CN114257313B - Fiber grating sensing system noise reduction method based on deep learning and related equipment thereof - Google Patents
Fiber grating sensing system noise reduction method based on deep learning and related equipment thereof Download PDFInfo
- Publication number
- CN114257313B CN114257313B CN202111556911.9A CN202111556911A CN114257313B CN 114257313 B CN114257313 B CN 114257313B CN 202111556911 A CN202111556911 A CN 202111556911A CN 114257313 B CN114257313 B CN 114257313B
- Authority
- CN
- China
- Prior art keywords
- noise
- signal
- deep learning
- noise reduction
- fbg
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
- H04B10/60—Receivers
- H04B10/66—Non-coherent receivers, e.g. using direct detection
- H04B10/69—Electrical arrangements in the receiver
- H04B10/697—Arrangements for reducing noise and distortion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
- H04B10/25—Arrangements specific to fibre transmission
Abstract
The application provides a fiber grating sensing system noise reduction method based on deep learning, electronic equipment and a computer readable storage medium, wherein the method comprises the following steps: preprocessing a plurality of collected FBG reflected signal samples to obtain a plurality of first FBG reflected signals; adding preset noise into the plurality of first FBG reflection signals to obtain a plurality of groups of second FBG reflection signals, wherein the preset noise comprises at least one of device interference noise, white Gaussian noise, alpha stable distribution noise and non-stable cable wind dance noise; dividing a plurality of groups of second FBG (fiber Bragg Grating) reflection signals into a training set and a verification set so as to train a deep learning noise reduction model; and carrying out noise reduction treatment on the FBG reflected signals acquired by the fiber bragg grating sensing system based on the trained deep learning noise reduction model. The deep learning noise reduction model based on training can be used for denoising FBG (fiber Bragg Grating) reflection signals interfered by different degrees, and is high in robustness and good in denoising effect.
Description
Technical Field
The application relates to the technical field of optical fiber sensing, in particular to a deep learning-based optical fiber grating sensing system noise reduction method, electronic equipment and a computer readable storage medium.
Background
The long-distance large-capacity fiber grating sensing system uses the weak-reflectivity fiber grating sensors to increase the multiplexing number, and combines the optical amplification technology to prolong the sensing distance, so that the requirement of ultra-long-distance dense monitoring application can be met. However, due to accumulation of long-distance optical fiber transmission loss and weak reflectivity of the optical fiber grating, the received sensing signal is very weak and is seriously affected by external environment and internal noise of a system, and the sensing signal is possibly buried in the noise, so that a high-quality spectrum cannot be provided for high-precision measurement of multiple parameters such as temperature and strain of the sensing signal.
In recent years, many researchers have denoised the fiber grating sensing signal using wavelet denoising and empirical mode analysis (EMD). In wavelet denoising, whether a threshold can be correctly set is a key for improving the denoising effect, and although various threshold processing methods exist, the wavelet denoising is not accurate enough; empirical mode analysis decomposes a signal into a plurality of components, but in the decomposition process, the selection of a mode mixing problem and an effective component has a great influence on the denoising performance. The two methods have the advantages of complex calculation process, large calculation amount, long running time and insufficient denoising effect on the weak signals of the long-distance and large-capacity fiber bragg grating sensing system. Therefore, a better noise reduction method needs to be explored to realize multi-parameter high-precision measurement of the system.
Disclosure of Invention
In view of the above, the present application provides a method for reducing noise of a fiber grating sensing system based on deep learning, an electronic device and a computer readable storage medium, which can solve the problem that the existing long-distance large-capacity fiber grating sensing system has poor signal.
An embodiment of the present application provides a fiber grating sensing system noise reduction method based on deep learning, including: collecting a plurality of Fiber Bragg Grating (FBG) reflected signal samples, and preprocessing the plurality of FBG reflected signal samples to obtain a plurality of first FBG reflected signals; adding preset noise to the plurality of first FBG reflection signals to obtain a plurality of groups of second FBG reflection signals, wherein the preset noise comprises at least one of device interference noise with random intensity, gaussian white noise, alpha stable distribution noise and non-stable cable wind dance noise, and each group of second FBG reflection signals in the plurality of groups of second FBG reflection signals comprises an FBG reflection signal not containing the preset noise and an FBG reflection signal containing the preset noise; dividing the plurality of groups of second FBG reflected signals into a training set and a verification set; training a deep learning noise reduction model based on the training set and the verification set until determining model parameters of the deep learning noise reduction model; and carrying out noise reduction treatment on the FBG reflected signals acquired by the fiber bragg grating sensing system based on the trained deep learning noise reduction model.
In some embodiments, the pre-processing the plurality of FBG reflected signal samples to obtain a plurality of first FBG reflected signals includes: converting the binary data file of the FBG reflected signal samples into a decimal data file, and converting data in the decimal data file into a two-dimensional array by taking the effective sampling points of the spectrum as the column number; and visually displaying the two-dimensional array according to the rows to select the FBG reflected signal with the spectral signal-to-noise ratio larger than a preset threshold value as the first FBG reflected signal.
In some embodiments, the number of effective sampling points of the spectrum is 512, and the preset threshold is 4dB.
