High-speed Fabry-Perot signal demodulation method and system based on sparse spectrum

Abstract

The invention belongs to the technical field of optical fiber sensing, and discloses a high-speed Fabry-Perot signal demodulation method and system based on a sparse spectrum. The method comprises the following steps: selecting a cavity length range of a Fabry-Perot cavity, and generating a cavity length sequence; repeating simulation to obtain reflection spectrums of the Fabry-Perot cavity under all cavity lengths in the cavity length sequence; adding noise to the reflection spectrum to obtain an analog acquisition spectrum; selecting a channel according to the output wavelength of the wavelength division multiplexer, obtaining a sparse spectrum, and constructing a data set; training the neural network model through the data of the data set; after training is completed, a reflection signal of the Fabry-Perot cavity is obtained through a wavelength division multiplexer and a photoelectric detector array and converted into a sparse spectrum; and inputting the obtained sparse spectrum into the neural network model to obtain the cavity length of the optical fiber Fabry-Perot sensor and realize Fabry-Perot signal demodulation. The demodulation rate and the monitoring efficiency of the optical fiber Fabry-Perot sensor are improved, and an efficient and reliable remote monitoring solution is provided for multiple fields.

Description

Technical Field

[0001] The present invention relates to the field of optical fiber sensing technology, and more specifically, to a high-speed Fabry-Perot signal demodulation method and system based on sparse spectrum. Background Art

[0002] With the rapid development of aerospace, industrial manufacturing and other fields, the monitoring of high time-varying physical quantities such as noise, vibration, strain, and pressure has put forward higher requirements for sensors and corresponding signal processing technologies. Fiber Fabry-Perot sensors have become the first choice for exploring high-speed measurement sensors due to their advantages such as anti-electromagnetic interference, fast response, high sensitivity, intrinsically passive, and high temperature resistance.

[0003] There are many demodulation methods for fiber Fabry-Perot sensors, which can be divided into intensity demodulation, phase demodulation and wavelength demodulation based on the demodulation information. Different demodulation methods have different applicable scenarios. In the face of the high frequency and large impact test characteristics of rocket engines, a demodulation method with high speed and large dynamic range is needed. The existing intensity demodulation method can achieve fast demodulation by means of working point control or phase orthogonality, but it is only applicable to Fabry-Perot signals with small cavity length changes. The phase demodulation method can achieve a large dynamic range by means of full spectrum information or reference interferometer, but the low rate of the spectrometer and the complex program limit its demodulation rate. Wavelength demodulation uses sparse spectra to obtain cavity length information, avoiding the limitation that the working point can only be in the linear working area, and does not require a spectrometer to obtain full spectrum information. However, due to the limited sparse spectral characteristics, the resolution of this method is relatively low. In addition, the existing technologies do not have the functions of multi-terminal display and remote monitoring, and cannot efficiently demodulate fiber Fabry-Perot sensors. Therefore, it is necessary to explore demodulation methods that meet the use requirements.

[0004] Based on this, it is necessary to provide a method and a corresponding system for high-speed and high-precision demodulation of Fabry-Perot signals to solve the above problems. Summary of the invention

[0005] In order to solve the shortcomings of the prior art fiber Fabry-Perot sensor demodulation method in demodulation speed and demodulation accuracy, the present invention proposes a high-speed Fabry-Perot signal demodulation method and system based on sparse spectrum to improve the demodulation speed and accuracy and achieve high-speed and high-precision demodulation.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a high-speed Fabry-Perot signal demodulation method based on sparse spectrum, comprising the following steps:

[0007] Step 1: Select the cavity length range of the Fabry-Perot cavity and generate a cavity length sequence;

[0008] Step 2: Extract a cavity length value from the cavity length sequence, perform multi-beam interference simulation, and obtain the reflection spectrum of the Fabry-Perot cavity at the cavity length; repeat the simulation to obtain the reflection spectrum of the Fabry-Perot cavity at all cavity lengths in the cavity length sequence;

