LSTM layer respectively. Each forward LSTM layer processes the input sequence in the forward order, and each backward LSTM layer processes the input sequence in the reverse order. Finally, the vector formed by concatenating the two output vectors is used as the final feature representation, that is, the output result of the BiLSTM layer is obtained. ; In this embodiment, the forward LSTM layer and the backward LSTM layer adopt the existing LSTM layer structure.

[0040] Step 303: Fuse the first feature map and the second feature map of each event to obtain the third feature map of each event; As shown in Figure 11 , specifically, the local time-domain features extracted by MTCB-ECA are used as , and the global spatio-temporal features extracted by BiLSTM are used as and input into the gated spatio-temporal feature fusion mechanism. First, the local time-domain features and the global spatio-temporal features are linearly projected to the same dimension, and then the Sigmoid gating weight dynamically controls the ratio of the features extracted by the two modules: , where, is the encoded feature in space, is the encoded feature in time domain, is the feature weighted and fused according to the gating ratio; Compared with traditional feature concatenation, the dynamic gating mechanism can adaptively suppress the noise-dominated modes (such as instantaneous interference in vibration signals); automatically learn the modal correlation through end-to-end training without manual setting of weights.

[0041] Step 304: Normalize the third feature map to generate the classification probability distribution of vibration events; Specifically, use the LogSoftmax layer to normalize the third feature map to generate the classification probability distribution of vibration events, expressed as: , where, is the original score of the i th class, represents the sum over all classes, is the jThe original scores of each class.

[0042] When training the vibration event classification model, the training set data is input into the network model for training, and parameter optimization is carried out by adjusting parameters such as the training batch and learning rate. The weights with the best effect during the training process are saved for the intelligent monitoring of the optical cable.

[0043] Use the test set to test the trained network model. Precision, Recall, and F1Score are used to evaluate the performance of the classification model.

[0044] Among them, the calculation formula for Precision is: , The calculation formula for Recall is: , The calculation formula for F1Score is: , Among them, the category where the predicted label is a positive sample and the true label is also a positive sample is abbreviated as TP, the category where the predicted label is a positive sample and the true label is a negative sample is abbreviated as FP, the category where the predicted label is a negative sample and the true label is a positive sample is abbreviated as FN, and the category where the predicted label is a negative sample and the true label is a negative sample is abbreviated as TN.

[0045] Step 4: Use the trained vibration event classification model to classify the data to be detected to obtain the classification result of the vibration event.

[0046] In order to verify the classification effect of the method of the present invention, an ablation experiment comparison is carried out with the classification effect of the existing algorithms. The ablation experiment comparison results are shown in Table 2: Table 2 Ablation Experiment Comparison Results

[0047] It can be seen from Table 2 that the method of the present invention is superior to the classification effects of existing classifiers such as TCN, TCN-ECA, TCN-BiLSTM, TCN-Improve, and TCN-Improve-ECA models. It is verified that the present invention improves the event recognition accuracy and reduces the false alarm rate by improving the feature extraction ability of time series signals.

[0048] Embodiment 2 This embodiment provides a Φ-OTDR vibration event classification system based on feature fusion, including: A data acquisition module, which is used to acquire spatio-temporal data of multiple types of disturbance events in a phase-sensitive optical time domain reflectometry (Φ-OTDR) distributed optical fiber sensing system; A data preprocessing module, which is used to divide the spatio-temporal data matrix of multiple types of disturbance events after preprocessing into single-channel time-domain data of multiple types of disturbance events; A vibration event classification model training module, which is used to train the constructed vibration event classification model based on the single-channel time-domain data of multiple types of disturbance events to obtain a trained vibration event classification model; wherein, the construction process of the vibration event classification model includes: extracting a first feature map and a second feature map of each type of disturbance event based on the single-channel time-domain data; fusing the first feature map and the second feature map of each event to obtain a fused feature of each type of disturbance event, and classifying according to the fused feature; A classification module, which is used to classify the data to be detected based on the trained vibration event classification model to obtain a classification result.

[0049] It should be noted that the specific implementation manner of the Φ-OTDR vibration event classification system based on feature fusion in the embodiments of the present invention is similar to the specific implementation manner of the Φ-OTDR vibration event classification method based on feature fusion in the embodiments of the present invention. For details, please refer to the description in the method part. To reduce redundancy, it will not be elaborated here.

[0050] Embodiment III This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps in the Φ-OTDR vibration event classification method based on feature fusion as described above.

[0051] Embodiment IV This embodiment provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the Φ-OTDR vibration event classification method based on feature fusion as described above.

