Optical frequency domain reflectometry distributed strain sensing signal denoising method, apparatus and medium

Abstract

The present application relates to the distributed strain sensing signal denoising method, device and medium of optical frequency domain reflectometry, the method comprises S1, analog optical frequency domain reflectometer strain sensing signal and corresponding noise data, add strain sensing signal and corresponding noise data, the generated analog noise sensing data and strain sensing signal are used as training set;S2, train the denoising network based on training set, obtain the denoising model of training completion;S3, obtain actual optical frequency domain reflectometer sensing signal, input actual optical frequency domain reflectometer sensing signal into denoising model, obtain the optical frequency domain reflectometer sensing signal after denoising.Compared with the prior art, the present application has the advantages of improving denoising performance and the like.

Description

Technical Field

[0001] This invention relates to the field of fiber optic sensor signal noise reduction technology, and in particular to a method, apparatus and medium for noise reduction of distributed strain sensor signals from an optical frequency domain reflectometer. Background Technology

[0002] Distributed strain sensing (DSS) technology based on optical frequency domain reflectometers (OFDR) offers advantages such as fully distributed measurement, high spatial resolution, and high strain resolution. It is widely used in distributed strain monitoring, such as monitoring the health status of large building structures and monitoring the deformation of battery packs. However, in practical applications, due to the influence of noise such as the nonlinearity of the light source sweep frequency, its strain resolution and other performance characteristics significantly decrease with increasing sensing distance.

[0003] Many researchers have conducted in-depth studies on noise reduction techniques to improve the performance of OFDR-based DSS systems. The main noise source in this system is the sweep frequency nonlinearity of the light source, and the impact of this nonlinearity increases with the measurement distance. For a certain location on the optical fiber, the noise distribution of multiple measurements follows a Gaussian distribution, but its standard deviation increases with the measurement distance, even to the point of being undemodulation-capable. To accurately measure strain, it is necessary to effectively separate the signal from the noise. Traditional noise reduction methods mainly include those based on various image processing algorithms and those based on discrete wavelet transform. However, image processing algorithm-based noise reduction methods, because their principle is based on the consistency of Rayleigh signal characteristics at adjacent locations, will lose some feature information of the measured location. Discrete wavelet transform-based noise reduction methods require manual selection of appropriate wavelet bases and signal decomposition levels based on the signal characteristics of specific application scenarios, thus lacking universal applicability to different scenarios. At the same time, there is still room for further improvement in the noise reduction capabilities of existing noise reduction methods. Therefore, existing noise reduction methods all have certain limitations and shortcomings in terms of the noise reduction performance of OFDR-based DSS systems.

[0004] In the field of signal denoising, the development of artificial intelligence algorithms has provided new possibilities for solving noise problems in RDTS systems. Regarding the application of AI algorithms in signal denoising, researchers have proposed a denoising scheme based on a one-dimensional convolutional neural network (1DCNN). This scheme simulates a distributed optical fiber sensing scenario, using randomly generated sensing signals with certain rules and Gaussian white noise with the same intensity as the actual noise. Through deep learning, a deep learning denoiser suitable for sensor signal denoising is trained. This method has been proven effective in several studies; however, this scheme uses a stacked small convolutional kernel design, resulting in a relatively small receptive field. It can only process a small range of sensing information at a time, leading to room for improvement in signal denoising performance.

[0005] In summary, existing technologies using neural networks to denoise distributed strain sensing signals from optical frequency domain reflectometers suffer from low efficiency due to the small receptive field. Summary of the Invention

[0006] The purpose of this invention is to provide a method, device, and medium for denoising distributed strain sensor signals of an optical frequency domain reflectometer in order to improve the denoising efficiency of sensor signals.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A method for denoising distributed strain sensor signals from an optical frequency domain reflectometer, the method comprising:

[0009] S1. Simulate the strain sensing signal and corresponding noise data of the optical frequency domain reflectometer, add the strain sensing signal and corresponding noise data, and use the generated simulated noisy sensing data and strain sensing signal as the training set.

