Hybrid neural network identification system and method for distributed optical fiber sound wave sensing events

Abstract

The invention belongs to the field of optical fiber sensing, and discloses a hybrid neural network identification system and method for distributed optical fiber sound wave sensing events. The system comprises a preliminary feature extraction module, a depth feature extraction module and an event recognition module. The preliminary feature extraction module comprises a preliminary feature extraction model based on a quantum neural network, the deep feature extraction module comprises a deep feature extraction model based on a deep learning neural network, and the event recognition module comprises an event recognition model based on the quantum neural network. The identification method mainly comprises the steps of extracting preliminary features of an intrusion event, extracting depth features and identifying the event according to the depth features. According to the method, the traditional neural network and the quantum neural network are combined, a complex data mode which is difficult to express by the traditional neural network can be captured, the recognition accuracy of the intrusion event is effectively improved, and the method has relatively high adaptability in a strong-noise complex environment, has a good intelligent level and is worthy of wide popularization and application.

Description

Technical Field

[0001] The present invention relates to the field of optical fiber sensing, and in particular to a hybrid neural network recognition system and method for distributed optical fiber acoustic wave sensing events. Background Art

[0002] A distributed fiber acoustic sensing (DAS) system uses optical fibers as sensing elements for distributed, multi-point monitoring. The DAS system exploits Rayleigh scattering in optical fibers to accurately sense and analyze acoustic signals. The basic principle is that when external vibrations occur along the optical fiber, the phase of the light wave at the corresponding location changes, and this phase change is linearly related to the external vibration. Because different types of external vibrations have distinct characteristics, data processing techniques can be used to demodulate the phase and identify the specific type of external vibration.

[0003] In perimeter security and oil and gas pipeline monitoring, DAS systems offer significant advantages over traditional monitoring methods, including long-distance transmission and immunity to electromagnetic interference. Intrusions typically generate different types of acoustic signals. By capturing and analyzing these signal characteristics, the type and location of the intrusion can be identified. Applying DAS systems to intrusion monitoring enables distributed, all-weather monitoring and early warning, reducing losses caused by intrusions.

[0004] Although DAS systems have great application potential in intrusion event monitoring, they still face some challenges in accurately identifying intrusion events and improving identification efficiency: 1. Currently, optical fibers are mainly buried in underground environments. Due to the complexity of the underground environment and the influence of external noise sources, they are easily interfered with sensor signals, resulting in signal quality degradation and increasing the difficulty of data processing. 2. The signals of intrusion events themselves are non-stationary, with large frequency variations and complex characteristics, making signal feature extraction difficult. Traditional time-frequency domain signal analysis methods are difficult to extract valuable features. 3. In actual applications, different types of intrusion events may generate similar acoustic wave signals. For example, the signals generated by excavation construction and the passage of certain large vehicles may be very similar in the low-frequency part. The subtle differences in these similar signals make it challenging to accurately identify different event types. 4. DAS systems have high requirements for real-time monitoring and generate huge amounts of data when monitoring over long distances. Existing technologies still have certain bottlenecks in processing large amounts of data quickly and efficiently, which affects the performance and application effectiveness of the system. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a hybrid neural network recognition system and method for distributed optical fiber acoustic wave sensing events, which can quickly and accurately identify intrusion events in strong background noise.

[0006] To solve the above technical problems, the present invention first proposes a hybrid neural network recognition system for distributed fiber optic acoustic wave sensing events, which includes a preliminary feature extraction module, a deep feature extraction module and an event recognition module; The preliminary feature extraction module includes a preliminary feature extraction model based on a quantum neural network, which is used to extract preliminary features of the intrusion event using the intrusion event data collected by the distributed optical fiber acoustic wave sensing system as input; The deep feature extraction module includes a deep feature extraction model based on a deep learning neural network, which is used to extract deep features of intrusion events using the preliminary features as input; The event recognition module includes an event recognition model based on a quantum neural network, which is used to identify corresponding intrusion events using the deep features of intrusion event data as input.