[0052] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A Φ-OTDR vibration event classification method based on feature fusion, characterized in that: The steps include: Acquire spatiotemporal data of multiple types of disturbance events in phase-sensitive optical time-domain reflectometry Φ-OTDR distributed fiber sensing systems; Divide the preprocessed spatiotemporal data matrix of multiple types of disturbance events into single-channel time domain data of multiple types of disturbance events; The vibration event classification model constructed is trained based on single-channel time domain data of multiple types of disturbance events to obtain a trained vibration event classification model; wherein the construction process of the vibration event classification model includes: A first characteristic map and a second characteristic map of each type of disturbance event are obtained based on single-channel time domain data extraction; The first feature map and the second feature map of each event are fused to obtain the fused features of each type of disturbance event, and classification is performed based on the fused features; The detection data is classified based on the trained vibration event classification model to obtain the classification result.

2. The Φ-OTDR vibration event classification method based on feature fusion as claimed in claim 1, characterized in that: The first characteristic diagram of each type of disturbance event extracted based on single-channel time domain data includes: Extract multi-scale features based on single-channel time domain data, and obtain multi-scale fusion features after fusion; The multi-scale fusion features are input into the BatchNorm layer, and the mean and variance of each feature dimension are calculated in the BatchNorm layer to obtain the standardized features; Globally pool the standardized features to obtain a C-dimensional vector; Then, through a size of k The one-dimensional convolution layer processes the C-dimensional vector, and applies the Sigmoid activation function to the processed result to obtain the channel weight; Multiply the channel weights by the original feature map channel by channel to obtain the enhanced feature map.

3. The Φ-OTDR vibration event classification method based on feature fusion as claimed in claim 2, characterized in that, The multi-scale features of single-channel time domain data using the multi-scale TCB module include: The multi-scale TCB module includes a parallel first causal convolution branch, a second causal convolution branch, and a third causal convolution branch; The first causal convolution branch takes as input single-channel time domain data and outputs high-frequency transient event features as The input of the second causal convolution branch is single-channel time domain data, and the output medium-periodic event feature is recorded as The input of the third causal convolution branch is single-channel time domain data, and the output long-period event feature is recorded as ; The features output by the first causal convolution branch, the second causal convolution branch, and the third causal convolution branch , and After passing through the corresponding global pooling GAP layer, the corresponding scale features are output.

4. The Φ-OTDR vibration event classification method based on feature fusion as claimed in claim 1, characterized in that: The second characteristic diagram of each type of disturbance event is obtained based on the single-channel time domain data extraction, including: Process the single-channel time domain data in segments; The segmented single-channel time domain data are input into the corresponding forward LSTM layer and reverse LSTM layer respectively. Each forward LSTM layer processes the input sequence in forward order, and each reverse LSTM layer processes the input sequence in reverse order. Finally, the vector formed by concatenating the two output vectors is used as the final feature.

5. The Φ-OTDR vibration event classification method based on feature fusion as claimed in claim 1, characterized in that: When classifying based on the fused features, the LogSoftmax layer is used to normalize the classification features and generate the classification probability distribution of the vibration events.

6. The Φ-OTDR vibration event classification method based on feature fusion as claimed in claim 1, characterized in that: The preprocessing of spatiotemporal data of multiple types of disturbance events includes normalization and first-order difference processing.

7. The Φ-OTDR vibration event classification system based on feature fusion is characterized by: include: A data acquisition module, which is used to acquire spatiotemporal data of multiple types of disturbance events in a phase-sensitive optical time domain reflectometry Φ-OTDR distributed optical fiber sensing system; A data preprocessing module, which is used to divide the preprocessed spatiotemporal data matrix of multiple types of disturbance events into single-channel time domain data of multiple types of disturbance events; A vibration event classification model training module is used to train the constructed vibration event classification model based on single-channel time domain data of multiple types of disturbance events to obtain a trained vibration event classification model; wherein the construction process of the vibration event classification model includes: extracting a first feature map and a second feature map of each type of disturbance event based on the single-channel time domain data; fusing the first feature map and the second feature map of each event to obtain a fusion feature of each type of disturbance event, and classifying according to the fusion feature; The classification module is used to classify the data to be detected based on the trained vibration event classification model to obtain a classification result.

8. The Φ-OTDR vibration event classification system based on feature fusion as claimed in claim 7, characterized in that: The Φ-OTDR distributed optical fiber sensing system includes an ultra-narrow linewidth laser, a signal generator, an acousto-optic modulator, an erbium-doped optical fiber amplifier, a circulator, a sensing optical fiber, a photodetector, a data acquisition card and a signal processing module; The output ends of the ultra-narrow linewidth laser and the signal generator are connected to the input end of the acousto-optic modulator, the output end of the acousto-optic modulator is connected to the input end of the erbium-doped fiber amplifier, the output end of the erbium-doped fiber amplifier is connected to one end of the circulator, the other end of the circulator is respectively connected to the input ends of the sensing fiber and the photodetector, the output end of the photodetector is connected to the input end of the data acquisition card, and the output end of the data acquisition card is connected to the signal processing module.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the Φ-OTDR vibration event classification method based on feature fusion as described in any one of claims 1 to 6 are implemented.

10. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the Φ-OTDR vibration event classification method based on feature fusion as described in any one of claims 1 to 6 are implemented.

Images