[0010] S2. Train the denoising network based on the training set to obtain the trained denoising model;

[0011] S3. Obtain the actual optical frequency domain reflector sensor signal, input the actual optical frequency domain reflector sensor signal into the noise reduction model, and obtain the noise-reduced optical frequency domain reflector sensor signal.

[0012] The specific steps for training the denoising network based on the training set are as follows:

[0013] The noisy sensor data in the training set passes sequentially through the input layer, the initial module, multiple large convolutional processing modules, and the output layer in the denoising network. A transition layer is set between every two large convolutional processing modules. The loss function is calculated by combining the output of the output layer and the strain sensor signal used as a label. The denoising network is then optimized and trained based on the loss function.

[0014] Furthermore, assuming there are n large convolutional processing modules, then there are n-1 transition layers, where n is a positive integer greater than 1. The specific steps for the noisy sensor data in the training set to pass through the input layer, the initial module, multiple large convolutional processing modules, and the output layer in the denoising network are as follows:

[0015] The noisy sensor data in the training set enters the input layer. The output of the input layer enters the start module. The output of the start module enters the first large convolutional processing module. The output of the first large convolutional processing module enters the first transition layer. The output of the first transition layer enters the next large convolutional processing module, and so on, until the output of the (n-1)th large convolutional processing module enters the (n-1)th transition layer. The output of the (n-1)th transition layer enters the nth large convolutional processing module, and the output of the nth large convolutional processing module enters the output layer.

[0016] Furthermore, the starting module includes a first convolution, a first one-dimensional depthwise separable convolution, a second convolution, and a second one-dimensional depthwise separable convolution. After the output of the input layer enters the starting module, it sequentially passes through the first convolution, the first one-dimensional depthwise separable convolution, the second convolution, and the second one-dimensional depthwise separable convolution. The output of the starting module then enters the first large convolution processing module.

[0017] Furthermore, the large convolution processing module includes multiple large convolutional layers, each including a large convolutional kernel layer and a convolutional feedforward layer. After the output of the initial module enters the large convolution processing module, it passes through multiple large convolutional layers in sequence. In each large convolutional layer, it passes through a large convolutional kernel layer and a convolutional feedforward layer in sequence.

[0018] Furthermore, the large convolutional kernel layer includes a first normalization layer, a third convolution, a third one-dimensional depthwise separable convolution, and a fourth convolution arranged sequentially;

[0019] The convolutional feedforward layer consists of a second normalized layer, a fifth convolution, and an activation function arranged sequentially. The activation function is a Gaussian error linear unit.

[0020] The specific steps of passing through the large convolutional kernel layer and the convolutional feedforward layer are as follows: sequentially passing through the first normalization layer, the third convolution, the third one-dimensional depthwise separable convolution, and the fourth convolution, and then sequentially passing through the second normalization layer, the fifth convolution, and the activation function.

[0021] Furthermore, the transition layer includes a sixth convolution and a fourth one-dimensional depthwise separable convolution.

[0022] The specific steps for the output of the large convolutional processing module to enter the transition layer are as follows:

[0023] The output of the large convolutional processing module passes through the sixth convolution and the fourth one-dimensional depthwise separable convolution in sequence, and then the output of the transition layer enters the next large convolutional processing module.

[0024] Furthermore, the output layer includes a seventh convolution, a fifth one-dimensional depthwise separable convolution, and a batch normalization layer;

[0025] After the output of the nth large convolutional processing module enters the output layer, it passes through the seventh convolution, the fifth one-dimensional depthwise separable convolution, and the batch normalization layer in sequence.

[0026] Furthermore, the noise data is the product of a random positive integer and a noise base, and the intensity of the noise data is the same as the noise intensity in the actual optical frequency domain reflectometer sensing signal.

[0027] Furthermore, the noise floor is:

[0028]

[0029] Where ΔF is the frequency sweep, N is the number of cross-correlation points, and pm / με is the sensitivity coefficient of the fiber optic sensor.