[0007] Furthermore, the preliminary feature extraction model, deep feature extraction model, and event recognition model need to be trained before they are officially used. The training method includes: A certain amount of training data is obtained as the input of the preliminary feature extraction model, so that the preliminary feature extraction model, the deep feature extraction model and the event recognition model are trained according to their respective model features until each model converges.

[0008] The preliminary feature extraction model, the deep feature extraction model and the event recognition model are trained using a cross entropy loss and a cosine annealing algorithm.

[0009] Optimally, the identification system further includes a data preprocessing module for performing denoising preprocessing on the intrusion event data collected by the distributed optical fiber acoustic wave sensing system before the intrusion event data is input into the preliminary feature extraction module.

[0010] The present invention also provides a hybrid neural network identification method for distributed optical fiber acoustic wave sensing events, comprising the following steps: The feature extraction module uses its preliminary feature extraction model based on quantum neural network to extract preliminary features of the intrusion event data collected by the distributed fiber optic acoustic wave sensing system; The deep feature extraction module uses its deep feature extraction model based on deep learning neural network to extract deep features of intrusion event data with the preliminary features as input; The event recognition model built on the quantum neural network in the event recognition module takes the deep features of the intrusion event data as input to identify the corresponding intrusion events.

[0011] Preferably, the preliminary feature extraction includes the following steps: First, use m (m=3,5,7) quantum bits , RY rotation quantum gate encoding is performed for each m data point, Perform RZ rotation quantum gate operation on the encoded m quantum bits. Applying controlled NOT gates to achieve entanglement between quantum bits, Perform RZ rotation quantum gate operation on m quantum bits, m qubits The Pauli expectation observations are the convolution output values of the quantum neural network for m different channels, i.e., the preliminary features of the intrusion event data.

[0012] Preferably, the event recognition model uses the deep features of the intrusion event data as input to identify the corresponding intrusion events, including the following steps: Using n quantum bits , corresponding to n types of intrusion events, flatten the deep features, The flattened depth features are output through the fully connected layer to obtain an n-dimensional vector. The corresponding n quantum bits are encoded by RY rotating quantum gate angle. Pairs of qubits Control the RZ rotation quantum gate operation to achieve mutual entanglement of quantum bit pairs; Perform RY rotation quantum gate operations on all qubits; Pairs of qubits Control the RZ rotation quantum gate operation, Measuring each qubit Pauli expected observation value, Compare these n probability values and select the event corresponding to the largest probability value as the final recognition result of the system.

[0013] Further optimization is to perform denoising preprocessing on the intrusion event data before it is input into the preliminary feature extraction module.

[0014] Preferably, the denoising preprocessing includes spectral subtraction and wavelet packet denoising methods.

[0015] Compared with the prior art, the present invention has the following beneficial effects: (1) At present, most DAS intrusion event identification methods mainly rely on traditional machine learning or deep learning technology. However, in an environment with strong background noise, the recognition accuracy is low, and there are difficulties in identifying events with small feature differences or similar events. The present invention includes data preprocessing and noise reduction, hybrid quantum neural network feature extraction, and quantum neural network output. By introducing quantum neural networks and utilizing quantum properties such as quantum bit entanglement, it can more accurately capture the key features in the signal in noisy signals, enhance the sensitivity to different events, and thus make efficient and accurate classifications in complex environments. The present invention can be used in important scenarios such as perimeter intrusion warning and monitoring of oil and gas pipelines. By laying monitoring optical cables and collecting vibration signals in real time, it can accurately identify external intrusion events such as mechanical excavation and manual construction, providing more reliable protection for public safety.