In some embodiments, the deep learning noise reduction model includes an input layer, a signal encoding unit, a residual puncturing unit, a signal decoding unit, and an output layer, and the training the deep learning noise reduction model based on the training set and the verification set includes: using a set of training data in the training set as input to the input layer; encoding, at the signal encoding unit, the FBG reflected signal containing the preset noise in the set of training data, performing soft thresholding on the encoding result by the residual error puncturing unit to obtain a noise characteristic, decoding the processing result by the residual error puncturing unit by the signal decoding unit, and outputting a reconstructed sample by the output layer; reversely propagating the reconstruction error of the FBG reflected signal which does not contain the preset noise in the reconstruction sample and the set of training data, and judging whether the loss is greater than the preset error; if the loss is larger than the preset error, taking the next group of training data in the training set as the input of the input layer to continue training the deep learning noise reduction model; if the loss is smaller than or equal to the preset error, parameters of the deep learning noise reduction model are reserved, and the deep learning noise reduction model is tested by using the verification set.
In some embodiments, the device interferes with the expression I of noise n (λ) is:
wherein, in the expression I n In (lambda), I 0 Lambda is the wavelength satisfying the Bragg condition, n is the refractive index of the medium, I (lambda) is the light intensity reflected by the fiber grating, nL is the optical path difference of the light wave, a 2 In order to be the intensity of the noise,
phase change of the reflection point caused by external environment;
characteristic function of alpha stable distribution noise
Satisfies the following conditions:
wherein, in the characteristic function
Where i is complex imaginary unit, t is signal time unit, alpha is characteristic index, beta is symmetric parameter, gamma is dispersion coefficient, delta is position parameter, 0<α≤2,-1≤β≤1，γ>0，-∞<δ<∞；
Expression n of non-stationary cable wind-dance noise wind Comprises the following steps:
n wind ＝[1+acos(2πft)]n g (t)+n 0 (t)；
wherein, in the expression n wind In, n g (t) is white Gaussian noise, n 0 (t) represents ambient noise, and a is the signal non-stationary modulation amplitude.
In some embodiments, the signal encoding unit comprises two convolutional layers, the operating principle of each convolutional layer is described as:
z＝f(x)＝S f (wx+b z )；
wherein, x = { x 1 ,x 2 ,x 3 ,…,x N Is the signal input to the convolutional layer, z = { z = 1 ,z 2 ,z 3 ,…,z N Is a characteristic signal obtained by processing the convolution layer, S f In order to activate the function(s),
b z w is a weight matrix for an offset vector in the encoding process; />
The signal decoding unit comprises two deconvolution layers, and the working principle of each deconvolution layer is described as follows:
y＝g(z)＝S g (w’z+b y )；
wherein y = { y = 1 ,y 2 ,y 3 ,…,y N Is a reconstructed signal obtained by deconvolution layer processing, S g In order to activate the function(s),
b y w' is a weight matrix for the offset vector in the decoding process.
In some embodiments, the residual puncturing unit comprises three residual puncturing networks, and the operation principle of each residual puncturing network is described as follows:
wherein x is input characteristics, y is output characteristics, tau is a threshold, in the residual shrinkage network, the characteristic signals after the global mean pooling are input into a two-layer full-connection network to obtain a scaling parameter z, the scaling parameter z is normalized between 0 and 1 through a Sigmoid function to obtain a new scaling parameter alpha, alpha is multiplied by the average value of the absolute values of the characteristic signals to obtain the threshold tau,
i. j, c represent the width, length and channel number of the characteristic signal x, respectively.
In some embodiments, the expression of the Loss function Loss of reconstruction error:
wherein y' = { y = 1 ’,y 2 ’,y 3 ’,…,y N ' } is the expected output value, and the loss function optimization objective is:
λ is the fixed parameter of the model, N is the number of signal sampling points, x i For the ith component of the input convolutional layer signal, the objective w is optimized using a gradient descent solution function * The iteration function is:
α is the learning rate.
An embodiment of the present application provides a computer-readable storage medium, which stores computer instructions, and when the computer instructions are executed on an electronic device, the electronic device executes the above noise reduction method for a fiber grating sensing system based on deep learning.
An embodiment of the present application provides an electronic device, where the electronic device includes a processor and a memory, where the memory is configured to store instructions, and the processor is configured to call the instructions in the memory, so that the electronic device executes the above method for reducing noise of a fiber bragg grating sensing system based on deep learning.
According to the fiber bragg grating sensing system noise reduction method based on deep learning, the electronic equipment and the computer readable storage medium, the feature expression of FBG (fiber bragg grating) reflection signals is learned based on a deep learning network, the noise features are processed by using a residual shrinkage unit, finally, input signals are reconstructed to obtain denoised signals, the deep learning denoising model adopts noise interference signals with random types and uncertain energy sizes and action ranges as training samples, model parameters are updated to minimize reconstruction errors in continuous iterative optimization to enable the model output signals to be close to target signals, the FBG reflection signals interfered to different degrees can be denoised by the trained deep learning denoising model, and the method is suitable for denoising of weak signals of long-distance and large-capacity fiber bragg grating sensing systems, high in robustness and good in denoising effect.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the description below are some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
Fig. 1 is a flowchart illustrating steps of a noise reduction method for a fiber grating sensing system based on deep learning according to an embodiment of the present application.
Fig. 2a to 2b are schematic diagrams illustrating an architecture of a deep learning noise reduction model in an embodiment of the present application.
