Distributed acoustic wave sensor data compression method, apparatus, equipment and storage medium

Abstract

A method, apparatus, device, and storage medium for compressing distributed acoustic wave sensing data are disclosed. The method includes: inputting the original beat frequency signal into a pre-trained encoder module, which directly outputs a low-dimensional compressed feature vector, thus achieving efficient compression of distributed acoustic wave sensing data. The encoder module, through end-to-end training, learns how to map the high-dimensional original signal into a low-dimensional feature vector, preserving the key physical features of the signal in the process. Compared to traditional fixed downsampling or threshold compression, this application can adapt to the dynamic characteristics of the signal, effectively maintaining signal fidelity while compressing the data volume, thereby reducing the burden of subsequent storage, transmission, and computational processing.

Description

Technical Field

[0001] This application relates to the field of fiber optic sensing and signal processing technology, specifically to a distributed acoustic wave sensing data compression method, apparatus, device, and computer-readable storage medium. Background Technology

[0002] Distributed acoustic wave sensing systems based on phase-sensitive optical time-domain reflectometry (Φ-OTDR) are widely used in geological exploration, perimeter security, and traffic monitoring due to their long sensing distance, high sensitivity, and scalability. In these systems, to satisfy the Nyquist sampling theorem, sampling rates of 400 to 600 MSa / s are often required when acquiring beat frequency signals with center frequencies typically ranging from tens to hundreds of MHz. This high sampling rate, combined with high-resolution acquisition typically ranging from 8 to 16 bits, leads to the generation of massive amounts of time-series data, resulting in significant hardware costs and computational burdens for data storage, transmission, and real-time processing.

[0003] Existing distributed acoustic sensing data compression frameworks typically follow a "compression-decompression-processing" model. For example, dictionary-based lossless compression algorithms (such as LZW) have limited compression ratios; while control-precision compression frameworks like DASPack can improve compression ratios, subsequent processing still requires the recovery of high-speed waveforms, failing to reduce online processing latency. Traditional direct downsampling or undersampling techniques can significantly compress data, but due to their fixed sampling and quantization settings, they are difficult to adapt to different noise levels and bandwidth limitations, often leading to a decrease in signal-to-noise ratio and the loss of subtle vibration characteristics, making it difficult to simultaneously achieve high compression ratios and effective preservation of subtle vibration characteristics. Summary of the Invention

[0004] This application provides a distributed acoustic wave sensing data compression method, apparatus, device, and computer-readable storage medium, which can solve the technical problem of difficulty in balancing high compression ratio and signal fidelity in the prior art.

[0005] In a first aspect, embodiments of this application provide a distributed acoustic wave sensing data compression method, the distributed acoustic wave sensing data compression method comprising: Obtain the raw beat frequency signal to be compressed; The original beat frequency signal is input into a pre-trained encoder module to obtain a low-dimensional compressed feature vector output by the encoder module, which is then used for subsequent processing. The encoder module is trained in the following way: a neural network model containing an encoder module and a decoder module is trained with the goal of minimizing the composite loss function obtained from the input of the encoder module and the output of the decoder module, thereby obtaining a pre-trained encoder module.

[0006] In conjunction with the first aspect, in one embodiment, the neural network model includes an encoder module and a decoder module. The encoder module includes a first convolutional layer, a first residual block, a pooling layer, a second convolutional layer, and a second residual block connected in sequence. The decoder module includes a third convolutional layer, a third residual block, an upsampling layer, a fourth convolutional layer, and a fourth residual block connected in sequence. The training process of the encoder module includes: The beat frequency signal used for training is obtained as a training sample. The training sample is input into the encoder module and processed by the first convolutional layer, the first residual block, the pooling layer, the second convolutional layer and the second residual block to output a compressed feature vector. The compressed feature vector is input into the decoder module, and after processing by the third convolutional layer, the third residual block, the upsampling layer, the fourth convolutional layer, and the fourth residual block, the reconstructed signal is output. Calculate the composite loss function between the training samples and the reconstructed signal, wherein the composite loss function includes at least two of the following: time-domain loss function, frequency-domain loss function, phase loss function, and gradient loss function; The trainable parameters of the encoder module and the decoder module are updated simultaneously using the backpropagation algorithm to minimize the composite loss function. Repeat the above steps iteratively until the convergence condition is met to obtain the trained encoder module.