[0009] Step 3: Add noise to the reflection spectrum to obtain a simulated acquisition spectrum;

[0010] Step 4: Select a channel according to the output wavelength of the wavelength division multiplexer, extract the reflected light intensity corresponding to each channel from the simulated acquisition spectrum, and form a sparse spectrum; take each cavity length in the cavity length sequence as output and the corresponding sparse spectrum as input to construct a data set; preprocess the data set;

[0011] Step 5: Initialize the neural network, take the sparse spectrum corresponding to each cavity length as input, and the cavity length as output, and train the neural network model through the data set data;

[0012] Step 6: After the training is completed, the sensing signal output by the optical fiber Fabry-Perot sensor is received through the wavelength division multiplexer, and the reflection signal output by each channel of the wavelength division multiplexer is received by the photodetector array, and converted into a sparse spectrum;

[0013] Step 7: Input the obtained sparse spectrum into the neural network model, obtain the cavity length of the optical fiber Fabry-Perot sensor through the neural network model, and realize Fabry-Perot signal demodulation.

[0014] In step 2, a multi-beam interference simulation is performed on the Fabry-Perot cavity through a Python script to obtain the reflection spectrum of the Fabry-Perot cavity in the C band; the emission spectrum is composed of the reflection output intensity I R Composition, I R It is expressed as:

[0015]

[0016] Where F is the finesse factor of the Fabry-Perot cavity, δ represents the phase difference between adjacent light rays, and I0 represents the incident light intensity.

[0017] In step 3, the specific method of adding noise to the reflection spectrum to obtain the simulated acquisition spectrum is:

[0018] Gaussian white noise is generated by the random number generator in the Numpy library, and it is superimposed with the reflection spectrum to obtain a spectrum as a simulated acquisition spectrum.

[0019] In step 4, the step of preprocessing the data set includes:

[0020] Normalize the data of the dataset;

[0021] The normalized dataset is divided into a training set and a test set.

[0022] In step 5, the specific method of training and testing the neural network model using the data set data is:

[0023] Step 5.1: Initialize the neural network and select the number of hidden layers and neurons;

[0024] Step 5.2: Calculate the loss function LOSS for the prediction result obtained by the neural network calculation and the actual cavity length, and use the Adam optimizer to update the weight and bias, set the number of loops, and obtain the final loss function LOSS through continuous loop iteration;

[0025] Step 5.3: Determine whether the final loss function LOSS meets expectations. If so, the training ends and the neural network model is derived. If not, the number of hidden layers and neurons of the neural network is modified, and the cycle is iterated again until a neural network model with a loss function LOSS that meets the expected effect is obtained.

[0026] In addition, the present invention also provides a high-speed Fabry-Perot signal demodulation system based on sparse spectrum, which is used to execute the high-speed Fabry-Perot signal demodulation method based on sparse spectrum, comprising: a light source, an optical fiber Fabry-Perot sensor, an optical fiber circulator, a wavelength division multiplexer, a photodetector array, a synchronous acquisition card, and a computing unit; the wavelength division multiplexer is a dense wavelength division multiplexer, and its output port is provided with n channels with different wavelengths;

[0027] The C-band wide-spectrum light output by the light source is incident on the optical fiber Fabry-Perot sensor after passing through the optical fiber circulator. The reflected signal generated in the optical fiber Fabry-Perot sensor is incident on the wavelength division multiplexer after passing through the optical fiber circulator again. The reflected signal is spectrally divided by the wavelength division multiplexer. Then, the reflected signal output by each channel of the wavelength division multiplexer is received by the photoelectric detector array and photoelectrically converted to obtain a sparse spectrum with multiple different wavelengths of light. The sparse spectrum is collected by the synchronous acquisition card and then sent to the computing unit for demodulation.

[0028] The value range of n is 3-20.