Fig. 3 is a schematic diagram of an architecture of a residual error puncturing network according to an embodiment of the present application.
Fig. 4 a-4 b are schematic diagrams illustrating the structure of an activation function according to an embodiment of the present application.
Fig. 5a is a schematic waveform diagram of an FBG reflected signal collected in an embodiment of the present application before denoising.
Fig. 5b is a schematic waveform diagram of an FBG reflected signal collected in an embodiment of the present application after denoising.
FIG. 6 is a functional block diagram of an electronic device according to an embodiment of the present application.
Detailed Description
In order that the above objects, features and advantages of the present application can be more clearly understood, a detailed description of the present application will be made below with reference to the accompanying drawings and detailed description. In addition, the embodiments and features of the embodiments of the present application may be combined with each other without conflict.
In the following description, numerous specific details are set forth to provide a thorough understanding of the present application, and the described embodiments are merely a subset of the embodiments of the present application, rather than all embodiments.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the description of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.
The fiber bragg grating sensing system noise reduction method based on deep learning can be applied to one or more electronic devices. The electronic device is a device capable of automatically performing numerical calculation and/or information processing according to a preset or stored instruction, and the hardware includes, but is not limited to, a Processor, a micro programmed Control Unit (MCU), an Application Specific Integrated Circuit (ASIC), a Programmable Gate Array (FPGA), a Digital Signal Processor (DSP), an embedded device, and the like.
The electronic device may be a desktop computer, an optical communication device, a server, or other computing device.
Fig. 1 is a flowchart illustrating steps of an embodiment of a noise reduction method for a fiber grating sensing system based on deep learning according to the present application. The order of the steps in the flow chart may be changed and some steps may be omitted according to different needs.
Referring to fig. 1, the method for reducing noise of a fiber grating sensing system based on deep learning may include the following steps.
And S11, collecting a plurality of FBG (fiber Bragg Grating) reflected signal samples, and preprocessing the plurality of FBG reflected signal samples to obtain a first FBG reflected signal.
In some embodiments, the FBG reflected signal of the long-distance large-capacity fiber grating sensing system collected in a period of time can be used as an FBG reflected signal sample.
In some embodiments, pre-processing the plurality of FBG reflected signal samples may include: converting a binary data file of a plurality of FBG (fiber Bragg Grating) reflected signal samples into a decimal data file, and converting data in the decimal data file into a two-dimensional array by taking the effective sampling points of a spectrum as column numbers; and visually displaying the two-dimensional array by rows to select the FBG reflection signal with the spectral signal-to-noise ratio larger than a preset threshold value as a first FBG reflection signal. The number of effective sampling points and the preset threshold value can be set according to the actual use scene, for example, the number of effective sampling points is 512, and the preset threshold value is 4dB.
And S12, adding preset noise to the plurality of first FBG reflected signals to obtain a plurality of groups of second FBG reflected signals.
In some embodiments, the preset noise may include at least one of random intensity device interference noise, white gaussian noise, alpha stationary distributed noise, and non-stationary cable wind-dance noise. Each of the plurality of sets of second FBG reflected signals may include an FBG reflected signal not including the predetermined noise and an FBG reflected signal including the predetermined noise.
And S13, dividing the plurality of groups of second FBG reflected signals into a training set and a verification set.
In some embodiments, the division ratio may be set according to an actual usage scenario, so as to divide the plurality of sets of the second FBG reflected signals into the training set and the verification set based on the division ratio. For example, the division ratio is 5.
And S14, training the deep learning noise reduction model based on the training set and the verification set until the model parameters of the deep learning noise reduction model are determined.
In some embodiments, the deep learning noise reduction model includes an input layer, a signal encoding unit, a residual puncturing unit, a signal decoding unit, and an output layer, and may be trained by: taking a group of training data in the training set as input of an input layer; the method comprises the steps that FBG (fiber Bragg Grating) reflection signals containing preset noise in a set of training data are coded in a signal coding unit, soft thresholding processing noise characteristics are carried out on coding results through a residual error shrinking unit, processing results of the residual error shrinking unit are decoded through a signal decoding unit, and a reconstructed sample is output through an output layer; reversely propagating the reconstruction error of the FBG reflected signal which does not contain the preset noise in the reconstruction sample and the set of training data, and judging whether the loss is greater than the preset error or not; if the loss is larger than the preset error, taking the next group of training data in the training set as the input of the input layer to continue training the deep learning noise reduction model; and if the loss is less than or equal to the preset error, reserving the parameters of the deep learning noise reduction model, and testing the deep learning noise reduction model by using the verification set.
And S15, carrying out noise reduction treatment on the FBG reflected signals collected by the fiber bragg grating sensing system based on the trained deep learning noise reduction model.
In some embodiments, when the deep learning noise reduction model training is completed, the FBG reflected signal collected by the long-distance large-capacity fiber grating sensing system may be input to the deep learning noise reduction model, so as to perform noise reduction processing on the collected FBG reflected signal.