[0007] In conjunction with the first aspect, in one implementation, each residual block includes a fifth convolutional layer, a first activation function layer, a Dropout layer, a sixth convolutional layer, an addition module, and a second activation function layer connected in sequence, wherein: The fifth convolutional layer performs a convolution operation on the input of the residual block; The first activation function layer applies a non-linear activation to the output of the fifth convolutional layer; The Dropout layer randomly sets some neurons in the output of the first activation function layer to zero with a preset probability, while keeping the rest unchanged, in order to prevent overfitting. The sixth convolutional layer performs a convolution operation on the output of the Dropout layer; The addition module adds the input of the residual block to the output of the sixth convolutional layer element by element; The second activation function layer applies a nonlinear activation to the output of the addition module, which serves as the final output of the residual block.

[0008] In conjunction with the first aspect, in one implementation, the first / second activation function layer employs a parameterized modified linear unit, which has a learnable slope parameter on the negative half-axis.

[0009] In conjunction with the first aspect, in one embodiment, after obtaining the low-dimensional compressed feature vector output by the encoder module, the method further includes: The low-dimensional compressed feature vector is quantized to reduce the data bit width.

[0010] In conjunction with the first aspect, in one implementation, the quantization process employs a μ-law quantization algorithm.

[0011] In conjunction with the first aspect, in one implementation, the subsequent processing includes: The low-dimensional compressed feature vector is sequentially subjected to frequency offset compensation and low-pass filtering to extract the baseband signal; After grouping and reconstructing the baseband signal, a rotating vector summation algorithm is used to eliminate the effects of interference fading. Phase extraction and unwrapping are performed on the signal processed by the rotating vector summation algorithm, and vibration information is restored by calculating the phase difference.

[0012] Secondly, embodiments of this application provide a distributed acoustic wave sensing data compression device, the distributed acoustic wave sensing data compression device comprising: The acquisition module is used to acquire the raw beat frequency signal to be compressed; The compression module is used to input the original beat frequency signal into the pre-trained encoder module and obtain the low-dimensional compressed feature vector output by the encoder module for subsequent processing based on the low-dimensional compressed feature vector. The encoder module is trained in the following way: a neural network model containing an encoder module and a decoder module is trained with the goal of minimizing the composite loss function obtained from the input of the encoder module and the output of the decoder module, thereby obtaining a pre-trained encoder module.

[0013] Thirdly, embodiments of this application provide a distributed acoustic wave sensor data compression device, which includes a processor, a memory, and a distributed acoustic wave sensor data compression program stored in the memory and executable by the processor. When the distributed acoustic wave sensor data compression program is executed by the processor, it implements the steps of the distributed acoustic wave sensor data compression method as described in the first aspect.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing a distributed acoustic wave sensing data compression program, wherein when the distributed acoustic wave sensing data compression program is executed by a processor, it implements the steps of the distributed acoustic wave sensing data compression method as described in the first aspect.

