OFDR large strain rapid measurement method based on machine learning
Through a machine learning-based method, parallel FFT and IFFT processing, combined with coarse demodulation and fine demodulation, the challenges of OFDR fiber sensing technology in large strain and rapid measurement are solved, and efficient strain demodulation is achieved.
Patent Information
- Application Number
- CN202510091301.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-21
- Publication Date
- 2025-05-30
AI Technical Summary
The existing OFDR fiber sensing technology has challenges in large strain and rapid measurement, making it difficult to achieve both large strain measurement range and high demodulation speed.
Using a machine learning-based method, by training the first and second machine learning models, coarse demodulation and fine demodulation are performed respectively, and combined with parallel FFT and IFFT processing, rapid measurement of large strains is achieved.
The amount of calculated data is greatly reduced, the strain demodulation speed is improved, and it is suitable for strain demodulation in large strain scenarios, and the strain accuracy is maintained.
Smart Images
Figure CN120063145A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of optical fiber sensing, and more specifically, relates to a method for rapid measurement of large strain in OFDR based on machine learning. Background Art
[0002] The optical fiber distributed sensing technology based on optical frequency domain reflectometry (OFDR) has superior performance in terms of spatial resolution, sensitivity, signal-to-noise ratio, etc., and is widely used in fields such as aerospace, petrochemical industry, civil engineering, and perimeter security. This technology uses a tunable laser with a narrow linewidth. The wavelength-scanned light emitted by the light source is divided into two paths and enters the reference arm and the measurement arm respectively. The local oscillator light in the reference arm and the backscattered Rayleigh signal returned from the fiber under test in the measurement arm generate a beat frequency. The position and environmental parameter information carried in the fiber is obtained by processing the frequency and phase information of the time-domain beat signal respectively. OFDR can be divided into phase demodulation and intensity demodulation according to the demodulation method. Among them, intensity demodulation has better anti-noise performance and higher demodulation accuracy. However, intensity demodulation faces challenges in large strain and rapid measurement.
[0003] Although some existing methods can achieve rapid demodulation, they can only achieve accurate measurement of small strains; some methods can demodulate large strains, but the demodulation speed is slow. Considering that in the actual optical fiber strain sensing scenario, it is necessary to simultaneously meet the large strain measurement range and high demodulation speed. Therefore, how to achieve rapid measurement of large strain is an urgent problem to be solved currently. Summary of the Invention
[0004] In view of the above defects or improvement requirements of the prior art, the present invention provides a method for rapid measurement of large strain in OFDR based on machine learning, aiming to solve the problem that it is difficult for existing demodulation methods to simultaneously achieve a large strain measurement range and a fast demodulation speed.
[0005] To achieve the above object, according to the first aspect of the present invention, there is provided a method for rapid measurement of large strain in OFDR based on machine learning, including:
[0006] Training stage:
[0007] Perform FFT processing on the time-domain reference signal before the preset strain acts on the distributed optical fiber sensor to obtain the frequency-domain reference signal, and perform windowing processing on it to decompose it into the frequency-domain reference signals of each sensing segment; after applying a phase shift corresponding to the preset strain to the frequency-domain reference signals of each sensing segment, perform IFFT processing to obtain the wavelength-domain signals of each sensing segment;
[0008] Using the wavelength-domain signals of the respective sensing segments as samples and the preset strain as labels, train the first machine learning model; using the local wavelength-domain signals near the reflection peaks of the wavelength-domain signals of the respective sensing segments, which account for a preset proportion of the complete wavelength-domain signal, as samples and the preset strain as labels, train the second machine learning model;
[0009] Application stage:
[0010] S1. Obtain the time-domain measurement signal after the strain to be measured acts on the distributed optical fiber sensor;
[0011] S2. Perform parallel FFT processing on the time-domain measurement signal to obtain the frequency-domain measurement signal, perform windowing processing on it to obtain the frequency-domain measurement signals of the respective sensing segments, perform serial-to-parallel conversion, zero-padding to make their lengths all N, and IFFT processing on the frequency-domain measurement signals of the respective sensing segments in sequence to obtain the wavelength-domain signals of the respective sensing segments;
[0012] S3. Input the wavelength-domain signals of the respective sensing segments into the trained first machine learning model, and the trained first machine learning model performs parallel processing on them to obtain the rough demodulated strain values of the respective sensing segments; determine the reflection peak positions of the respective wavelength-domain signals according to the respective rough strain values, intercept the local wavelength-domain signals near the reflection peaks, which account for a preset proportion of the complete wavelength-domain signal, and input them into the trained second machine learning model, and the trained second machine learning model performs parallel processing on them to obtain the fine demodulated strain values of the respective sensing segments; add the rough demodulated strain values and the fine demodulated strain values of the respective sensing segments to obtain the strain values to be measured of the respective sensing segments.