[0030] The present invention also proposes a noise reduction device for distributed strain sensing signals of an optical frequency domain reflectometer, comprising a memory, a processor, and a program stored in the memory, wherein the processor executes the program to implement the above-described method.

[0031] This invention also proposes a noise reduction device for distributed strain sensor signals from an optical frequency domain reflectometer, comprising a training set generation module, a noise reduction model training module, and an actual noise reduction module, wherein...

[0032] The training set generation module is used to simulate the strain sensing signal and corresponding noise data of the optical frequency domain reflectometer. The strain sensing signal and the corresponding noise data are added together, and the generated simulated noisy sensing data and strain sensing signal are used as the training set.

[0033] The noise reduction model training module is used to train the noise reduction network based on the training set to obtain the trained noise reduction model;

[0034] The actual noise reduction module is used to acquire the actual optical frequency domain reflectometer sensing signal, input the actual optical frequency domain reflectometer sensing signal into the noise reduction model, and obtain the noise-reduced optical frequency domain reflectometer sensing signal.

[0035] The specific steps for training the denoising network based on the training set are as follows:

[0036] The noisy sensor data in the training set passes sequentially through the input layer, the initial module, multiple large convolutional processing modules, and the output layer in the denoising network. A transition layer is set between every two large convolutional processing modules. The loss function is calculated by combining the output of the output layer and the strain sensor signal used as a label. The denoising network is then optimized and trained based on the loss function.

[0037] The present invention also proposes a computer-readable storage medium having a program stored thereon, which, when executed, implements the above-described method.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] (1) The noise reduction model of the present invention combines a large convolution processing module, and uses shallow large convolution to replace deep small convolution, which effectively improves the receptive field of the network, thereby improving the efficiency of noise reduction and making the network have better noise reduction performance.

[0040] (2) In this invention, the training set for training the noise reduction model is obtained through simulated data. The simulated noise data is the product of a randomized integer and the noise base, so that the intensity of the simulated noise is consistent with the intensity of the actual noise. Using such simulated actual noise data to train the model enhances the model's ability to reduce noise in actual noise and allows the model to adapt to various different actual noise conditions, thereby improving noise reduction performance and applicability. Attached Figure Description

[0041] Figure 1 This is a flowchart of the present invention;

[0042] Figure 2 This is a structural diagram of the noise reduction model of the present invention;

[0043] Figure 3 This is a flowchart of the noise reduction network training process of the present invention;

[0044] Figure 4 This is a structural diagram of the distributed strain sensor signal noise reduction device for the optical frequency domain reflectometer of the present invention;

[0045] Figure 5 This is a structural diagram of another optical frequency domain reflectometer distributed strain sensor signal noise reduction device according to the present invention. Detailed Implementation

[0046] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0047] Example 1

[0048] When analyzing strain using distributed strain sensing (DSS) based on optical frequency domain reflectometer (OFDR), the accuracy of the analysis decreases significantly with increasing sensing distance.

[0049] Generally, distributed strain sensing (DSS) analysis using an optical frequency domain reflectometer (OFDR) includes the following steps:

[0050] The OFDR signal acquired before strain occurs is used as a reference signal, and its expression is:

[0051] ∑ i A i cos[2πγτ i t+C i +θ(t)-θ(t-τ i )]

[0052] Here A i and τ iRepresenting the intensity and group delay of Rayleigh backscattered light at different locations, respectively, γ is the sweep rate, and C i The phase constant is related to the scattering position, and θ(t) is the phase noise caused by the finite linewidth of the light source and the nonlinearity of the sweep frequency mechanism, which is the main noise source of the OFDR-based DSS system.