[0015] The beneficial effects of the technical solutions provided in this application include: In this embodiment, efficient compression of distributed acoustic wave sensing data is achieved by inputting the original beat frequency signal into a pre-trained encoder module and directly outputting a low-dimensional compressed feature vector. The encoder module, through end-to-end training, learns how to map the high-dimensional original signal into a low-dimensional feature vector, preserving the key physical characteristics of the signal in the process. Compared to traditional fixed downsampling or threshold compression, this embodiment can adapt to the dynamic characteristics of the signal, effectively maintaining signal fidelity while compressing the data volume, thereby reducing the burden of subsequent storage, transmission, and computational processing. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an embodiment of the distributed acoustic wave sensor data compression method of this application; Figure 2 This is a schematic diagram of the architecture of a distributed acoustic wave sensing system in one embodiment; Figure 3 This is a schematic diagram of a signal compression and reconstruction architecture based on a residual neural network in one embodiment; Figure 4 This is a flowchart of a signal processing and demodulation method in one embodiment; Figure 5 This is a schematic diagram showing the location of the vibration point in one embodiment; Figure 6 This is a time-domain diagram of the demodulated vibration point signal in one embodiment; Figure 7 This is a schematic diagram of the functional modules of an embodiment of the distributed acoustic wave sensor data compression device of this application; Figure 8 This is a schematic diagram of the hardware structure of the distributed acoustic wave sensing data compression device involved in the embodiments of this application. Detailed Implementation

[0017] 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 are within the scope of protection of the present application.

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0019] In a first aspect, embodiments of this application provide a distributed acoustic wave sensing data compression method.

[0020] In one embodiment, reference is made to Figure 1 , Figure 1 This is a schematic flowchart of an embodiment of the distributed acoustic wave sensor data compression method of this application. Figure 1 As shown, the distributed acoustic wave sensor data compression method includes: Step S10: Obtain the original beat frequency signal to be compressed; Step S20: Input the original beat frequency signal into the pre-trained encoder module to obtain the low-dimensional compressed feature vector output by the encoder module, so as to perform subsequent processing based on the low-dimensional compressed feature vector; The encoder module is trained in the following way: a neural network model containing an encoder module and a decoder module is trained with the goal of minimizing the composite loss function obtained from the input of the encoder module and the output of the decoder module, thereby obtaining a pre-trained encoder module.

[0021] In this embodiment, the distributed acoustic wave sensing data specifically refers to the beat frequency signal output by the balanced photodetector in the distributed acoustic wave sensing system and digitized by the data acquisition card. The beat frequency signal is the original signal carrier carrying external vibration information. (Refer to...) Figure 2 , Figure 2 This is a schematic diagram of the architecture of a distributed acoustic wave sensing system in one embodiment. Figure 2 As shown, the light source is a narrow linewidth laser (NLL) with a center wavelength of 1550 nm. The output light is split into two paths after passing through an isolator (ISO) and a 90:10 coupler (OC1). 90% of the probe light is modulated by an acousto-optic modulator (AOM) to generate a 200MHz frequency-shifted pulse, which is then injected into the fiber under test (FUT) through an erbium-doped fiber amplifier (EDFA) and a circulator. At about 1.65 km in the fiber, a piezoelectric ceramic (PZT) driven by a signal generator (SG) is placed as a simulated vibration source. The returned Rayleigh backscattered light is amplified and filtered, and beats with the reference light at a 50:50 coupler (OC2). Finally, it is received by a balanced photodetector (BPD) and digitally acquired by a data acquisition card (DAQ). An arbitrary waveform generator (AWG) serves as the core of the system's timing control. One output is connected to an acousto-optic modulator (AOM) to generate electrical pulse signals with specific pulse widths and repetition frequencies, driving the AOM to modulate continuous probe light into pulsed light. The other output of the AWG is connected to the trigger input of a data acquisition card (DAQ) to send a synchronization trigger signal to the DAQ, ensuring precise timing synchronization between data acquisition and optical pulse emission, thereby achieving accurate positioning of scattered signals at various locations along the optical fiber. The sampling rate of the data acquisition card is, for example, 500 MSa / s, and the sampling bit width is, for example, 8 bits.

[0022] In one application scenario, the beat frequency signal collected by the distributed acoustic wave sensing system is acquired as the original beat frequency signal to be compressed. The original beat frequency signal is then input into a pre-trained encoder module. This encoder module has learned how to map the high-dimensional original signal into a low-dimensional feature vector through pre-training. The low-dimensional compressed feature vector output by the encoder module can then be obtained for subsequent processing based on the low-dimensional compressed feature vector.