[0029] The high-speed Fabry-Perot signal demodulation system based on sparse spectrum also includes a power supply and a voltage conversion module; the power supply is used to provide voltage, and the voltage conversion module is used to perform voltage conversion on the voltage provided by the power supply and then supply power to the system.

[0030] The high-speed Fabry-Perot signal demodulation system based on sparse spectrum also includes: a network card, a cloud platform and a display unit; after the computing unit completes data demodulation, the demodulated Fabry-Perot cavity length data is uploaded to the cloud platform through the network card; the cloud platform is used to draw the Fabry-Perot cavity length data in real time on a web page, and the display unit is connected to the Internet to access the web page to realize remote multi-terminal monitoring.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. The present invention utilizes the sparse spectrum of the optical fiber Fabry-Perot sensor for demodulation instead of using the complete interference spectrum, thereby solving the problem of low demodulation rate caused by the sampling rate limitation of the spectrometer and avoiding the waste of redundant information generated by the complete spectrum, thereby improving the demodulation rate of the Fabry-Perot signal.

[0033] 2. The present invention uses a neural network algorithm to achieve end-to-end demodulation from sparse spectra to Fabry-Perot cavity length, solving the problem of complex calculations and long calculation time caused by traditional demodulation algorithms relying on loop processing, phase correction, etc. At the same time, the TensorRT framework is used to accelerate the inference of the model to ensure that the calculation time of batch data is much shorter than the acquisition time, avoiding the memory crisis caused by data accumulation caused by the long-term operation of traditional algorithms, and providing reference value for the stable demodulation of long-term Fabry-Perot signals.

[0034] 3. The present invention uses the Internet of Things technology to achieve remote multi-terminal monitoring of Fabry-Perot sensor demodulation. Traditional demodulation methods rely on local hardware and manual operation, have poor monitoring flexibility, low real-time performance, and high maintenance costs. Through remote communication, multi-terminal synchronous access, intelligent early warning and fault diagnosis, dynamic configuration optimization, and easy expansion and maintenance, the present invention not only improves monitoring efficiency and response speed, promotes information sharing and teamwork, but also achieves optimal resource allocation and reduces operating costs, providing efficient and reliable remote monitoring solutions for multiple fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A schematic flow chart of a high-speed Fabry-Perot signal demodulation method based on sparse spectrum provided in Embodiment 1 of the present invention;

[0036] Figure 2 It is a schematic diagram of the demodulation principle of the neural network in the present invention;

[0037] Figure 3 Schematic diagram of the structure of the demodulation device used in the first embodiment of the present invention;

[0038] Figure 4 Comparison chart for neural network model predictions;

[0039] Figure 5A schematic structural diagram of a high-speed Fabry-Perot signal demodulation system based on sparse spectrum provided in Embodiment 1 of the present invention;

[0040] Figure 6 It is a flowchart of remote multi-terminal monitoring;

[0041] In the figure: 1-power supply, 2-voltage conversion module, 3-light source, 4-fiber Fabry-Perot sensor, 5-fiber circulator, 6-wavelength division multiplexer, 7-photodetector, 8-synchronous acquisition card, 9-computing unit, 10-network card, 11-cloud platform, 12-display unit. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0043] Embodiment 1

[0044] like Figure 1 As shown, the first embodiment of the present invention provides a high-speed Fabry-Perot signal demodulation method based on sparse spectrum, comprising the following steps:

[0045] Step 1: Select the cavity length range of the Fabry-Perot cavity and generate a cavity length sequence.

[0046] Different cavity lengths will produce different reflection spectra. A larger cavity length range will bring more data sets and increase the training pressure of the model, but it will improve the model's demodulation ability for Fabry-Perot cavities with different cavity lengths. Cavity length sequences are generated at certain intervals within the selected cavity length range. The smaller the interval, the higher the demodulation resolution, but the greater the model training pressure.