[0013] According to the second aspect of the present invention, there is provided an electronic device, including: a computer-readable storage medium and a processor;
[0014] The computer-readable storage medium is used to store executable instructions;
[0015] The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the method as described in the first aspect.
[0016] According to the third aspect of the present invention, there is provided a computer-readable storage medium, and the computer-readable storage medium stores computer instructions, and the computer instructions are used to cause the processor to execute the method as described in the first aspect.
[0017] According to the fourth aspect of the present invention, there is provided a computer program product, including a computer program or instructions, and when the computer program or instructions are executed by a processor, the method as described in the first aspect is implemented.
[0018] Generally speaking, compared with the prior art by the above technical solution conceived by the present invention, the following beneficial effects can be achieved:
[0019] The present invention provides a method for fast measurement of large strain based on machine learning for OFDR. This method uses an OFDR-based fiber optic sensing technology to obtain a time-domain signal sequence, and then uses a serial-to-parallel conversion device to perform parallel FFT calculations to obtain a frequency-domain signal. After serial-to-parallel conversion, the wavelength-domain signals of each sensing segment are obtained through IFFT, and the strain values of all sensing segments are demodulated in parallel to accelerate the demodulation speed. Among them, the demodulation includes: first, taking the complete spectrum as the input of the first machine learning model to calculate the roughly demodulated strain value with lower resolution, then intercepting the local spectrum according to the roughly demodulated strain value and taking this local spectrum as the input of the second machine learning model to calculate the finely demodulated strain value with higher resolution, and finally summing the above two strain values to obtain the final strain value for output. Generally speaking, this method greatly reduces the amount of data to be calculated, effectively improves the strain demodulation speed, and is applicable to strain demodulation in large strain scenarios without degrading the strain accuracy. Description of the Drawings
[0020] Figure 1 It is a flowchart of the method for fast measurement of large strain based on machine learning for OFDR provided by an embodiment of the present invention;
[0021] Figure 2 It is a schematic diagram of the principle of OFDR-based fiber optic sensing technology;
[0022] Figure 3 It is a schematic diagram of the principle of strain demodulation based on machine learning provided by an embodiment of the present invention;
[0023] Figure 4 It is a diagram of the OFDR fiber optic strain sensing device adopted by an embodiment of the present invention;
[0024] Figure 5 It is a flowchart of the method for fast measurement of large strain based on cascaded ANN and GPU for OFDR provided by an embodiment of the present invention;
[0025] Figure 6 It is a flowchart of the algorithm of cascaded ANN provided by an embodiment of the present invention;
[0026] Figure 7 It is a comparison diagram of strain errors between the demodulation method based on cascaded ANN and GPU and the traditional demodulation method based on cross-correlation and CPU provided by an embodiment of the present invention;
[0027] Figure 8 It is a comparison diagram of demodulation time consumption between the demodulation method based on cascaded ANN and GPU and the traditional demodulation method based on cross-correlation and CPU provided by an embodiment of the present invention. Detailed Embodiments
[0028] In order to make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0029] An embodiment of the present invention provides a method for fast measurement of large strain based on machine learning for OFDR, including:
[0030] Training stage:
[0031] Perform FFT processing on the time-domain reference signal before the preset strain acts on the distributed optical fiber sensor to obtain the frequency-domain reference signal, and perform windowing processing on it to decompose it into the frequency-domain reference signals of each sensing segment; after applying a phase shift corresponding to the preset strain to the frequency-domain reference signals of each sensing segment, perform IFFT processing to obtain the wavelength-domain signals of each sensing segment.
[0032] Use the wavelength-domain signals of each sensing segment as samples and the preset strain as labels to train the first machine learning model; use the local wavelength-domain signals (i.e., local spectra) that account for a preset proportion of the complete wavelength-domain signal (i.e., complete spectrum) near the reflection peak of the wavelength-domain signals of each sensing segment as samples and the preset strain as labels to train the second machine learning model.