[0023] Furthermore, in one embodiment, after obtaining the low-dimensional compressed feature vector output by the encoder module, the method further includes: The low-dimensional compressed feature vector is quantized to reduce the data bit width.

[0024] In this embodiment, the encoder module compresses the data volume of the original beat frequency signal in the time dimension. Furthermore, to further reduce the data bit width, the low-dimensional compressed feature vector is quantized. The quantization process employs a μ-law quantization algorithm.

[0025] As described above, the key to achieving distributed acoustic wave sensing data compression lies in the pre-trained encoder module. The pre-trained encoder module is obtained by training a neural network model that includes an encoder module and a decoder module, with the goal of minimizing the composite loss function obtained from the input of the encoder module and the output of the decoder module.

[0026] Further, in one embodiment, the neural network model includes an encoder module and a decoder module. The encoder module includes a first convolutional layer, a first residual block, a pooling layer, a second convolutional layer, and a second residual block connected in sequence. The decoder module includes a third convolutional layer, a third residual block, an upsampling layer, a fourth convolutional layer, and a fourth residual block connected in sequence. The training process of the encoder module includes: The beat frequency signal used for training is obtained as a training sample. The training sample is input into the encoder module and processed by the first convolutional layer, the first residual block, the pooling layer, the second convolutional layer and the second residual block to output a compressed feature vector. The compressed feature vector is input into the decoder module, and after processing by the third convolutional layer, the third residual block, the upsampling layer, the fourth convolutional layer, and the fourth residual block, the reconstructed signal is output. Calculate the composite loss function between the training samples and the reconstructed signal, wherein the composite loss function includes at least two of the following: time-domain loss function, frequency-domain loss function, phase loss function, and gradient loss function; The trainable parameters of the encoder module and the decoder module are updated simultaneously using the backpropagation algorithm to minimize the composite loss function. Repeat the above steps iteratively until the convergence condition is met to obtain the trained encoder module.

[0027] In this embodiment, refer to Figure 3 , Figure 3 This is a schematic diagram of a signal compression and reconstruction architecture based on a residual neural network in one embodiment. Based on... Figure 3 The training process for one embodiment of the architecture shown is as follows: The beat frequency signal used for training is acquired as training samples. These training samples are divided into normalized signal segments of fixed length (4096 sampling points each), which serve as the input tensor of the encoder module, denoted as... ,in, For batch size.

[0028] The data processing flow of the encoder module is as follows: Will Input the first convolutional layer, which uses 16 convolutional kernels with a preset kernel size (e.g., 7) and a stride of 1. The first convolutional layer will... The number of channels is expanded from 1 to 16 to extract local features and output feature maps. ; feature map Input the first residual block, process it, and output the result. Each residual block comprises a fifth convolutional layer, a first activation function layer, a Dropout layer, a sixth convolutional layer, an addition module, and a second activation function layer, connected in sequence, wherein: The fifth convolutional layer performs a convolution operation on the input of the residual block; The first activation function layer applies a non-linear activation to the output of the fifth convolutional layer; The Dropout layer randomly sets some neurons in the output of the first activation function layer to zero with a preset probability, while keeping the rest unchanged, in order to prevent overfitting. The sixth convolutional layer performs a convolution operation on the output of the Dropout layer; The addition module adds the input of the residual block to the output of the sixth convolutional layer element by element; The second activation function layer applies a nonlinear activation to the output of the addition module, which serves as the final output of the residual block.

[0029] Process the feature map using the first residual block. For example: Fifth convolutional layer pair Perform convolution operations to output intermediate feature maps; The first activation function layer applies nonlinear activation to the intermediate feature map. The first / second activation function layer uses a parameterized modified linear unit (PReLU), which has a learnable slope parameter on the negative half-axis to adapt to large dynamic changes in the signal.