As shown in fig. 2a to 2b, the deep learning noise reduction model includes an input layer, a signal encoding unit, a residual shrinking unit, a signal decoding unit, and an output layer. The signal encoding unit may be stacked by 2 convolutional layers, each of which may include a structure of a convolution function + an activation function, and one convolution operation and one activation process are 1 convolutional layer. The signal decoding unit may be stacked by 2 deconvolution layers, each of which may include a structure of a deconvolution function + an activation function, and one deconvolution operation and one activation process are 1 deconvolution layer. The residual error contraction unit is formed by stacking 3 layers of residual error contraction networks, and each residual error contraction network can comprise a convolution layer, a rectification linear unit activation function, batch standardization, global mean pooling and a full-connection output layer. The structure of the residual shrinkage network can be as shown in fig. 3.
The activation function is used for realizing activation, and is a process of mapping input of the neuron to output by using a preset function and adding a nonlinear factor to enhance the nonlinear fitting capability of the model, for example, the preset function can be a Sigmoid function, a tanh function, a Relu function and the like. As shown in fig. 4a and 4b, taking the activation process of a single neuron and a plurality of neurons as examples, x1, x2, and x3 are used as initial inputs, and a deviation with a size of b is added, and an output h obtained through the neurons is obtained w,b (x) Comprises the following steps:
where the function f is called the activation function and W is the weight matrix.
In some embodiments, the training data may enter the signal encoding unit from the input layer for encoding, the residual shrinking unit processes the noise feature to obtain a feature signal, and the signal decoding unit decodes the feature signal to generate a reconstructed sample. The deep learning noise reduction model can determine parameters such as weight and bias in the signal encoding unit and the signal decoding unit according to reconstruction errors of the reconstructed sample and the target sample (FBG reflected signals which do not contain preset noise in training data).
Due to the fact that the overfitting problem may occur during training of the deep learning model, the effect of the trained model on test data is poor, in the process of constructing the training set, the noise of sample data is guaranteed to be complex enough, all possible noise situations are contained as much as possible, and the denoising effect of the model is improved.
For example, multiple FBG reflection signals with good signal-to-noise ratio can be acquired through an experimental system, and at least one of device interference noise with random intensity, white gaussian noise, alpha stably distributed noise and non-stationary cable wind-dance noise is added to the FBG reflection signals, so that random noise, undefined noise interference of energy magnitude and action range, which may be encountered when the monitoring system is used for outdoor high-altitude real-time cable monitoring in an actual use scene, can be simulated as much as possible, and 51000 FBG reflection signals (hereinafter referred to as train _ x signals) containing noise and 51000 clean FBG reflection signals (hereinafter referred to as train _ y) corresponding to the FBG reflection signals containing noise under a simulated real environment are generated. 51000 train _ x signals and 51000 train _ y signals serve as training sets. 20000 strips of FBG reflection signals containing noise (hereinafter referred to as "differentiation _ x signals") are generated by the same processing method, and 20000 strips of clean FBG reflection signals (hereinafter referred to as "differentiation _ y signals") corresponding to these FBG reflection signals containing noise are generated. 20000 valid _ x signals and 20000 valid _ y signals are used as verification sets. In the model training process, a train _ x signal is used as the input of the deep learning noise reduction model to obtain a reconstructed signal. And performing error calculation on the reconstructed signal and the corresponding train _ y signal, and judging whether the loss meets the index of less than or equal to a preset error.
If the model parameter does not meet the index, the model is continuously trained by using the next train _ x on the basis of the current model parameter, the parameters of the model are repeatedly trained, judged and updated in such a way, finally, the deep learning noise reduction model meets the goal of minimizing the loss function, and the determination of the model parameter is completed. When the test of the deep learning noise reduction model which is trained is passed, the FBG transmitting signal which is acquired by the long-distance large-capacity fiber bragg grating sensing system can be used as the input of the model, and the FBG reflecting signal after noise reduction can be obtained.
If the index is met, the parameters of the model are reserved, the input of the deep learning noise reduction model is switched to an identification _ x signal, and model testing is carried out. 20000 validity _ x signals are used as the input of the deep learning noise reduction model, a corresponding number of reconstruction signals can be obtained, based on the obtained reconstruction signals and the corresponding 20000 validity _ y signals, whether the noise reduction performance of the deep learning noise reduction model meets the requirements or not can be evaluated, if the noise reduction performance meets the requirements, the training of the model can be ended, if the noise reduction performance does not meet the requirements, the parameters of the model can be updated, and the model is trained by using the data of a training set.
The following is a detailed explanation of related concepts and calculation processes related to the deep learning noise reduction model according to the embodiments of the present application.
In some embodiments, the expression for a convolutional layer may be:
z＝f(x)＝S f (wx+b z )；
wherein, x = { x 1 ,x 2 ,x 3 ,…,x N Is the signal input to the convolutional layer, z = { z = 1 ,z 2 ,z 3 ,…,z N Is a characteristic signal obtained by processing the convolution layer, S f To activate a function, b z W is the weight matrix for the offset vector in the encoding process.
In some embodiments, the expression for batch normalization in a residual shrinkage network may be:
wherein x is n And y n Respectively representing the input and output characteristics of the nth sample in a small batch, gamma and beta represent two trainable parameters of the scaling and translation profile, respectively, and epsilon is a positive number close to zero.
In some embodiments, the expression of the rectified linear cell activation function in the residual shrinkage network may be:
y = max (x, 0), where x and y represent the input and output characteristics of the activation function, respectively.