[0047] Step 2: Extract a cavity length value from the cavity length sequence, perform multi-beam interference simulation, and obtain the reflection spectrum of the Fabry-Perot cavity at this cavity length; repeat the simulation to obtain the reflection spectrum of the Fabry-Perot cavity at all cavity lengths in the cavity length sequence.

[0048] In step 2, a multi-beam interference simulation is performed on the Fabry-Perot cavity through a Python script to obtain the reflection spectrum of the Fabry-Perot cavity in the C band; the emission spectrum is composed of the reflection output intensity I R Composition, I R It is expressed as:

[0049]

[0050] Where F is the finesse factor of the Fabry-Perot cavity, δ represents the phase difference between adjacent light rays, and I0 represents the incident light intensity. The expression of the finesse factor F is:

[0051]

[0052] Where R is the reflectivity of the Fabry-Perot cavity.

[0053] When the medium in the Fabry-Perot cavity is air, that is, the refractive index of the medium is about 1, the phase difference δ of adjacent light rays can be expressed as:

[0054]

[0055] Where h is the Fabry-Perot cavity length and λ is the wavelength.

[0056] Step 3: Add noise to the reflection spectrum to obtain the simulated acquisition spectrum.

[0057] In step 3, the specific method of adding noise to the reflection spectrum to obtain the simulated acquisition spectrum is:

[0058] Gaussian white noise is generated by the random number generator in the Numpy library, and it is superimposed with the reflection spectrum to obtain a spectrum as a simulated acquisition spectrum.

[0059] Step 4: Select a channel according to the output wavelength of the wavelength division multiplexer, extract the reflected light intensity corresponding to each channel from the simulated acquisition spectrum, and form a sparse spectrum; take each cavity length in the cavity length sequence as output and the corresponding sparse spectrum as input to construct a data set; preprocess the data set.

[0060] In step 4, the step of preprocessing the data set includes:

[0061] Normalize the data of the dataset;

[0062] The normalized dataset is divided into a training set and a test set.

[0063] Step 5: Initialize the neural network, take the sparse spectrum corresponding to each cavity length as input, and the cavity length as output, and train the neural network model using the data set data.

[0064] like Figure 2 As shown, this is a schematic diagram of the principle of neural network data training.

[0065] In step 5, the specific method of training and testing the neural network model using the data set data is:

[0066] Step 5.1: Initialize the neural network and select the number of hidden layers and neurons;

[0067] Step 5.2: Calculate the loss function LOSS for the prediction result obtained by the neural network calculation and the actual cavity length, and use the Adam optimizer to update the weight and bias, set the number of loops, and obtain the final loss function LOSS through continuous loop iteration;

[0068] Step 5.3: Determine whether the final loss function LOSS meets expectations. If so, the training ends and the neural network model is derived. If not, the number of hidden layers and neurons of the neural network is modified, and the cycle is iterated again until a neural network model with a loss function LOSS that meets the expected effect is obtained.

[0069] The model prediction results are compared Figure 4 As shown, the linear fitting line R 2 =1, which has a lower prediction error, indicating that the neural network model of the embodiment of the present invention can realize accurate demodulation of the cavity length.

[0070] Step 6: After the training is completed, the sensing signal output by the optical fiber Fabry-Perot sensor is received through a wavelength division multiplexer, and the reflected signal output by each channel of the wavelength division multiplexer is received by a photodetector array and converted into a sparse spectrum.

[0071] like Figure 3 As shown, it is a schematic diagram of using a wavelength division multiplexer and a photodetector array to receive a reflected signal and obtain a sparse spectrum. Among them, the wavelength division multiplexer 6 is a dense wavelength division multiplexer, and its output port is provided with n channels with different wavelengths. The wavelength of each channel is the same as the wavelength when the reflected light intensity is taken out from the simulated acquisition spectrum in step 4 to form a sparse spectrum.