[0030] The Dropout layer randomly sets some neurons in the output of the first activation function layer to zero with a preset probability, while leaving the rest unchanged, in order to prevent overfitting. The sixth convolutional layer performs a convolution operation on the output of the Dropout layer to restore it to the number of input channels; The addition module will input the first residual block. Add element-wise to the output of the sixth convolutional layer; The second activation function layer applies a nonlinear activation to the output of the addition module, which becomes the final output of the first residual block. .

[0031] Output of the first residual block The pooling layer uses a pooling window size of 2 and a stride of 2. It selects the maximum value within each window in the time dimension to achieve data downsampling and output a feature map. ; Feature map The second convolutional layer uses 32 kernels, expanding the number of channels from 16 to 32, to extract higher-level abstract features and output a feature map. ; Feature map Entering the second residual block (whose internal structure is the same as the first residual block), the same feature transformation and nonlinear mapping are performed, and the feature map is output. ; At this point, the encoder module outputs the sample compressed feature vector. Its time dimension was reduced from the original 4096 to 2048, achieving a 2x compression.

[0032] The decoder module is used to restore the sample compressed feature vectors to the original signal dimension. It is only used during the training phase, and its data processing flow is as follows: Sample compressed feature vector The third convolutional layer uses 16 kernels, reducing the number of channels from 32 to 16. This layer performs preliminary feature deconstruction and mapping on the compressed features, outputting a feature map. ; Feature map Entering the third residual block (whose internal structure is the same as the first residual block), perform nonlinear transformation and enhancement of the features, and output the feature map. ; The signal then enters the upsampling layer, which uses nearest neighbor interpolation or linear interpolation to enlarge the time dimension from 2048 to 4096, thereby restoring the signal dimension and outputting a feature map. ; Feature map The fourth convolutional layer uses a single kernel, reducing the number of channels from 16 to 1, and outputs a feature map. ; Feature map Entering the fourth residual block (whose internal structure is the same as the first residual block), the final feature transformation and nonlinear mapping are performed, and the reconstructed signal is output. .

[0033] Then, the input tensor With reconstruction signal Input loss calculation module to calculate composite loss ,For example: calculate and The mean square error on the time-domain waveform yields the time-domain loss. ; right and Perform Fourier transforms on each spectrum and calculate the mean square error between the spectral amplitudes to obtain the frequency domain loss. ; Extract separately and The in-phase component (I) and quadrature component (Q), for and The phase loss is obtained by summing the squares of the I / Q component errors. ; calculate and The difference in the time gradient domain yields the gradient loss. .

[0034] right , , as well as We perform a weighted summation to obtain the compound loss. .

[0035] The calculated composite loss Perform backpropagation: The gradient first flows through the decoder module, updating all trainable parameters (convolutional kernel weights, PReLU learnable slope parameters, etc.) in the third-level convolutional layer, the third residual block, the fourth-level convolutional layer, and the fourth residual block. The gradient then continues to propagate back to the encoder module, updating all trainable parameters in the first-level convolutional layer, the first residual block, the second-level convolutional layer, and the second residual block. The above process is iteratively executed using a gradient descent optimization algorithm (such as Adam) until the composite loss converges to a preset threshold or reaches a preset number of iterations.

[0036] At this point, the trained encoder module is obtained.

[0037] Furthermore, in one embodiment, the subsequent processing includes: The low-dimensional compressed feature vector is sequentially subjected to frequency offset compensation and low-pass filtering to extract the baseband signal; After grouping and reconstructing the baseband signal, a rotating vector summation algorithm is used to eliminate the effects of interference fading. Phase extraction and unwrapping are performed on the signal processed by the rotating vector summation algorithm, and vibration information is restored by calculating the phase difference.