[0016] (2) The present invention combines spectral subtraction and wavelet packet denoising technology to enhance the noise suppression effect. Spectral subtraction can suppress background noise to a certain extent and improve the signal-to-noise ratio of the signal by processing the signal in the frequency domain and estimating the noise amplitude; wavelet packet denoising can further decompose the signal in the time domain and remove unnecessary noise. When processing signals that have been processed by spectral subtraction and wavelet packet technology for noise reduction, quantum neural networks can better capture the key features in the signal through quantum superposition states, reduce the impact of noise on recognition results, and improve recognition accuracy and stability in strong background noise environments.

[0017] (3) The present invention combines quantum neural networks in feature extraction. Through the quantum gate encoding and rotation quantum gate operation of the quantum neural network, it can effectively extract the potential laws and complex features of the data, process high-dimensional data features that traditional neural networks cannot fully explore, and avoid the problem of feature loss in high-dimensional data analysis. In particular, it has significant feature extraction advantages for non-stationary signals collected by DAS systems.

[0018] (4) The present invention uses n qubits in the quantum neural network output module, corresponding to n intrusion events, and achieves efficient entanglement between qubits by controlling the RY rotation quantum gate and the RZ rotation quantum gate. The final model compares the probability values of the n events and selects the event corresponding to the maximum probability as the recognition result. This entanglement between qubits enhances the system's sensitivity to subtle differences, thereby enabling the distinction between events with similar signals and improving the accuracy and robustness of classification.

[0019] (5) The present invention optimizes the learning rate by introducing the cosine annealing algorithm, making the training process more efficient. The cosine annealing algorithm can dynamically adjust the learning rate according to the different stages of the training process, avoiding overfitting while accelerating the convergence of the model. This allows quantum neural networks to be trained efficiently in a short period of time, improving the system's response speed and computational efficiency in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0021] Figure 1 Schematic diagram of the hardware layout of the distributed fiber optic acoustic wave sensing system involved in the present invention.

[0022] Figure 2 This is an organizational structure diagram of the main functional modules of the hybrid neural network recognition system for distributed optical fiber acoustic wave sensing events of the present invention.

[0023] Figure 3 This is a structural diagram of the relationship between the preliminary feature extraction model based on quantum neural network, the deep feature extraction model based on deep learning neural network, and the event recognition model of quantum neural network of the present invention.

[0024] Figure 4 This is an overall flow chart of the hybrid neural network identification method for distributed optical fiber acoustic wave sensing events provided by the present invention.

[0025] Figure 5 Schematic diagram of the mechanism of the preliminary feature extraction model for intrusion events based on quantum neural networks provided by the present invention.

[0026] Figure 6 Schematic diagram of the intrusion event recognition model mechanism based on quantum neural network provided by the present invention. DETAILED DESCRIPTION

[0027] Generally, the present invention relates to the hardware arrangement of a distributed fiber acoustic sensing (DAS) system, such as Figure 1 As shown in the figure, the distributed fiber acoustic sensing (DAS) system, tested at different times and locations on the test fiber, includes: a narrow-linewidth laser 1, a 10:90 fiber coupler 2, an acousto-optic modulator 3, an erbium-doped fiber amplifier 4, a fiber Bragg grating 5, a circulator 6, an erbium-doped fiber amplifier 7, a fiber Bragg grating 8, a 50:50 fiber coupler 9, a balanced photodetector 10, and a data acquisition card 11. The DAS system collects data from various types of mechanical construction, human operations, and other behaviors that could interfere with or threaten the monitored area in scenarios such as perimeter security and oil and gas pipeline monitoring. The collected data is then divided into training and test sets.