[0072] Step 7: Input the obtained sparse spectrum into the neural network model, obtain the cavity length of the optical fiber Fabry-Perot sensor through the neural network model, and realize Fabry-Perot signal demodulation.

[0073] Specifically, the neural network model is set in a computing unit, and specifically, the computing unit is a desktop GPU or an embedded GPU.

[0074] Specifically, step 7 includes the following steps: converting the model obtained in step 5 to obtain a model in a universal format, using TensorRT to accelerate reasoning, thereby deploying the neural network model to a desktop GPU or an embedded GPU, and then inputting the obtained sparse spectrum into the desktop GPU or the embedded GPU, obtaining the cavity length of the optical fiber Fabry-Perot sensor through the neural network model in the desktop GPU or the embedded GPU, and realizing Fabry-Perot signal demodulation.

[0075] Embodiment 2

[0076] like Figure 5As shown, embodiment 2 of the present invention provides a high-speed Fabry-Perot signal demodulation system based on sparse spectrum, which is used to execute a high-speed Fabry-Perot signal demodulation method based on sparse spectrum described in embodiment 1, including: a light source 3, an optical fiber Fabry-Perot sensor 4, an optical fiber circulator 5, a wavelength division multiplexer 6, a photodetector array 7, a synchronous acquisition card 8, and a computing unit 9; the wavelength division multiplexer 6 is a dense wavelength division multiplexer, and its output port is provided with n channels with different wavelengths.

[0077] The C-band wide spectrum light output by the light source 3 is incident on the optical fiber Fabry-Perot sensor 4 after passing through the optical fiber circulator 5. The reflected signal generated in the optical fiber Fabry-Perot sensor 4 is incident on the wavelength division multiplexer 6 after passing through the optical fiber circulator 5 again. The reflected signal is spectrally divided by the wavelength division multiplexer 6. Then, the reflected signal output by each channel of the wavelength division multiplexer is received by the photoelectric detector array 7 and photoelectrically converted to obtain a sparse spectrum with multiple different wavelengths of light. The sparse spectrum is collected by the synchronous acquisition card 8 and then sent to the computing unit 9 for demodulation. The computing unit 9 is provided with a neural network model for demodulation.

[0078] Specifically, in this embodiment, the value range of n is 3-20.

[0079] Furthermore, a high-speed Fabry-Perot signal demodulation system based on sparse spectrum in this embodiment also includes a power supply 1 and a transformer module 2; the power supply 1 is used to provide voltage, and the transformer module 2 is used to convert the voltage provided by the power supply 1 to power the system.

[0080] In this embodiment, the computing unit 9 uses a deployed neural network model to achieve high-speed demodulation of the sparse spectrum, wherein the input of the neural network model is the light intensity of n wavelengths contained in the sparse spectrum, and the output is the Fabry-Perot cavity length.

[0081] Furthermore, the high-speed Fabry-Perot signal demodulation system based on sparse spectrum also includes: a network card 10, a cloud platform 11 and a display unit 12; after the computing unit 9 completes data demodulation, the demodulated Fabry-Perot cavity length data is uploaded to the cloud platform 11 through the network card 10; the cloud platform 11 is used to draw the Fabry-Perot cavity length data in real time on the web page, and the display unit 12 is connected to the Internet to access the web page to realize remote multi-terminal monitoring.

[0082] Specifically, Figure 6 As shown, the working steps are as follows:

[0083] 1. Turn on the system and collect the sparse spectrum from the optical fiber Fabry-Perot sensor 4 at high speed through the synchronous acquisition card 8;

[0084] 2. Input the sparse spectrum into the deployed neural network model and output the Fabry-Perot cavity length;

[0085] 3. The compressed Fabry-Perot cavity length data is uploaded to the cloud platform 11 via the MQTT protocol;

[0086] 4. The web page refreshes and displays batch cavity length data;

[0087] 5. Access the web page to view the cavity length data of the fiber optic Fabry-Perot sensor 4

[0088] 6. Obtain batches of sparse spectra and repeat steps 2-4 to achieve real-time remote monitoring.