[0033] Application stage, as Figure 1 shown, includes:
[0034] S1, obtain the time-domain measurement signal after the strain to be measured acts on the distributed optical fiber sensor.
[0035] Specifically, use the optical fiber sensing technology based on OFDR to respectively obtain the time-domain reference signal sequence I r (n) and the measurement signal sequence I m (n).
[0036] As Figure 2 shown, the principle of the optical fiber sensing technology based on OFDR is as follows: The continuously frequency-swept light output by the tunable laser 1 enters the coupler 2 and is divided into two beams of light, one of which enters the reference arm to become the reference light. The other beam enters the signal arm, passes through the circulator 3 and enters the optical fiber under test 4. The light entering the optical fiber under test 4 will undergo Rayleigh scattering to generate a backward scattering signal light. The backward scattering signal light passes through the circulator 3 and reaches the photosensitive surface of the photodetector 6 through different paths with the reference light through the coupler 5 to interfere. The interference light will carry the frequency difference and phase difference between the two beams of light and be detected on the surface of the photodetector to obtain the OFDR time-domain signal sequence including the time-domain reference signal and the time-domain measurement signal.
[0037] Among them, the sweep range of the continuous swept-frequency light determines the strain measurement range of the OFDR, and its formula is shown in Equation (1).
[0038]
[0039] Among them, |ε max | is the strain measurement range of the OFDR system, Δλ is the sweep range of the laser, λ is the central frequency of the laser, and K ε is the strain coefficient of the optical fiber.
[0040] S2. Perform parallel FFT processing on the time-domain measurement signal to obtain a frequency-domain measurement signal, perform windowing processing on it to obtain the frequency-domain measurement signals of each sensing section, and perform serial-to-parallel conversion, zero-padding to make their lengths all N, and IFFT processing on the frequency-domain measurement signals of each sensing section in sequence to obtain the wavelength-domain signals of each sensing section.
[0041] Among them, the number of zero-padding affects the time-consuming of the IFFT step. Therefore, preferably, the value of N is of the order of 10 2 magnitude.
[0042] S3. Input the wavelength-domain signals of each sensing section into the trained first machine learning model, and the trained first machine learning model performs parallel processing on them to obtain the roughly demodulated strain values of each sensing section; determine the reflection peak positions of the wavelength-domain signals according to the rough strain values, and intercept the local wavelength-domain signals near the reflection peaks that account for a preset proportion of the complete wavelength-domain signals, and input them into the trained second machine learning model, and the trained second machine learning model performs parallel processing on them to obtain the finely demodulated strain values of each sensing section; add the roughly demodulated strain values and the finely demodulated strain values of each sensing section to obtain the strain values to be measured of each sensing section.
[0043] In step S3, obtain strain values with lower resolution through the trained first machine learning model to complete rough strain demodulation, obtain strain values with higher resolution through the trained second machine learning model to complete fine strain demodulation, and add the strain values of rough demodulation and fine demodulation to obtain the final strain values.
[0044] Specifically, input the wavelength-domain signals of each sensing section into the trained first machine learning model to obtain the rough demodulation results (i.e., the strain values of rough demodulation). Among them, in the sample library for training the first machine learning model, the strain range of the samples is consistent with the strain range of the OFDR system, and the strain resolution value setting should be less than the strain resolution determined by the number of zero-padding in IFFT.
[0045] The rough demodulation result output by the first machine learning model is used to find the position of the reflection peak (the translation amount of the reflection peak is proportional to the strain value of the optical fiber, so the position of the reflection peak can be found according to the strain value obtained by rough demodulation). A local spectrum that accounts for a preset proportion (such as 3%) of the complete spectrum near the reflection peak is intercepted, and this local spectrum is input into the trained second machine learning model to obtain the fine demodulation result (i.e., the strain value obtained by fine demodulation). Among them, in the sample library for training the second machine learning model, the strain range of the sample library is determined by the maximum strain error of the first-stage demodulation result, and the strain resolution is set to the strain resolution that the OFDR system needs to achieve. The output fine demodulation result is used to compensate for the deviation between the rough demodulation result and the actual strain value.