[0038] In this embodiment, refer to Figure 4 , Figure 4 This is a flowchart of a signal processing and demodulation method in one embodiment. Figure 4 As shown, firstly, the original beat frequency signal to be compressed is processed using a distributed acoustic wave sensing data compression method, and then the data bit width is further reduced through a quantization module. The quantized data is then processed sequentially through frequency offset compensation and a low-pass filter to extract the baseband signal. After grouping and data reconstruction, the rotating vector summation (RVS) algorithm is used to eliminate the effects of interference fading. The optimized signal is then filtered by moving average, followed by phase extraction and unwrapping. By calculating the phase difference, the core physical quantities reflecting the vibration characteristics are reconstructed, and the position and waveform of the vibration point are extracted. (Refer to...) Figure 5 , Figure 5 This is a schematic diagram illustrating the location of the vibration point in one embodiment. For example... Figure 5 As shown in the vibration point location map, the location of the vibration point can be clearly identified. (Refer to...) Figure 6 , Figure 6 This is a time-domain plot of the demodulated vibration point signal in one embodiment. For example... Figure 6 As shown, after finding the vibration point, the time-domain waveform of the vibration point is recovered. It can be seen that the sine waveform of the vibration point can also be accurately recovered.

[0039] In this embodiment, efficient compression of distributed acoustic wave sensing data is achieved by inputting the original beat frequency signal into a pre-trained encoder module and directly outputting a low-dimensional compressed feature vector. The encoder module, through end-to-end training, learns how to map the high-dimensional original signal into a low-dimensional feature vector, preserving the key physical characteristics of the signal in the process. Compared to traditional fixed downsampling or threshold compression, this embodiment can adapt to the dynamic characteristics of the signal, effectively maintaining signal fidelity while compressing the data volume, thereby reducing the burden of subsequent storage, transmission, and computational processing.

[0040] Secondly, embodiments of this application also provide a distributed acoustic wave sensing data compression device.

[0041] In one embodiment, reference is made to Figure 7 , Figure 7 This is a functional module diagram of an embodiment of the distributed acoustic wave sensor data compression device of this application. Figure 7 As shown, the distributed acoustic wave sensing data compression device includes: Acquisition module 10 is used to acquire the original beat frequency signal to be compressed; Compression module 20 is used to input the original beat frequency signal into a pre-trained encoder module to obtain a low-dimensional compressed feature vector output by the encoder module for subsequent processing based on the low-dimensional compressed feature vector. The encoder module is trained in the following way: a neural network model containing an encoder module and a decoder module is trained with the goal of minimizing the composite loss function obtained from the input of the encoder module and the output of the decoder module, thereby obtaining a pre-trained encoder module.

[0042] Further, in one embodiment, the neural network model includes an encoder module and a decoder module. The encoder module includes a first convolutional layer, a first residual block, a pooling layer, a second convolutional layer, and a second residual block connected in sequence. The decoder module includes a third convolutional layer, a third residual block, an upsampling layer, a fourth convolutional layer, and a fourth residual block connected in sequence. The training process of the encoder module includes: The beat frequency signal used for training is obtained as a training sample. The training sample is input into the encoder module and processed by the first convolutional layer, the first residual block, the pooling layer, the second convolutional layer and the second residual block to output a compressed feature vector. The compressed feature vector is input into the decoder module, and after processing by the third convolutional layer, the third residual block, the upsampling layer, the fourth convolutional layer, and the fourth residual block, the reconstructed signal is output. Calculate the composite loss function between the training samples and the reconstructed signal, wherein the composite loss function includes at least two of the following: time-domain loss function, frequency-domain loss function, phase loss function, and gradient loss function; The trainable parameters of the encoder module and the decoder module are updated simultaneously using the backpropagation algorithm to minimize the composite loss function. Repeat the above steps iteratively until the convergence condition is met to obtain the trained encoder module.

[0043] Further, in one embodiment, each residual block includes a fifth convolutional layer, a first activation function layer, a Dropout layer, a sixth convolutional layer, an addition module, and a second activation function layer connected in sequence, wherein: The fifth convolutional layer performs a convolution operation on the input of the residual block; The first activation function layer applies a non-linear activation to the output of the fifth convolutional layer; The Dropout layer randomly sets some neurons in the output of the first activation function layer to zero with a preset probability, while keeping the rest unchanged, in order to prevent overfitting. The sixth convolutional layer performs a convolution operation on the output of the Dropout layer; The addition module adds the input of the residual block to the output of the sixth convolutional layer element by element; The second activation function layer applies a nonlinear activation to the output of the addition module, which serves as the final output of the residual block.