In some embodiments, global mean pooling in the residual shrinkage network may reduce the number of weights for the fully connected output layers, thereby reducing the risk of network overfitting, and also making the network less susceptible to noise-induced FBG signal center wavelength shifts when learning features. The expression for global mean pooling may be:
In some embodiments, the working principle of soft thresholding in a residual shrinkage network can be described as:
Soft thresholding can directly set features close to zero, leaving negative, useful features. Inputting the feature signal after the global mean pooling into a two-layer full-connection network to obtain a scaling parameter z, regulating the scaling parameter to be between 0 and 1 through a Sigmoid function to obtain a new scaling parameter alpha, and multiplying the alpha by the mean value of the absolute value of the feature signal to obtain a threshold tau. The expression of the scaling parameter α and the threshold τ may be:
in some embodiments, the expression for the deconvolution layer may be:
y＝g(z)＝S g (w’z+b y )；
wherein y = { y = 1 ,y 2 ,y 3 ,…,y N Is a reconstructed signal obtained by deconvolution layer processing, S g To activate a function, b y Is a decoding processW' is a weight matrix.
In some embodiments, the deep learning noise reduction model may use a LeakyReLU function as an activation function of the signal encoding unit and the signal decoding unit, and the LeakyReLU function may be expressed as:
in some embodiments, in order to minimize the error between the reconstructed signal obtained after model decoding and the target signal (train _ y signal), a deep learning noise reduction model satisfying the requirement is obtained. A mean square error loss function (MSE) may be used as the loss function,
Let an input signal be
Then the output signal is>
Wherein, w * Is a model parameter, w, required for learning from the network * ＝{w *1 ,w *2 ,w *3 ,…,w *N },d * Representing the reconstructed signal obtained after the encoding and decoding operations, d * ＝{d *1 ,d *2 ,d *3 ,…,d *N And then the loss function can be found as:
in order to minimize the loss, the function optimization objective is,
In some embodiments, the gradient descent method may be such that the desired model parameters are obtained with a minimum number of back propagation, and the resulting weight values and bias are the result of adjusting the error in the gradient direction. w is a * Is a one-dimensional vector containing all parameters, and can initialize a w * Above this value, the gradient descent method is used to find the value of the next group. The value of the loss function decreases due to the decreasing gradient. When iterating to a certain extent, w * Tends to be stable, at which time w * I.e. the value to be found.
The iteration function is:
At each iteration, the current w may be used *j Solve the value to the right of the equation and cover the current w *j And obtaining a new value after iteration.
According to 2a-2b, the deep learning noise reduction model mainly comprises three parts of coding, noise characteristic processing and decoding, the model overall presents a symmetrical structure, an input signal is coded in a signal coding unit, the noise characteristic is processed by a residual shrinkage unit, and the signal decoding unit decodes the input signal, so that low-dimensional characteristic information is restored into a clean signal.
In some embodiments, in the process of constructing the training set, in order to enable the model to learn the signal characteristics interfered by noise with random types and uncertain energy size and action range as comprehensively as possible, the long-distance large-capacity fiber grating sensing system can be used for collecting 500 laboratory environment FBG reflected signals with better quality and capable of being regarded as pure signals at the sampling rate of 4kHz, random-size device interference noise, white gaussian noise and two types of non-gaussian distributed noise including alpha stable distributed noise and non-stable cable wind-dancing noise are randomly added to the laboratory environment FBG reflected signals, noise interference with random types, uncertain energy size and action range, which is possibly encountered when the system is used for outdoor high-altitude real-time cable monitoring, is simulated as much as possible, 51000 FBG reflected signals under a simulated real environment are formed to be used in the training set, and 20000 FBG reflected signals are generated by the same process to be used in the verification set.
Wherein, the expression I of the interference noise of the device n (λ) may be:
in the above expression I n In (λ), I 0 Lambda is the wavelength satisfying the Bragg condition, n is the refractive index of the medium, I (lambda) is the light intensity reflected by the fiber grating, nL is the optical path difference of the light wave, a 2 In order to be the intensity of the noise,
the phase change of the reflection point caused by the external environment. />
Characteristic function of alpha stable distribution noise
Satisfies the following conditions:
in the above characteristic function
Where i is complex imaginary unit, t is signal time unit, alpha is characteristic index, beta is symmetric parameter, gamma is dispersion coefficient, delta is position parameter, 0<α≤2,-1≤β≤1，γ>0，-∞<δ<∞；
Expression n of non-stationary cable wind-dance noise wind Comprises the following steps:
n wind ＝[1+acos(2πft)]n g (t)+n 0 (t)；
in the above formula, n g (t) represents white Gaussian noise, n 0 (t) represents ambient noise, and a is the non-stationary modulation amplitude of the signal.
In some embodiments, after the trained deep learning noise reduction model is obtained, the FBG reflected signal acquired by the long-distance large-capacity fiber grating sensing system may be used as an input of the model, so as to obtain the noise-reduced FBG reflected signal. As shown in fig. 5a to 5b, fig. 5a shows the FBG reflected signal which is acquired and has poor signal quality and needs to be denoised, and fig. 5b shows the FBG reflected signal which is obtained by using the deep learning denoising model and is denoised. As can be seen from fig. 5b, the FBG reflection signal obtained by the deep learning noise reduction model processing is clean and smooth, and 44 FBG wavelength information of the 1535nm-1565nm wavelength band is completely recovered, especially the weak FBG reflection signal at the edge is substantially invisible and is obviously seen, and the noise reduction effect is significant.