[0046] It should be noted that the acquisition method of the dataset for the machine learning model is as follows: First, a time-domain reference signal is obtained through experiments. The time-domain reference signal is subjected to FFT to obtain a frequency-domain reference signal, and it is windowed to obtain the frequency-domain signal of a single sensing section. Then, a phase shift corresponding to a preset strain value is applied to the frequency-domain signal, and then IFFT is performed to obtain a wavelength-domain signal. At this time, the wavelength-domain signal will produce a translation corresponding to the strain value. Finally, for the first machine learning model, this wavelength-domain signal is used as a sample, and for the second machine learning model, a local spectrum that accounts for a preset proportion of the complete spectrum near the reflection peak of this wavelength-domain signal is intercepted as a sample, and the sample and the preset strain value are stored in the dataset as a sample-label pair.
[0047] Preferably, the first and second machine learning models include, but are not limited to, ANN, convolutional neural network, recurrent neural network, etc., and the first and second machine learning models are cascaded with each other.
[0048] Considering that the signal-to-noise ratio of the all-grating optical fiber is relatively high and there are reflection peaks in the spectrum, preferably, the sensing optical fiber used in the distributed optical fiber sensor is an all-grating optical fiber.
[0049] The sensing optical fiber used in the distributed optical fiber sensor can also be a standard single-mode optical fiber or a polarization-maintaining optical fiber. At this time, data enhancement needs to be performed on the frequency-domain measurement signal to make the wavelength-domain signals of each sensing section generate reflection peaks. The data enhancement includes: taking the dot product of the conjugate of the frequency-domain measurement signal and the frequency-domain reference signal; among them, the frequency-domain reference signal is obtained by performing parallel FFT processing on the time-domain reference signal before the strain to be measured acts on the distributed optical fiber sensor.
[0050] That is, when the sensing optical fiber used in the distributed optical fiber sensor adopts a standard single-mode optical fiber or a polarization-maintaining optical fiber, after multiplying the conjugate of the frequency-domain measurement signal by the frequency-domain reference signal and then performing windowing processing, the frequency-domain measurement signals of each sensing segment are obtained. The frequency-domain measurement signals of each sensing segment are sequentially subjected to serial-to-parallel conversion, zero-padding to make their lengths all N, and IFFT processing to obtain the wavelength-domain signals of each sensing segment. Only at this time can the wavelength-domain signals generate reflection peaks.
[0051] Preferably, the parallel FFT processing and the serial-to-parallel conversion are both implemented by a serial-to-parallel conversion device, and the first and second machine learning models are both deployed on the serial-to-parallel conversion device;
[0052] The serial-to-parallel conversion device is a GPU or an FPGA.
[0053] Specifically, the above-mentioned parallel FFT processing and serial-to-parallel conversion are both implemented by a serial-to-parallel conversion device. The serial-to-parallel conversion device is a GPU or a field-programmable gate array (FPGA), etc., which is used for parallel acceleration of matrix multiplication operations in FFT, IFFT, and the first and second machine learning models, etc., and can be used for serial-to-parallel conversion of frequency-domain signals to achieve parallel calculation of the strain values of each sensing segment in step S3.
[0054] Next, a specific example is used to further illustrate the method provided by the present invention.
[0055] In this example, both the first and second machine learning models adopt ANN, and a GPU is used as the serial-to-parallel conversion device.
[0056] The experimental device based on OFDR adopted in this example is as Figure 4 shown. The tunable laser 1 generates a wavelength-tunable light beam with a wavelength tuning range of 60 nm, a tuning speed of 60 nm / s, and an output power of 3.5 dBm. The wavelength-tunable light is divided into two beams after passing through the coupler 7 (splitting ratio 10:90), where 10% enters the auxiliary interferometer and 90% enters the main interferometer. The auxiliary interferometer consists of a coupler 8 (splitting ratio 50:50), a delay optical fiber 9 (length 10 m), and a coupler 10 (splitting ratio 50:50), and is detected by a photodetector 11. The main interferometer consists of a coupler 12 (splitting ratio 10:90), a polarization controller 13, a circulator 3, a fiber under test (i.e., the sensing optical fiber) 4 (length 3 m), a coupler 14 (splitting ratio 50:50), a first polarization beam splitter 15, a second polarization beam splitter 16, a first balanced detector 17, and a second balanced detector 18, and uses polarization diversity reception to suppress polarization fading. Finally, the data acquisition and processing module 19 is used to collect and process the OFDR signals.