[0044] Furthermore, in one embodiment, the first / second activation function layer employs a parameterized modified linear unit, which has a learnable slope parameter on the negative half-axis.

[0045] Furthermore, in one embodiment, the distributed acoustic wave sensing data compression device further includes a quantization module, used for: The low-dimensional compressed feature vector is quantized to reduce the data bit width.

[0046] Furthermore, in one embodiment, the quantization process employs a μ-law quantization algorithm.

[0047] Furthermore, in one embodiment, the subsequent processing includes: The low-dimensional compressed feature vector is sequentially subjected to frequency offset compensation and low-pass filtering to extract the baseband signal; After grouping and reconstructing the baseband signal, a rotating vector summation algorithm is used to eliminate the effects of interference fading. Phase extraction and unwrapping are performed on the signal processed by the rotating vector summation algorithm, and vibration information is restored by calculating the phase difference.

[0048] The functions of each module in the above-mentioned distributed acoustic wave sensing data compression device correspond to the steps in the above-mentioned distributed acoustic wave sensing data compression method embodiment, and their functions and implementation processes will not be described in detail here.

[0049] Thirdly, embodiments of this application provide a distributed acoustic wave sensing data compression device, which can be a device with data processing capabilities such as a personal computer (PC), a laptop computer, or a server.

[0050] Reference Figure 8 , Figure 8 This is a schematic diagram of the hardware structure of the distributed acoustic wave sensor data compression device involved in the embodiments of this application. In the embodiments of this application, the distributed acoustic wave sensor data compression device may include a processor, a memory, a communication interface, and a communication bus.

[0051] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0052] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces used for interconnecting devices within the distributed acoustic wave sensor data compression device, as well as interfaces used for interconnecting the distributed acoustic wave sensor data compression device with other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.

[0053] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0054] The processor can be a general-purpose processor, which can call the distributed acoustic wave sensor data compression program stored in the memory and execute the distributed acoustic wave sensor data compression method provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the distributed acoustic wave sensor data compression program is called can be referred to in the various embodiments of the distributed acoustic wave sensor data compression method of this application, and will not be repeated here.

[0055] Those skilled in the art will understand that Figure 8 The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0056] Fourthly, embodiments of this application also provide a computer-readable storage medium.

[0057] The present application provides a computer-readable storage medium storing a distributed acoustic wave sensor data compression program, wherein when the distributed acoustic wave sensor data compression program is executed by a processor, it implements the steps of the distributed acoustic wave sensor data compression method described above.

[0058] The method implemented when the distributed acoustic wave sensor data compression program is executed can be referred to in various embodiments of the distributed acoustic wave sensor data compression method of this application, and will not be repeated here.

[0059] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0060] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0061] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0062] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0063] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0064] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0065] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method of distributed acoustic wave sensor data compression, characterized by, The distributed acoustic wave sensing data compression method includes: Obtain the raw beat frequency signal to be compressed; The original beat frequency signal is input into a pre-trained encoder module to obtain a low-dimensional compressed feature vector output by the encoder module, which is then used for subsequent processing. The encoder module is trained in the following way: a neural network model containing an encoder module and a decoder module is trained with the goal of minimizing the composite loss function obtained from the input of the encoder module and the output of the decoder module, thereby obtaining a pre-trained encoder module.