The long-distance large-capacity fiber grating sensing system uses the weak-reflectivity fiber grating sensor to increase the multiplexing number, and combines the optical amplification technology to prolong the sensing distance, but due to accumulation of long-distance fiber transmission loss and weak reflectivity of the fiber grating, the received sensing signal is quite weak and is easily influenced by external environment and system internal noise, and the sensing signal is possibly buried in the noise. The deep learning noise reduction model is obtained based on deep learning algorithm training, the model can comprehensively learn the characteristics of a noise-containing optical fiber grating sensing signal, the capacity of unsupervised denoising of the optical fiber grating sensing signal is achieved, the deep learning noise reduction model uses a residual shrinkage unit to process the noise characteristics, an input signal is reconstructed to obtain a denoised signal, the model training adopts a noise interference signal with random type, uncertain energy and action range as a training sample, in continuous iterative optimization, model parameters are updated to minimize reconstruction errors, so that the model output signal is close to a target signal, the signal quality is obviously improved, the noise reduction model can reconstruct FBG reflection signals interfered in different degrees through a large amount of sample training, the noise reduction model has the advantage of obvious noise reduction under the condition of complex system environment, the model robustness is high, the noise reduction problem of a weak signal of a long-distance large-capacity optical fiber grating sensing system can be solved, and a high-quality spectrum is provided for high-precision measurement of multiple parameters such as temperature and strain.
FIG. 6 is a diagram of an electronic device according to a preferred embodiment of the present application.
The electronic device 100 comprises a memory 20, a processor 30 and a computer program 40 stored in the memory 20 and executable on the processor 30. The processor 30, when executing the computer program 40, implements the steps in the above-described fiber grating sensing system noise reduction method embodiment based on deep learning, such as the steps S11 to S15 shown in fig. 1.
Illustratively, the computer program 40 may be partitioned into one or more modules/units that are stored in the memory 20 and executed by the processor 30 to accomplish the present application. The one or more modules/units may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution of the computer program 40 in the electronic device 100.
The electronic device 100 may be a desktop computer, an optical communication device, a server, or like computing device. Those skilled in the art will appreciate that the schematic diagram is merely an example of the electronic device 100, does not constitute a limitation of the electronic device 100, and may include more or less components than those shown, or combine certain components, or different components, for example, the electronic device 100 may further include a signal acquisition device, a communication bus, etc.
The Processor 30 may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components, etc. The general purpose processor may be a microprocessor or the processor 30 may be any conventional processor or the like, the processor 30 being the control center for the electronic device 100 and the various interfaces and lines connecting the various parts of the overall electronic device 100.
The memory 20 may be used to store the computer program 40 and/or the module/unit, and the processor 30 may implement various functions of the electronic device 100 by running or executing the computer program and/or the module/unit stored in the memory 20 and calling data stored in the memory 20. The memory 20 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function, and the like; the storage data area may store data created according to the use of the electronic apparatus 100, and the like. In addition, the memory 20 may include high speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), at least one magnetic disk storage device, a Flash memory device, or other non-volatile solid state storage device.
The integrated modules/units of the electronic device 100 may be stored in a computer-readable storage medium if implemented in the form of software functional units and sold or used as separate products. Based on such understanding, all or part of the processes in the methods of the embodiments described above may be implemented by a computer program, which may be stored in a computer-readable storage medium and used by a processor to implement the steps of the embodiments of the methods described above. Wherein the computer program comprises computer program code, which may be in the form of source code, object code, an executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U.S. disk, removable hard disk, magnetic diskette, optical disk, computer Memory, read-Only Memory (ROM), random Access Memory (RAM), electrical carrier wave signal, telecommunications signal, software distribution medium, etc. It should be noted that the computer readable medium may contain other components which may be suitably increased or decreased as required by legislation and patent practice in jurisdictions, for example, in some jurisdictions, in accordance with legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunications signals.
In the several embodiments provided in the present application, it should be understood that the disclosed electronic device and method may be implemented in other ways. For example, the above-described embodiments of the electronic device are merely illustrative, and for example, the division of the units is only one logical function division, and there may be other division ways in actual implementation.
In addition, functional units in the embodiments of the present application may be integrated into the same processing unit, or each unit may exist alone physically, or two or more units are integrated into the same unit. The integrated unit can be realized in a form of hardware, or in a form of hardware plus a software functional module.
It will be evident to those skilled in the art that the application is not limited to the details of the foregoing illustrative embodiments, and that the present application may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive. Furthermore, it will be obvious that the term "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. The terms first, second, etc. are used to denote names, but not any particular order.
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present application and not for limiting, and although the present application is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions can be made on the technical solutions of the present application without departing from the spirit and scope of the technical solutions of the present application.