[0057] The data acquisition and processing module 19 performs non-linear tuning compensation on the signals obtained by the first balance detector 17 and the second balance detector 18 to obtain a time-domain reference signal and a time-domain measurement signal.
[0058] The non-linear sweep compensation method includes a hardware compensation method and a software compensation method, such as an auxiliary interferometer trigger sampling method, a resampling method, a matched Fourier transform method, and a dechirp filtering method, etc.
[0059] Figure 5 It is a flowchart of the OFDR large strain rapid measurement method based on cascaded ANN and GPU adopted in the specific embodiment of the present invention. First, use the Figure 4 experimental device based on OFDR as shown to obtain a time-domain reference signal and a measurement signal, input the time-domain reference signal and the measurement signal into the GPU, and perform subsequent calculation steps in parallel in the GPU. First, obtain the frequency-domain signal through parallel FFT, perform serial-parallel conversion on the frequency-domain signal to obtain the parallel frequency-domain signals of each sensing section, and demodulate all sensing sections in parallel. Obtain the wavelength-domain signal of each sensing section through zero-padding IFFT with a zero-padding number of 500, then use the cascaded ANN to demodulate the strain curve distributed along the length, and output it to the CPU.
[0060] Figure 6 It is a flowchart of the algorithm of the cascaded ANN adopted in the specific embodiment of the present invention. After obtaining the wavelength-domain signal of the sensing section, input it into the first-stage ANN to obtain the roughly demodulated strain value. Intercept the local spectrum near the reflection peak of the wavelength-domain signal according to the roughly demodulated strain value, input the local spectrum signal into the second-stage ANN to obtain the finely demodulated strain value, add the roughly demodulated strain value and the finely demodulated strain value, and output the strain demodulation result.
[0061] Among them, the strain range of the sample library of the first-stage ANN is 24000 με, the strain resolution is set to 100 με, and the sample length is 500, that is, the complete spectrum is adopted; the strain range of the sample library of the second-stage ANN is set to 100 με, the strain resolution is set to 1 με, and the sample length is 15, that is, the local spectrum is adopted.
[0062] As Figure 7 shown, before and after using the demodulation method based on cascaded ANN and GPU in the present invention, the comparison of strain errors under different strain magnitudes. First, use the Figure 4The OFDR-based fiber optic strain sensing device shown obtains a time-domain signal, and compares the strain error variation curves of the demodulation method based on cascaded ANN and GPU in the present invention and the traditional demodulation method based on cross-correlation and CPU with the change of strain magnitude. Taking a sensing length of 0.8 m, a spatial resolution of 1 mm, and a constant tensile strain of 500 - 3000 με applied in a 0.5 m area as an example, when using the traditional cross-correlation and CPU demodulation method and the demodulation method based on cascaded ANN and GPU, the strain error performance is very close at different strain magnitudes, indicating that this method does not deteriorate the strain accuracy.
[0063] As Figure 8 shown, the comparison of the demodulation time consumption at different sensing lengths before and after using the demodulation method based on cascaded ANN and GPU in the present invention. First, use the Figure 4 OFDR-based fiber optic strain sensing device shown to obtain a time-domain signal, and compare the curve of the time consumption varying with the sensing length of the demodulation method based on cascaded ANN and GPU and the demodulation method based on cross-correlation and CPU in the present invention. Taking a sensing length of 1 - 10 m and a spatial resolution of 1 mm as an example, if the traditional cross-correlation demodulation method is used, the demodulation time consumption increases linearly with the increase of the sensing length, in the order of 10 s, and the time consumption performance is poor. When the demodulation method of cascaded ANN and GPU in the present invention is adopted, the demodulation time consumption drops significantly, in the order of 10 ms, about 1.2% of that using the cross-correlation-based demodulation method, effectively improving the demodulation time consumption performance.
[0064] In summary, the method provided by the present invention uses the OFDR-based fiber optic sensing technology to obtain a time-domain signal sequence, then inputs the time-domain signal into a serial-to-parallel conversion device, obtains a frequency-domain signal sequence through FFT, converts the frequency-domain signal through serial-to-parallel conversion to obtain parallel frequency-domain signals of each sensing segment, and demodulates all sensing segments in parallel. The wavelength-domain signal of each sensing segment is obtained through IFFT, and the wavelength-domain signal is input into a machine learning model for demodulation. First, the complete spectrum is used as the input to calculate the rough demodulation strain value with lower resolution, then the local spectrum is intercepted according to the rough demodulation strain value and used as the input to calculate the fine demodulation strain value with higher resolution, and finally the sum of the above two strain values is output as the final strain value. Generally speaking, this method greatly reduces the amount of data to be calculated, effectively improves the strain demodulation speed, and is applicable to strain demodulation in large strain scenarios without deteriorating the strain accuracy.