[0053] The OFDR signal after strain occurs is acquired as the measurement signal. The OFDR signals of the reference state and the measurement state are transformed into the range domain using Fourier transform. The OFDR range domain data is segmented using a window of a certain length; therefore, the spatial resolution of the strain sensing system depends on the length of the window and the spatial resolution of the two OFDR points. The segmented local OFDR range domain signal is then transformed into the spectral domain using inverse Fourier transform (the spectral quantization resolution in the spectral domain can be improved by methods such as zero-padding in the range domain before the inverse Fourier transform). Since strain changes the phase of the Rayleigh backscattered light, the spectral shift caused by strain can be obtained using a cross-correlation algorithm, calculated as follows:

[0054]

[0055] here For cross-correlation calculations, the spectral shift is then converted into strain values ​​through a correspondence, ultimately yielding a curve showing strain variation with distance.

[0056] The measurement signals obtained in the above process will be affected by noise, which will result in a large error in the final strain-distance curve.

[0057] To overcome the above problems and improve the accuracy and performance of sensor signal denoising, this invention proposes a method for denoising distributed strain sensor signals from an optical frequency domain reflectometer. The flowchart of the method is as follows: Figure 1 As shown. The method includes:

[0058] S1. Simulate the strain sensing signal and corresponding noise data of the optical frequency domain reflectometer, add the strain sensing signal and corresponding noise data, and use the generated simulated noisy sensing data and strain sensing signal as the training set.

[0059] S2. Train the denoising network based on the training set to obtain the trained denoising model;

[0060] S3. Obtain the actual optical frequency domain reflector sensor signal, input the actual optical frequency domain reflector sensor signal into the noise reduction model, and obtain the noise-reduced optical frequency domain reflector sensor signal.

[0061] The specific steps for training the denoising network based on the training set are as follows:

[0062] The noisy sensor data in the training set sequentially passes through the input layer, initial module, multiple large convolutional processing modules, and output layer of the denoising network. A transition layer is placed between every two large convolutional processing modules. A loss function is calculated by combining the output of the output layer and the strain sensor signal used as a label. The denoising network is then optimized and trained based on this loss function. The flowchart for training the denoising network is as follows: Figure 2 As shown.

[0063] In S1, diverse OFDR signal data is generated by simulating strain sensing data and noise data from different scenarios, with a synthesis quantity of up to 50,000.

[0064] The sensing data for different strain scenarios are constructed using random numbers with an intensity range of Q, where Q is set with reference to the intensity range of the actual sensing application. The intensity consistency length is a random number between 1 and 250, which is used to simulate strain sensing activities that can occur at any location on the optical fiber.

[0065] Based on the characteristics of OFDR noise data, noise data with matching intensity is generated. The parameters of the OFDR hardware system determine the characteristics of the noise in the signal. The noise intensity is a random integer multiple of the base α, where α is expressed as follows, ΔF is the frequency sweep, and N is the number of cross-correlation points.

[0066]

[0067] The simulated noise signal is a product of a randomized integer and the noise base, and the intensity of the simulated noise is required to be consistent with the intensity of the actual noise.

[0068] The noisy data and strain sensing data are added together to generate simulated noisy sensing data. This noisy sensing data is normalized to a range of plus or minus one and used as the training data for the model, while the strain sensing data is used as the label data for model training. The training dataset can be divided into a training dataset and a test dataset in an 8:2 ratio.

[0069] The S2 noise reduction network consists of an input layer, a start module, a large convolutional layer processing module, a transition module, and an output layer.

[0070] In S2, if there are n large convolutional processing modules, then there are n-1 transition layers, where n is a positive integer greater than 1. The specific steps for the noisy sensor data in the training set to pass through the input layer, the initial module, multiple large convolutional processing modules, and the output layer in the denoising network are as follows:

[0071] The noisy sensor data in the training set enters the input layer. The output of the input layer enters the start module. The output of the start module enters the first large convolutional processing module. The output of the first large convolutional processing module enters the first transition layer. The output of the first transition layer enters the next large convolutional processing module, and so on, until the output of the (n-1)th large convolutional processing module enters the (n-1)th transition layer. The output of the (n-1)th transition layer enters the nth large convolutional processing module, and the output of the nth large convolutional processing module enters the output layer.