2. The distributed acoustic sensing data compression method of claim 1, wherein, The neural network model includes an encoder module and a decoder module. The encoder module includes a first convolutional layer, a first residual block, a pooling layer, a second convolutional layer, and a second residual block connected in sequence. The decoder module includes a third convolutional layer, a third residual block, an upsampling layer, a fourth convolutional layer, and a fourth residual block connected in sequence. The training process of the encoder module includes: The beat frequency signal used for training is obtained as a training sample. The training sample is input into the encoder module and processed by the first convolutional layer, the first residual block, the pooling layer, the second convolutional layer and the second residual block to output a compressed feature vector. The compressed feature vector is input into the decoder module, and after processing by the third convolutional layer, the third residual block, the upsampling layer, the fourth convolutional layer, and the fourth residual block, the reconstructed signal is output. Calculate the composite loss function between the training samples and the reconstructed signal, wherein the composite loss function includes at least two of the following: time-domain loss function, frequency-domain loss function, phase loss function, and gradient loss function; The trainable parameters of the encoder module and the decoder module are updated simultaneously using the backpropagation algorithm to minimize the composite loss function. Repeat the above steps iteratively until the convergence condition is met to obtain the trained encoder module.

3. The distributed acoustic sensing data compression method of claim 2, wherein, Each residual block comprises a fifth convolutional layer, a first activation function layer, a Dropout layer, a sixth convolutional layer, an addition module, and a second activation function layer, connected in sequence, wherein: The fifth convolutional layer performs a convolution operation on the input of the residual block; The first activation function layer applies a non-linear activation to the output of the fifth convolutional layer; The Dropout layer randomly sets some neurons in the output of the first activation function layer to zero with a preset probability, while keeping the rest unchanged, in order to prevent overfitting. The sixth convolutional layer performs a convolution operation on the output of the Dropout layer; The addition module adds the input of the residual block to the output of the sixth convolutional layer element by element; The second activation function layer applies a nonlinear activation to the output of the addition module, which serves as the final output of the residual block.

4. The distributed acoustic sensing data compression method of claim 1, wherein, The first / second activation function layer employs a parameterized modified linear unit, which has a learnable slope parameter on the negative half-axis.

5. The distributed acoustic wave sensing data compression method as described in claim 1, characterized in that, After obtaining the low-dimensional compressed feature vector output by the encoder module, the method further includes: The low-dimensional compressed feature vector is quantized to reduce the data bit width.

6. The distributed acoustic wave sensing data compression method as described in claim 5, characterized in that, The quantization process employs the μ-law quantization algorithm.

7. The distributed acoustic wave sensing data compression method according to any one of claims 1 to 6, characterized in that, The subsequent processing includes: The low-dimensional compressed feature vector is sequentially subjected to frequency offset compensation and low-pass filtering to extract the baseband signal; After grouping and reconstructing the baseband signal, a rotating vector summation algorithm is used to eliminate the effects of interference fading. Phase extraction and unwrapping are performed on the signal processed by the rotating vector summation algorithm, and vibration information is restored by calculating the phase difference.

8. A distributed acoustic wave sensing data compression device, characterized in that, The distributed acoustic wave sensing data compression device includes: The acquisition module is used to acquire the raw beat frequency signal to be compressed; The compression module is used to input the original beat frequency signal into the pre-trained encoder module and obtain the low-dimensional compressed feature vector output by the encoder module for subsequent processing based on the low-dimensional compressed feature vector. The encoder module is trained in the following way: a neural network model containing an encoder module and a decoder module is trained with the goal of minimizing the composite loss function obtained from the input of the encoder module and the output of the decoder module, thereby obtaining a pre-trained encoder module.

9. A distributed acoustic wave sensing data compression device, characterized in that, The distributed acoustic wave sensor data compression device includes a processor, a memory, and a distributed acoustic wave sensor data compression program stored in the memory and executable by the processor, wherein when the distributed acoustic wave sensor data compression program is executed by the processor, it implements the steps of the distributed acoustic wave sensor data compression method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a distributed acoustic wave sensor data compression program, wherein when the distributed acoustic wave sensor data compression program is executed by a processor, it implements the steps of the distributed acoustic wave sensor data compression method as described in any one of claims 1 to 7.

Images