[0028] When collecting signals, the output light of the narrow linewidth laser 1 is divided into two optical paths through the fiber coupler 2. The light with an energy share of 90% is used as the detection light, and the light with an energy share of 10% is used as the local oscillator light. The acousto-optic modulator 3 modulates the detection light into pulses, and the light pulses pass through the erbium-doped fiber amplifier 4 for energy amplification. The fiber Bragg grating 5 is used to filter out the amplified spontaneous radiation of the erbium-doped fiber amplifier 4. Then, the detection light enters the test fiber through the fiber circulator 6. The reflected back-Rayleigh scattered light passes through the fiber circulator 6 and is amplified by the erbium-doped fiber amplifier 7. The fiber Bragg grating 8 is used to filter out the amplified spontaneous radiation of the erbium-doped fiber amplifier 7. Next, the back-Rayleigh scattered light and the local oscillator light are coupled in the 50:50 fiber coupler 9 and pass through the balanced photodetector 10. The data acquisition process is then completed by the data acquisition card 11.

[0029] like Figure 2 , combined Figure 3 As shown, the hybrid neural network recognition system for distributed optical fiber acoustic wave sensing events of the present invention includes a data preprocessing module, a preliminary feature extraction module, a deep feature extraction module and an event recognition module.

[0030] The data preprocessing module is used to perform denoising preprocessing on the intrusion event data collected by the distributed optical fiber acoustic wave sensing system before the intrusion event data is input into the preliminary feature extraction module.

[0031] A preliminary feature extraction module, including a preliminary feature extraction model based on a quantum neural network, for extracting preliminary features of the intrusion event using the intrusion event data collected by the distributed fiber optic acoustic wave sensing system as input; The deep feature extraction module includes a deep feature extraction model based on a deep learning neural network, which is used to extract deep features of intrusion events using the preliminary features as input; The event recognition module includes an event recognition model based on a quantum neural network, which is used to identify corresponding intrusion events using the deep features of intrusion event data as input.

[0032] The preliminary feature extraction model, deep feature extraction model, and event recognition model must be trained before formal use. The training method includes: using a certain amount of training data as input to the preliminary feature extraction model, so that the preliminary feature extraction model, deep feature extraction model, and event recognition model are trained according to their respective model operation mechanisms until each model converges. During the training process, cross entropy loss and cosine annealing algorithms are used. Generally speaking, the accuracy of event recognition reaches 90% when the training reaches convergence.

[0033] The present invention also provides a hybrid neural network recognition method for distributed optical fiber acoustic wave sensing events, such as Figure 4 As shown, the following steps are included: Step 1: Pre-process the intrusion event data collected by the distributed fiber optic acoustic wave sensing system, including: spectral subtraction and wavelet packet denoising to improve the signal-to-noise ratio of the signal. Perform a fast Fourier transform, where For the Frame, get its spectrum The amplitude spectrum and phase spectrum of the signal are expressed as: ; Then the amplitude spectrum is processed to estimate the noise amplitude to obtain the amplitude spectrum after noise reduction , Indicates the The noise amplitude estimate of the frame signal, It is a parameter that controls the intensity of noise reduction, which converts the processed amplitude spectrum and the phase spectrum of the original signal into Combine to get the spectrum after noise reduction ,Finally, the denoised spectra of all frames are inversely transformed to obtain the denoised signal.

[0034] Wavelet packet denoising uses the Db2 wavelet to perform a three-layer wavelet packet decomposition of the signal. Each layer of decomposition separates the signal into two components: a low-frequency component and a high-frequency component. A hard threshold is set, and all wavelet coefficients below the threshold are set to zero, while the other coefficients remain unchanged. Finally, the remaining signal coefficients are used to reconstruct the signal, resulting in the denoised signal.

[0035] Step 2: Preliminary feature extraction module, which extracts preliminary features from the denoised intrusion event data. Figure 5 As shown in the figure, in the initial feature extraction, the quantum neural network first uses m (m=3,5,7) quantum bits , RY rotation quantum gate encoding is performed for each m data point. The matrix form of the RY rotation quantum gate is as follows: ; in, It is the data point The calculated angle is calculated as follows: ; Then, the RZ rotation quantum gate operation is performed on the encoded m quantum bits. The matrix form of the RZ rotation quantum gate is as follows: ; in The parameters that can be learned by training quantum neural networks are then applied to achieve entanglement between qubits. The CNOT gate is a quantum entanglement operation that usually consists of two qubits: a control qubit and a target qubit. If the control qubit is in the state , then the state of the target quantum bit will flip. Apply the CNOT gate to the quantum bits in turn. (control qubits) and (target qubit) on ( ). In this way, entanglement can be created between the quantum bits.