Claims (10)
1. A fiber grating sensing system noise reduction method based on deep learning is characterized by comprising the following steps:
collecting a plurality of FBG (fiber Bragg Grating) reflected signal samples, and preprocessing the plurality of FBG reflected signal samples to obtain a plurality of first FBG reflected signals;
adding preset noise to the plurality of first FBG reflection signals to obtain a plurality of groups of second FBG reflection signals, wherein the preset noise comprises at least one of device interference noise with random intensity, gaussian white noise, alpha stable distribution noise and non-stable cable wind dance noise, and each group of second FBG reflection signals in the plurality of groups of second FBG reflection signals comprises an FBG reflection signal not containing the preset noise and an FBG reflection signal containing the preset noise;
dividing the plurality of groups of second FBG reflected signals into a training set and a verification set;
training a deep learning noise reduction model based on the training set and the verification set until determining model parameters of the deep learning noise reduction model;
and carrying out noise reduction treatment on the FBG reflected signals acquired by the fiber bragg grating sensing system based on the trained deep learning noise reduction model.
2. The method for noise reduction of the deep learning based fiber grating sensing system according to claim 1, wherein the preprocessing the plurality of FBG reflected signal samples to obtain a plurality of first FBG reflected signals comprises:
converting the binary data file of the FBG reflected signal samples into a decimal data file, and converting data in the decimal data file into a two-dimensional array by taking the effective sampling points of the spectrum as the column number;
and visually displaying the two-dimensional array according to the rows to select the FBG reflected signal with the spectral signal-to-noise ratio larger than a preset threshold value as the first FBG reflected signal.
3. The deep learning-based fiber grating sensing system noise reduction method according to claim 2, wherein the number of effective sampling points of the spectrum is 512, and the preset threshold is 4dB.
4. The method according to any one of claims 1 to 3, wherein the deep learning noise reduction model comprises an input layer, a signal encoding unit, a residual shrinking unit, a signal decoding unit and an output layer, and the training the deep learning noise reduction model based on the training set and the verification set comprises:
using a set of training data in the training set as input to the input layer;
encoding, at the signal encoding unit, the FBG reflected signal containing the preset noise in the set of training data, performing soft thresholding on the encoding result by the residual error puncturing unit to obtain a noise characteristic, decoding the processing result by the residual error puncturing unit by the signal decoding unit, and outputting a reconstructed sample by the output layer;
reversely propagating the reconstruction error of the FBG reflected signal which does not contain the preset noise in the reconstruction sample and the set of training data, and judging whether the loss is greater than the preset error;
if the loss is larger than the preset error, taking the next group of training data in the training set as the input of the input layer to continue training the deep learning noise reduction model;
if the loss is smaller than or equal to the preset error, parameters of the deep learning noise reduction model are reserved, and the deep learning noise reduction model is tested by using the verification set.
5. The deep learning-based fiber grating sensing system noise reduction method according to claim 4, wherein the expression I of the device interference noise n (λ) is:
wherein, in the expression I n In (λ), I 0 Lambda is the wavelength satisfying the Bragg condition, n is the refractive index of the medium, I (lambda) is the light intensity reflected by the fiber grating, nL is the optical path difference of the light wave, a 2 In order to be the intensity of the noise,
phase change of the reflection point caused by external environment;
characteristic function of alpha stable distribution noise
Satisfies the following conditions:
wherein, in the characteristic function
Where i is complex imaginary unit, t is signal time unit, alpha is characteristic index, beta is symmetric parameter, gamma is dispersion coefficient, delta is position parameter, 0<α≤2,-1≤β≤1，γ>0，-∞<δ<∞；
Expression n of non-stationary cable wind-dance noise wind Comprises the following steps:
n wind ＝[1+acos(2πft)]n g (t)+n 0 (t)；
wherein, in the expression n wind In, n g (t) is white Gaussian noise, n 0 (t) represents ambient noise, and a is the non-stationary modulation amplitude of the signal.
6. The deep learning-based fiber grating sensing system noise reduction method according to claim 4, wherein the signal encoding unit includes two convolutional layers, and the operating principle of each convolutional layer is described as follows:
z＝f(x)＝S f (wx+b z )；
wherein, x = { x 1 ,x 2 ,x 3 ,…,x N Is the signal input to the convolutional layer, z = { z = 1 ,z 2 ,z 3 ,…,z N Is a characteristic signal obtained by processing the convolution layer, S f In order to activate the function(s),
b z is an offset vector in the encoding process, and w is a weight matrix;
the signal decoding unit comprises two deconvolution layers, and the working principle of each deconvolution layer is described as follows:
y＝g(z)＝S g (w’z+b y )；
wherein y = { y = 1 ,y 2 ,y 3 ,…,y N Is a reconstructed signal obtained by deconvolution layer processing, S g In order to activate the function(s),
b y w' is a weight matrix for the offset vector in the decoding process.
7. The deep learning-based fiber grating sensing system noise reduction method according to claim 4, wherein the residual shrinking unit comprises three residual shrinking networks, and the working principle of each residual shrinking network is described as follows:
wherein x is input characteristics, y is output characteristics, tau is a threshold value, and in the residual shrinkage network, the characteristic signals after global mean pooling are input into a two-layer fully-connected network to obtain scaling parametersAnd z, regulating the Sigmoid function to be between 0 and 1 to obtain a new scaling parameter alpha, and multiplying the scaling parameter alpha by the average value of the absolute values of the characteristic signals to obtain a threshold value tau, wherein the expression of the scaling parameter alpha is as follows:
the expression for the threshold τ is: />
i. j, c represent the width, length and channel number of the characteristic signal x, respectively. />
8. The deep learning-based fiber grating sensing system noise reduction method according to claim 4, wherein the expression of the Loss function Loss of the reconstruction error is as follows:
wherein y' = { y = 1 ’,y 2 ’,y 3 ’,…,y N ' } is the expected output value, and the penalty function optimization objective is:
λ is the fixed parameter of the model, N is the number of signal sampling points, x i For the ith component of the input convolutional layer signal, the objective w is optimized using a gradient descent method to solve the function * The iteration function is:
α is the learning rate.