[0072] The flowchart for training a noise reduction network is as follows: Figure 3 As shown.

[0073] The input layer is used to receive one-dimensional sensing signals as input to the network.

[0074] The start module (Stem) captures detailed information about the signal by starting with several convolutional layers.

[0075] The starting module includes a first convolution, a first one-dimensional depthwise separable convolution, a second convolution, and a second one-dimensional depthwise separable convolution. After the output of the input layer enters the starting module, it passes through the first convolution, the first one-dimensional depthwise separable convolution, the second convolution, and the second one-dimensional depthwise separable convolution in sequence. The output of the starting module enters the first large convolution processing module.

[0076] The first convolution is a one-dimensional convolution with a kernel of 3. The first and second one-dimensional depthwise separable convolutions have kernels of 3. The second convolution is a one-dimensional 1×1 convolution.

[0077] The large convolutional layer processing module is used to provide signal feature extraction with a large receptive field. The large convolutional processing module includes multiple large convolutional layers, including large convolutional kernel layers (LK_BLOCK) and convolutional feedforward layers (ConvFFN). After the output of the initial module enters the large convolutional processing module, it passes through multiple large convolutional layers in sequence. In each large convolutional layer, it passes through a large convolutional kernel layer and a convolutional feedforward layer in sequence.

[0078] In the large convolutional layer processing module, large convolutional kernel layers and convolutional feedforward layers are stacked alternately, and the depth of the module can be freely designed.

[0079] The large convolutional kernel layer consists of a first normalized layer (BN), a third convolution, a third one-dimensional depthwise separable convolution, and a fourth convolution, arranged sequentially. The third and fourth convolutions are 1×1 convolutions. The large convolutional kernel layer incorporates a shortcut design.

[0080] The convolutional feedforward layer consists of a second normalized layer (BN), a fifth convolution, and an activation function, which is a Gaussian error linear unit (GELU). The fifth convolution is a 1×1 convolution, and the GELU is as follows:

[0081]

[0082] The specific steps of passing through the large convolutional kernel layer and the convolutional feedforward layer are as follows: sequentially passing through the first normalization layer, the third convolution, the third one-dimensional depthwise separable convolution, and the fourth convolution, and then sequentially passing through the second normalization layer, the fifth convolution, and the activation function.

[0083] The transition layer consists of a sixth convolution and a fourth one-dimensional depthwise separable convolution. The sixth convolution is a 1×1 convolution.

[0084] The specific steps for the output of the large convolutional processing module to enter the transition layer are as follows:

[0085] The output of the large convolutional processing module passes through the sixth convolution and the fourth one-dimensional depthwise separable convolution in sequence, and then the output of the transition layer enters the next large convolutional processing module.

[0086] Large convolutional processing modules and transition layers are alternated, with a transition layer between every two large convolutional processing modules. The specific network depth can be freely designed according to the complexity of the data.

[0087] The output layer is used to optimize the processing of normalized data. The output layer includes a seventh convolution, a fifth one-dimensional depthwise separable convolution, and a batch normalization layer.

[0088] After the output of the nth large convolutional processing module enters the output layer, it sequentially passes through the seventh convolution, the fifth one-dimensional depthwise separable convolution, and the batch normalization layer (BN). The seventh convolution has a kernel of 3.

[0089] The loss function is calculated based on the output and labels of the output layer. The loss function for model training is:

[0090]

[0091] The loss function values ​​of the model output signal and the actual signal are calculated. The parameters in the algorithm are updated iteratively using the Adaptive Moment Estimation (Adam) optimizer to obtain the signal generation and denoising model with the optimal loss value. The actual OFDR strain sensing signal is input into the denoising model, and the denoised OFDR strain sensing signal is output.

[0092] This invention employs a novel noise reduction scheme. By simulating sensor data and noise, a signal denoising device is trained, allowing for efficient noise reduction of real signals using the denoising model. The training set for the denoising model is obtained through simulated data. The simulated noise data is the product of a randomized integer and the noise base, ensuring that the intensity of the simulated noise matches that of the actual noise. Using such simulated real-world noise data for model training enhances the model's noise reduction capability and allows it to adapt to various real-world noise conditions, improving denoising performance and applicability.

[0093] The denoising model of this invention incorporates a large convolution processing module, replacing deep small convolutions with shallow large convolutions, effectively increasing the network's receptive field and thus improving denoising efficiency and performance. In sub-millimeter OFDR system testing, the standard deviation of strain after denoising was as low as 0.78 micro-strains.

[0094] This invention can achieve highly efficient OFDR signal denoising, which traditional image-based methods struggle to achieve with such a short processing time.

[0095] The actual experiment will be conducted below:

[0096] Submillimeter-level spatial resolution OFDR sensing signals were acquired experimentally. Micro-strains of 0, 1000, and 2000 were applied to the tail end of a 140-meter-long fiber optic cable. Noise was set to white noise with a randomized integer deviation of 3.3 standard deviations. The initial module had 64 channels (C1), and two large convolutional processing modules (C2) with 128 channels were used. The kernel size of the first large convolutional module was 51, and the kernel size of the second was 47. Each large convolutional module contained two large convolutional kernel layers (LK_BLOCK) and a convolutional feedforward layer (ConvFFN). The batch size for model training was 64, the Adam optimizer learning rate was 0.01, the weight decay was set to 5e-3, and the number of training epochs was 300. This method can effectively reduce the noise of sensor signals and obtain sensor signals with a high signal-to-noise ratio. In the best case, the standard deviation of the signal is about 0.78 micro-strains.

[0097] Example 2

[0098] Based on the denoising method for distributed strain sensing signals from an optical frequency domain reflectometer in Embodiment 1, this invention proposes a denoising device for distributed strain sensing signals from a frequency domain reflectometer. The device includes a training set generation module, a denoising model training module, and an actual denoising module.

[0099] The training set generation module is used to simulate the strain sensing signal and corresponding noise data of the optical frequency domain reflectometer. The strain sensing signal and the corresponding noise data are added together, and the generated simulated noisy sensing data and strain sensing signal are used as the training set.

[0100] The noise reduction model training module is used to train the noise reduction network based on the training set to obtain the trained noise reduction model;

[0101] The actual noise reduction module is used to acquire the actual optical frequency domain reflectometer sensing signal, input the actual optical frequency domain reflectometer sensing signal into the noise reduction model, and obtain the noise-reduced optical frequency domain reflectometer sensing signal.

[0102] The specific steps for training the denoising network based on the training set are as follows:

[0103] The noisy sensor data in the training set passes sequentially through the input layer, the initial module, multiple large convolutional processing modules, and the output layer in the denoising network. A transition layer is set between every two large convolutional processing modules. The loss function is calculated by combining the output of the output layer and the strain sensor signal used as a label. The denoising network is then optimized and trained based on the loss function.

[0104] The structure of the frequency domain reflectometer distributed strain sensor signal noise reduction device is as follows: Figure 4 As shown.

[0105] This invention also proposes another noise reduction device for distributed strain sensing signals of an optical frequency domain reflectometer, including a memory, a processor, and a program stored in the memory. When the processor executes the program, it implements the method described in Embodiment 1. The structure of the other noise reduction device for distributed strain sensing signals of an optical frequency domain reflectometer is as follows: Figure 4 As shown. Another type of optical frequency domain reflectometer distributed strain sensor signal noise reduction device structure is as follows. Figure 5 As shown.

[0106] The device includes a processor, internal bus, network interface, memory, and non-volatile storage, and may also include other hardware required for the operation. The processor reads the corresponding computer program from the non-volatile storage into memory and then executes it to achieve the above. Figure 1 The method for noise reduction of distributed strain sensing signals from an optical frequency domain reflectometer is described above. Of course, besides software implementation, this invention does not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution entity of the following processing flow is not limited to individual logic units, but can also be hardware or logic devices.

[0107] Improvements in a technology can be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many improvements to the methodology can now be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that an improvement in methodology cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog are the most commonly used. Those skilled in the art should understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0108] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0109] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0110] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, in implementing this invention, the functions of each unit can be implemented in one or more software and / or hardware components.

[0111] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0112] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0113] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0114] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0115] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0116] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0117] Example 3

[0118] The present invention also proposes a computer-readable storage medium having a program stored thereon, which, when executed, implements the method in Embodiment 1.

[0119] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0120] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0121] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0122] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0123] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0124] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for denoising distributed strain sensing signals from an optical frequency domain reflectometer, characterized in that, The methods include: S1. Simulate the strain sensing signal and corresponding noise data of the optical frequency domain reflectometer, add the strain sensing signal and corresponding noise data, and use the generated simulated noisy sensing data and strain sensing signal as the training set. S2. Train the denoising network based on the training set to obtain the trained denoising model; S3. Obtain the actual optical frequency domain reflector sensor signal, input the actual optical frequency domain reflector sensor signal into the noise reduction model, and obtain the noise-reduced optical frequency domain reflector sensor signal. The specific steps for training the denoising network based on the training set are as follows: The noisy sensor data in the training set passes sequentially through the input layer, the initial module, multiple large convolutional processing modules, and the output layer in the denoising network. There is a transition layer between every two large convolutional processing modules. The loss function is calculated by combining the output of the output layer and the strain sensor signal used as a label. The denoising network is then optimized and trained based on the loss function. If there are n large convolutional processing modules, then there are n-1 transition layers, where n is a positive integer greater than 1. The specific steps for the noisy sensor data in the training set to pass through the input layer, the initial module, multiple large convolutional processing modules, and the output layer in the denoising network are as follows: The noisy sensor data in the training set enters the input layer. The output of the input layer enters the start module. The output of the start module enters the first large convolutional processing module. The output of the first large convolutional processing module enters the first transition layer. The output of the first transition layer enters the next large convolutional processing module, and so on, until the output of the (n-1)th large convolutional processing module enters the (n-1)th transition layer. The output of the (n-1)th transition layer enters the nth large convolutional processing module, and the output of the nth large convolutional processing module enters the output layer. The starting module includes a first convolution, a first one-dimensional depthwise separable convolution, a second convolution, and a second one-dimensional depthwise separable convolution. After the output of the input layer enters the starting module, it passes through the first convolution, the first one-dimensional depthwise separable convolution, the second convolution, and the second one-dimensional depthwise separable convolution in sequence. The output of the starting module enters the first large convolution processing module. The large convolution processing module includes multiple large convolutional layers, each containing a large convolutional kernel layer and a convolutional feedforward layer. After the output of the initial module enters the large convolution processing module, it passes through multiple large convolutional layers in sequence. In each large convolutional layer, it passes through a large convolutional kernel layer and a convolutional feedforward layer in sequence.

2. The method for noise reduction of distributed strain sensing signals from an optical frequency domain reflectometer according to claim 1, characterized in that, The large convolutional kernel layer consists of a first normalization layer, a third convolution, a third one-dimensional depthwise separable convolution, and a fourth convolution, arranged sequentially. The convolutional feedforward layer consists of a second normalized layer, a fifth convolution, and an activation function arranged sequentially. The activation function is a Gaussian error linear unit. The specific steps of passing through the large convolutional kernel layer and the convolutional feedforward layer are as follows: sequentially passing through the first normalization layer, the third convolution, the third one-dimensional depthwise separable convolution, and the fourth convolution, and then sequentially passing through the second normalization layer, the fifth convolution, and the activation function.

3. The method for noise reduction of distributed strain sensing signals from an optical frequency domain reflectometer according to claim 2, characterized in that, The transition layer includes a sixth convolution and a fourth one-dimensional depthwise separable convolution. The specific steps for the output of the large convolutional processing module to enter the transition layer are as follows: The output of the large convolutional processing module passes through the sixth convolution and the fourth one-dimensional depthwise separable convolution in sequence, and then the output of the transition layer enters the next large convolutional processing module.

4. The method for noise reduction of distributed strain sensing signals from an optical frequency domain reflectometer according to claim 3, characterized in that, The output layer includes a seventh convolution, a fifth one-dimensional depthwise separable convolution, and a batch normalization layer; After the output of the nth large convolutional processing module enters the output layer, it passes through the seventh convolution, the fifth one-dimensional depthwise separable convolution, and the batch normalization layer in sequence.

5. The method for noise reduction of distributed strain sensing signals from an optical frequency domain reflectometer according to claim 1, characterized in that, The noise data is the product of a random positive integer and the noise base, and the intensity of the noise data is the same as the noise intensity in the actual optical frequency domain reflectometer sensing signal.

6. The method for noise reduction of distributed strain sensing signals in an optical frequency domain reflectometer according to claim 5, characterized in that, The noise floor is: ； in, It's a frequency sweep. For the number of cross-correlation points, This represents the sensitivity coefficient of the fiber optic sensor.

7. A noise reduction device for distributed strain sensing signals of an optical frequency domain reflectometer, comprising a memory, a processor, and a program stored in the memory, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-6.

8. A noise reduction device for distributed strain sensor signals of an optical frequency domain reflectometer, characterized in that, It includes a training set generation module, a noise reduction model training module, and an actual noise reduction module, among which, The training set generation module is used to simulate the strain sensing signal and corresponding noise data of the optical frequency domain reflectometer. The strain sensing signal and the corresponding noise data are added together, and the generated simulated noisy sensing data and strain sensing signal are used as the training set. The noise reduction model training module is used to train the noise reduction network based on the training set to obtain the trained noise reduction model; The actual noise reduction module is used to acquire the actual optical frequency domain reflectometer sensing signal, input the actual optical frequency domain reflectometer sensing signal into the noise reduction model, and obtain the noise-reduced optical frequency domain reflectometer sensing signal. The specific steps for training the denoising network based on the training set are as follows: The noisy sensor data in the training set passes sequentially through the input layer, the initial module, multiple large convolutional processing modules, and the output layer in the denoising network. There is a transition layer between every two large convolutional processing modules. The loss function is calculated by combining the output of the output layer and the strain sensor signal used as a label. The denoising network is then optimized and trained based on the loss function. If there are n large convolutional processing modules, then there are n-1 transition layers, where n is a positive integer greater than 1. The specific steps for the noisy sensor data in the training set to pass through the input layer, the initial module, multiple large convolutional processing modules, and the output layer in the denoising network are as follows: The noisy sensor data in the training set enters the input layer. The output of the input layer enters the start module. The output of the start module enters the first large convolutional processing module. The output of the first large convolutional processing module enters the first transition layer. The output of the first transition layer enters the next large convolutional processing module, and so on, until the output of the (n-1)th large convolutional processing module enters the (n-1)th transition layer. The output of the (n-1)th transition layer enters the nth large convolutional processing module, and the output of the nth large convolutional processing module enters the output layer. The starting module includes a first convolution, a first one-dimensional depthwise separable convolution, a second convolution, and a second one-dimensional depthwise separable convolution. After the output of the input layer enters the starting module, it passes through the first convolution, the first one-dimensional depthwise separable convolution, the second convolution, and the second one-dimensional depthwise separable convolution in sequence. The output of the starting module enters the first large convolution processing module. The large convolution processing module includes multiple large convolutional layers, each containing a large convolutional kernel layer and a convolutional feedforward layer. After the output of the initial module enters the large convolution processing module, it passes through multiple large convolutional layers in sequence. In each large convolutional layer, it passes through a large convolutional kernel layer and a convolutional feedforward layer in sequence.

9. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed, it implements the method as described in any one of claims 1-6.

Images