[0036] Then perform RZ rotation quantum gate operation on m qubits. The Pauli expectation observations are the convolution output values of the quantum neural network for m different channels.

[0037] Step 3: The deep feature extraction module uses its deep feature extraction model based on a deep learning neural network to extract deep features from the intrusion event data. Specifically, the preliminary features processed by the preliminary feature extraction model are input into the deep feature extraction model of the deep learning neural network for further convolution to extract deeper features, or deep features.

[0038] The deep learning neural network consists of multiple residual convolution blocks and attention mechanisms. The residual convolution block consists of repeated one-dimensional convolution, batch normalization layer, ReLU activation function, and maximum pooling.

[0039] After the stacked residual convolution blocks, a one-dimensional attention mechanism is applied. This attention mechanism consists of a channel attention mechanism and a spatial attention mechanism. The channel attention mechanism uses adaptive average pooling and max pooling operations to extract global channel information. The pooled results are processed through a one-dimensional convolution to further compress the information dimension. The ReLU activation function introduces nonlinearity, and the final channel weight is obtained through the Sigmoid function. The spatial attention mechanism uses a one-dimensional convolution to weight the input spatially to capture spatial features, and the Sigmoid activation function is used to calculate the attention weight for each position.

[0040] Step 4: Flatten the extracted deep features, then use the fully connected layer to reduce the dimension to n-dimensional, and then enter the event recognition model based on quantum neural network, which performs event recognition processing. Figure 6 As shown in the figure, the process includes: the quantum neural network output module uses n quantum bits, which correspond to n intrusion events in specific scenarios such as oil and gas pipeline monitoring and perimeter security monitoring, and performs RY rotation quantum gate angle encoding on the n-dimensional features, and then encodes the quantum bit pairs. 、 、 、 、 Perform a controlled RZ rotation quantum gate operation. The first qubit in the qubit pair is the control qubit, and the second qubit is the target qubit. The target qubit performs an RZ rotation quantum gate operation based on the state of the control qubit, and the rotation angle is determined by a parameter that can be learned by the quantum neural network. Determine and realize the mutual entanglement of quantum bit pairs.

[0041] Then perform RY rotation quantum gate operation on all qubits; then perform RY rotation quantum gate operation on qubit pairs. 、 、 、 、 Control the RZ rotation quantum gate operation. Finally, measure the The Pauli expected observation value, where the expected value of each qubit measured represents the probability of the corresponding event. By comparing these n values, the event type corresponding to the maximum value is selected as the final event recognition output.

[0042] In summary, the present invention combines quantum neural networks with traditional neural networks, and its main advantages are: (1) effectively improving the signal-to-noise ratio of the signal, enhancing the identifiability of intrusion events, and having good adaptability to complex environments in the real world; (2) being able to capture complex features that are difficult to obtain with classical neural networks, and improving the ability to represent features; (3) through the superposition and entanglement characteristics of quantum neural networks, it is able to more finely capture subtle differences in signals, effectively improving the system's ability to identify similar event signals; (4) through the quantum bits of the quantum neural network output module and the rotation quantum gate operation, fast and accurate classification is achieved, improving the accuracy and response speed of intrusion event identification.

[0043] Finally, it should be noted that the above specific implementation methods are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A hybrid neural network recognition system for distributed fiber optic acoustic wave sensing events, characterized in that: It includes preliminary feature extraction module, deep feature extraction module and event recognition module; The preliminary feature extraction module includes a preliminary feature extraction model based on a quantum neural network, which is used to extract preliminary features of the intrusion event using the intrusion event data collected by the distributed optical fiber acoustic wave sensing system as input; The deep feature extraction module includes a deep feature extraction model based on a deep learning neural network, which is used to extract deep features of intrusion events using the preliminary features as input; The event recognition module includes an event recognition model based on a quantum neural network, which is used to identify corresponding intrusion events using the deep features of intrusion event data as input.

2. The hybrid neural network recognition system for distributed optical fiber acoustic wave sensing events according to claim 1, characterized in that: The preliminary feature extraction model, deep feature extraction model, and event recognition model need to be trained before they are officially used. The training method includes: A certain amount of training data is obtained as the input of the preliminary feature extraction model, so that the preliminary feature extraction model, the deep feature extraction model and the event recognition model are trained according to their respective model features until each model converges.

3. The hybrid neural network recognition system for distributed optical fiber acoustic wave sensing events according to claim 1, characterized in that: The preliminary feature extraction model, the deep feature extraction model and the event recognition model are trained using a cross entropy loss and a cosine annealing algorithm.

4. The hybrid neural network recognition system for distributed optical fiber acoustic wave sensing events according to claim 1, characterized in that: The identification system further comprises a data preprocessing module for performing denoising preprocessing on the intrusion event data collected by the distributed optical fiber acoustic wave sensing system before the intrusion event data is input into the preliminary feature extraction module.

5. A hybrid neural network recognition method based on the distributed optical fiber acoustic wave sensing event of claim 1, characterized in that: The following steps are involved: The feature extraction module uses its preliminary feature extraction model based on quantum neural network to extract preliminary features of the intrusion event data collected by the distributed fiber optic acoustic wave sensing system; The deep feature extraction module uses its deep feature extraction model based on deep learning neural network to extract deep features of intrusion event data with the preliminary features as input; The event recognition model built on the quantum neural network in the event recognition module takes the deep features of the intrusion event data as input to identify the corresponding intrusion events.

6. The hybrid neural network identification method for distributed optical fiber acoustic wave sensing events according to claim 5, characterized in that: The preliminary feature extraction includes the following steps: First, use m (m=3,5,7) quantum bits , RY rotation quantum gate encoding is performed for each m data point, Perform RZ rotation quantum gate operation on the encoded m quantum bits. Applying controlled NOT gates to achieve entanglement between quantum bits, Perform RZ rotation quantum gate operation on m quantum bits, m qubits -Pauli expectation observations are respectively used as the convolution output values of the quantum neural network of m different channels, i.e., the preliminary features of the intrusion event data.

7. The hybrid neural network identification method for distributed optical fiber acoustic wave sensing events according to claim 6, characterized in that: The event recognition model uses the deep features of intrusion event data as input to identify corresponding intrusion events, including the following steps: Using n quantum bits , corresponding to n types of intrusion events, flatten the deep features, The flattened depth features are output through the fully connected layer to obtain an n-dimensional vector. The corresponding n quantum bits are encoded by RY rotating quantum gate angle. Pairs of qubits 、 、 、 、 Control the RZ rotation quantum gate operation to achieve mutual entanglement of quantum bit pairs; Perform RY rotation quantum gate operations on all qubits; Pairs of qubits 、 、 、 、 Control the RZ rotation quantum gate operation, Measuring each qubit - Pauli expected observation value, Compare these n probability values and select the event corresponding to the largest probability value as the final recognition result of the system.

8. The hybrid neural network identification method for distributed optical fiber acoustic wave sensing events according to claim 5, characterized in that: Before the intrusion event data is input into the preliminary feature extraction module, it is pre-processed for denoising.

9. The hybrid neural network identification method for distributed optical fiber acoustic wave sensing events according to claim 8, characterized in that: The denoising preprocessing includes spectrum subtraction and wavelet packet denoising methods.

Images