[0065] An embodiment of the present invention provides an electronic device, including: a computer-readable storage medium and a processor;
[0066] The computer-readable storage medium is used to store executable instructions;
[0067] The processor is configured to read the executable instructions stored in the computer-readable storage medium and execute the method as described in any of the above embodiments.
[0068] An embodiment of the present invention provides a computer-readable storage medium storing computer instructions for causing a processor to execute the method as described in any of the above embodiments.
[0069] An embodiment of the present invention provides a computer program product including a computer program or instructions, which when executed by a processor implement the method as described in any of the above embodiments.
[0070] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.
Claims
1. A method for rapid measurement of large strain using OFDR based on machine learning, characterized in that: include: Training phase: Performing FFT processing on the time domain reference signal before the preset strain acts on the distributed optical fiber sensor to obtain a frequency domain reference signal, and performing windowing processing on the time domain reference signal to decompose it into frequency domain reference signals of each sensing segment; After applying a phase shift corresponding to the preset strain to the frequency domain reference signal of each sensing segment, IFFT processing is performed to obtain a wavelength domain signal of each sensing segment; Using the wavelength domain signals of each sensing segment as samples and the preset strain as labels, training a first machine learning model; Using the local wavelength domain signal of each sensing segment near the reflection peak of the wavelength domain signal that accounts for a preset proportion of the complete wavelength domain signal as a sample and using the preset strain as a label to train the second machine learning model; Application phase: S1, obtaining the time domain measurement signal after the strain to be measured acts on the distributed optical fiber sensor; S2, performing parallel FFT processing on the time domain measurement signal to obtain a frequency domain measurement signal, performing windowing processing on the time domain measurement signal to obtain a frequency domain measurement signal of each sensing segment, and sequentially performing serial-to-parallel conversion, zero padding to make the length of the frequency domain measurement signal of each sensing segment N, and IFFT processing to obtain a wavelength domain signal of each sensing segment; S3, input the wavelength domain signal of each sensing segment into a trained first machine learning model, and the trained first machine learning model processes it in parallel to obtain the coarse demodulated strain value of each sensing segment; determine the reflection peak position of each wavelength domain signal according to each coarse strain value, and intercept the local wavelength domain signal near the reflection peak that occupies a preset proportion of the complete wavelength domain signal, and input it into a trained second machine learning model, and the trained second machine learning model processes it in parallel to obtain the fine demodulated strain value of each sensing segment; add the coarse demodulated strain value and the fine demodulated strain value of each sensing segment to obtain the strain value to be measured of each sensing segment.
2. The method according to claim 1, characterized in that The parallel FFT processing and serial-to-parallel conversion are both implemented by a serial-to-parallel conversion device, and the first and second machine learning models are both deployed on the serial-to-parallel conversion device; The serial-to-parallel conversion device is a GPU or FPGA.
3. The method according to claim 1 or 2, characterized in that The value of N is 10 2 Magnitude.
4. The method according to claim 1, characterized in that The sensing optical fiber used by the distributed optical fiber sensor is a full grating optical fiber.
5. The method according to claim 1, characterized in that The sensing optical fiber used in the distributed optical fiber sensor is a standard single-mode optical fiber or a polarization-maintaining optical fiber; In step S2, before windowing the frequency domain measurement signal to obtain the frequency domain measurement signal of each sensing segment, the method further includes: Performing a dot product of the frequency domain measurement signal and the conjugate of the frequency domain reference signal; The frequency domain reference signal is obtained by performing parallel FFT processing on the time domain measurement signal before the strain to be measured acts on the distributed optical fiber sensor.
6. The method according to claim 1, characterized in that The first and second machine learning models are ANN, convolutional neural network or recurrent neural network.
7. An electronic device, characterized in that: include: A computer readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the method according to any one of claims 1 to 6.
8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute the method according to any one of claims 1 to 6.
9. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the method according to any one of claims 1 to 6 is implemented.