[0089] In this embodiment, after obtaining a batch of cavity length data, the computing unit 9 compresses and packages the data to reduce the pressure on the network bandwidth. The network card 10 is used to provide networking function for the computing unit 9, and uploads it to the cloud platform 11 through the MQTT protocol. The cloud platform 11 creates a web page to receive and display the data. The display unit 12 is a display device with networking function, which accesses the web page through the Internet for remote monitoring.

[0090] The working principle of this embodiment is as follows: a large number of cavity lengths and corresponding sparse spectrum data sets are obtained through multi-beam interference simulation, which are divided into training sets and test sets. The training set is used to be put into the neural network for cyclic iteration to optimize the number of hidden layers and neurons. The neural network model after training is deployed to the computing unit with GPU after format conversion, TensorRT framework acceleration and other processes. When the wide-spectrum light source emitted by the light source 3 enters the optical fiber Fabry-Perot sensor 4 through the optical fiber circulator 5, the reflected spectrum enters the wavelength division multiplexer 6 through the optical fiber circulator 5 to complete the wavelength division and obtain the sparse spectrum. The photoelectric detector array 7 performs photoelectric conversion on the sparse spectrum, and the synchronous acquisition card 8 collects the sparse spectrum at high speed and transmits it to the computing unit 9. The neural network model calculates the Fabry-Perot cavity length and uploads it to the cloud platform 11 to achieve remote multi-terminal monitoring. The sampling rate limitation brought by the use of a spectrometer is avoided, and the sampling rate of the original data is improved from the hardware. At the same time, the accelerated processing of the neural network model effectively improves the software processing rate, solving the problem of inefficient calculation of serial calculation and cyclic correction in the traditional demodulation method. The use of Internet of Things technology to achieve remote multi-terminal monitoring of fiber-optic Fabry-Perot sensors not only improves monitoring efficiency and response speed, promotes information sharing and team collaboration, but also achieves optimal allocation of resources and reduces operating costs, providing efficient and reliable remote monitoring solutions for multiple fields.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A high-speed Fabry-Perot signal demodulation method based on sparse spectrum, characterized in that: The following steps are involved: Step 1: Select the cavity length range of the Fabry-Perot cavity and generate a cavity length sequence; Step 2: Extract a cavity length value from the cavity length sequence, perform multi-beam interference simulation, and obtain the reflection spectrum of the Fabry-Perot cavity at the cavity length; repeat the simulation to obtain the reflection spectrum of the Fabry-Perot cavity at all cavity lengths in the cavity length sequence; Step 3: Add noise to the reflection spectrum to obtain a simulated acquisition spectrum; Step 4: Select a channel according to the output wavelength of the wavelength division multiplexer, and extract the reflected light intensity corresponding to each channel from the simulated acquisition spectrum to form a sparse spectrum; Each cavity length in the cavity length sequence is taken as output, and the corresponding sparse spectrum is taken as input to construct a data set; the data set is preprocessed; Step 5: Initialize the neural network, take the sparse spectrum corresponding to each cavity length as input, and the cavity length as output, and train the neural network model through the data set data; Step 6: After the training is completed, the sensing signal output by the optical fiber Fabry-Perot sensor is received through the wavelength division multiplexer, and the reflection signal output by each channel of the wavelength division multiplexer is received by the photodetector array, and converted into a sparse spectrum; Step 7: Input the obtained sparse spectrum into the neural network model, obtain the cavity length of the optical fiber Fabry-Perot sensor through the neural network model, and realize Fabry-Perot signal demodulation.

2. A high-speed Fabry-Perot signal demodulation method based on sparse spectrum according to claim 1, characterized in that: In step 2, a multi-beam interference simulation is performed on the Fabry-Perot cavity through a Python script to obtain the reflection spectrum of the Fabry-Perot cavity in the C band; the emission spectrum is composed of the reflection output intensity I R Composition, I R It is expressed as: Where F is the finesse factor of the Fabry-Perot cavity, δ represents the phase difference between adjacent light rays, and I0 represents the incident light intensity.

3. A high-speed Fabry-Perot signal demodulation method based on sparse spectrum according to claim 1, characterized in that: In step 3, the specific method of adding noise to the reflection spectrum to obtain the simulated acquisition spectrum is: Gaussian white noise is generated by the random number generator in the Numpy library, and it is superimposed with the reflection spectrum to obtain a spectrum as a simulated acquisition spectrum.

4. A high-speed Fabry-Perot signal demodulation method based on sparse spectrum according to claim 1, characterized in that: In step 4, the step of preprocessing the data set includes: Normalize the data of the dataset; The normalized dataset is divided into a training set and a test set.

5. A high-speed Fabry-Perot signal demodulation method based on sparse spectrum according to claim 4, characterized in that: In step 5, the specific method of training and testing the neural network model using the data set data is: Step 5.1: Initialize the neural network and select the number of hidden layers and neurons; Step 5.2: Calculate the loss function LOSS for the prediction result obtained by the neural network calculation and the actual cavity length, and use the Adam optimizer to update the weight and bias, set the number of loops, and obtain the final loss function LOSS through continuous loop iteration; Step 5.3: Determine whether the final loss function LOSS meets expectations. If so, the training ends and the neural network model is derived. If not, the number of hidden layers and neurons of the neural network is modified, and the cycle is iterated again until a neural network model with a loss function LOSS that meets the expected effect is obtained.

6. A high-speed Fabry-Perot signal demodulation system based on sparse spectrum, characterized in that: A method for high-speed Fabry-Perot signal demodulation based on sparse spectrum according to claim 1, comprising: a light source (3), an optical fiber Fabry-Perot sensor (4), an optical fiber circulator (5), a wavelength division multiplexer (6), a photodetector array (7), a synchronous acquisition card (8), and a computing unit (9); the wavelength division multiplexer (6) is a dense wavelength division multiplexer, and its output port is provided with n channels with different wavelengths; The C-band wide-spectrum light output by the light source (3) passes through the optical fiber circulator (5) and is incident on the optical fiber Fabry-Perot sensor (4). The reflected signal generated in the optical fiber Fabry-Perot sensor (4) passes through the optical fiber circulator (5) again and is incident on the wavelength division multiplexer (6). The wavelength division multiplexer (6) performs spectral wavelength division on the reflected signal. The reflected signal output by each channel of the wavelength division multiplexer is then received by the photoelectric detector array (7) and photoelectrically converted to obtain a sparse spectrum with multiple different wavelengths of light. The sparse spectrum is collected by the synchronous acquisition card (8) and then sent to the calculation unit (9) for demodulation.

7. A high-speed Fabry-Perot signal demodulation system based on sparse spectrum according to claim 6, characterized in that: The value range of n is 3-20.

8. The high-speed Fabry-Perot signal demodulation system based on sparse spectrum according to claim 6, characterized in that: It also comprises a power supply (1) and a voltage conversion module (2); the power supply (1) is used to provide voltage, and the voltage conversion module (2) is used to perform voltage conversion on the voltage provided by the power supply (1) and then supply power to the system.

9. The high-speed Fabry-Perot signal demodulation system based on sparse spectrum according to claim 6, characterized in that: Also includes: A network card (10), a cloud platform (11) and a display unit (12); after the computing unit (9) completes data demodulation, the demodulated Fabry-Perot cavity length data is uploaded to the cloud platform (11) via the network card (10); the cloud platform (11) is used to draw the Fabry-Perot cavity length data in real time on a web page, and the display unit (12) is connected to the Internet to access the web page, thereby realizing remote multi-terminal monitoring.

Images