9. A computer readable storage medium storing computer instructions which, when executed on an electronic device, cause the electronic device to perform the method for noise reduction for a deep learning-based fiber grating sensing system according to any one of claims 1 to 8.
10. An electronic device comprising a processor and a memory, wherein the memory is configured to store instructions, and wherein the processor is configured to invoke the instructions in the memory, so that the electronic device executes the method for noise reduction for a deep learning-based fiber bragg grating sensing system of any one of claim 1 to claim 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111556911.9A CN114257313B (en) | 2021-12-18 | 2021-12-18 | Fiber grating sensing system noise reduction method based on deep learning and related equipment thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111556911.9A CN114257313B (en) | 2021-12-18 | 2021-12-18 | Fiber grating sensing system noise reduction method based on deep learning and related equipment thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114257313A CN114257313A (en) | 2022-03-29 |
| CN114257313B true CN114257313B (en) | 2023-04-07 |
Family
ID=80795758
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111556911.9A Active CN114257313B (en) | 2021-12-18 | 2021-12-18 | Fiber grating sensing system noise reduction method based on deep learning and related equipment thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114257313B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115014404A (en) * | 2022-05-09 | 2022-09-06 | 武汉理工大学 | High-precision high-speed fiber grating demodulation method based on deep learning |
| CN116861167B (en) * | 2023-06-12 | 2023-12-15 | 河北工程大学 | FBG spectrum cyclic denoising method based on deep learning |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111951505A (en) * | 2020-08-25 | 2020-11-17 | 青岛大学 | Fence vibration intrusion positioning and mode identification method based on distributed optical fiber system |
| CN113654642A (en) * | 2021-08-23 | 2021-11-16 | 之江实验室 | Distributed acoustic wave sensing noise reduction system and method based on reference sensor |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB201801825D0 (en) * | 2018-02-05 | 2018-03-21 | Univ Of Sussex | Control methods for optical systems |
-
2021
- 2021-12-18 CN CN202111556911.9A patent/CN114257313B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111951505A (en) * | 2020-08-25 | 2020-11-17 | 青岛大学 | Fence vibration intrusion positioning and mode identification method based on distributed optical fiber system |
| CN113654642A (en) * | 2021-08-23 | 2021-11-16 | 之江实验室 | Distributed acoustic wave sensing noise reduction system and method based on reference sensor |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114257313A (en) | 2022-03-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114257313B (en) | Fiber grating sensing system noise reduction method based on deep learning and related equipment thereof | |
| CN107730474B (en) | Image processing method, processing device and processing equipment | |
| CN112288193A (en) | Ocean station surface salinity prediction method based on GRU deep learning of attention mechanism | |
| CN113344295B (en) | Method, system and medium for predicting residual life of equipment based on industrial big data | |
| CN110705600A (en) | Cross-correlation entropy based multi-depth learning model fusion method, terminal device and readable storage medium | |
| CN115409252A (en) | Plant growth data multi-step prediction method, device, medium and equipment based on LSTM | |
| CN113988357A (en) | High-rise building wind-induced response prediction method and device based on deep learning | |
| CN117079005A (en) | Optical cable fault monitoring method, system, device and readable storage medium | |
| Bahri et al. | Time series forecasting using smoothing ensemble empirical mode decomposition and machine learning techniques | |
| CN116205863A (en) | Method for detecting hyperspectral image abnormal target | |
| CN115859048A (en) | Noise processing method and device for partial discharge signal | |
| CN115169747A (en) | Method and device for predicting non-stationary time sequence of power load and related equipment | |
| CN114187484A (en) | Sensor induction image recognition method, system, device and storage medium | |
| CN114187194A (en) | Sensor induction image noise reduction processing method, system and device and storage medium | |
| CN113657595A (en) | Neural network real-time pruning method and system and neural network accelerator | |
| CN113609445A (en) | Multi-source heterogeneous monitoring data processing method, terminal device and readable storage medium | |
| CN116879192B (en) | Water bloom prediction method, device, equipment and medium based on satellite remote sensing data | |
| CN117392564B (en) | River water quality inversion method based on deep learning, electronic equipment and storage medium | |
| CN117708507B (en) | Efficient alpha and beta ray identification and classification method based on artificial intelligence | |
| CN114997228B (en) | Action detection method and device based on artificial intelligence, computer equipment and medium | |
| CN117648537B (en) | Atmospheric pollution real-time monitoring method and system based on hyperspectral technology | |
| CN114745187B (en) | Internal network anomaly detection method and system based on POP flow matrix | |
| CN107506779A (en) | A kind of evaluation method and system of plant stem water content | |
| Rezvov et al. | Improvement of the AI-Based Estimation of Significant Wave Height Based on Preliminary Training on Synthetic X-Band Radar Sea Clutter Images | |
| CN117909384A (en) | Multivariable time sequence prediction method and system based on wavelet denoising and multiscale feature extraction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |