CN106503642A - A kind of model of vibration method for building up for being applied to optical fiber sensing system - Google Patents
A kind of model of vibration method for building up for being applied to optical fiber sensing system Download PDFInfo
- Publication number
- CN106503642A CN106503642A CN201610907115.8A CN201610907115A CN106503642A CN 106503642 A CN106503642 A CN 106503642A CN 201610907115 A CN201610907115 A CN 201610907115A CN 106503642 A CN106503642 A CN 106503642A
- Authority
- CN
- China
- Prior art keywords
- optical fiber
- signal
- vibration
- fiber
- model
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention provides a kind of method for building up of the model of vibration for being applied to optical fiber sensing system, comprises the following steps：Vibration signal is adjusted to by optical fiber interference signal output by fiber-optic vibration device, the time for producing signal, the distance for producing signal are combined using the signal intensity of output, signal period, signal condition figure is set up as the first coordinate axess, time for the second coordinate axess with distance of the fiber position with optical fiber starting point；Signal condition figure under the different Vibration Conditions of collection, label is manually marked to these data, the data message of manually mark label and signal condition figure is input to convolutional neural networks model carries out image recognition and CNN training, obtains the model of vibration for being applied to optical fiber sensing system；The image recognition of the convolutional neural networks model and CNN training include：First stage, forward propagation stage；Second stage, back-propagation stage.Technical scheme has the characteristics of discrimination is high, identification accuracy is good, identification ability is strong.
Description
Technical field
The invention belongs to technical field of optical fiber sensing, more particularly to a kind of model of vibration for being applied to optical fiber sensing system
Method for building up.
Background technology
With the fast development of society, important is more and more shown to the long-range monitoring and protection of some important areas, such as
Airport, petroleum pipeline, military base, nuclear power plant, prison, bank etc..It is heavy that fiber-optic vibration safety pre-warning system can gather these
The various vibration signals of area peripheral edge are wanted, by analyzing periphery vibration signal characteristics, vibration source type is drawn, if detecting to region
Harmful vibration source occurs, and can carry out early warning in time, and reports the particular location of hazardous events, reaches to important area or region
The real-time guard of periphery, the purpose for reducing property loss.Optical fiber is very sensitive to the vibration signal from the external world, if not to adopting
The vibration signal for collecting is processed, and optical fiber can be caused to carry out indiscriminate alarm to different classes of vibration event, bring compared with
High false alarm rate, reduces the practicality of optical fiber safety early-warning system, and vibration signal is identified being a kind of feasible process
Means.At present, the research about vibration signal processing and identification mainly has the blind recognition skill of blind source separate technology, vibration signal
Art, wavelet analysises and conversion, linear classifier support vector machine etc..It is applied to optical fiber early warning identifying system mainly to use
DTW algorithms.Due to the continuous development of system, recognition result is required constantly to be lifted, DTW algorithms can not meet system need
Ask, its exist subject matter be following some：
1st, current recognizer discrimination is relatively low, it is impossible to effectively recognized.
2nd, under the big environment of noise, discrimination is remarkably decreased, it is impossible to accurately recognized.
3rd, fiber distance is more remote, and the oscillation intensity for collecting data is lower, and recognition result is remarkably decreased.
4th, model is fairly simple, it is impossible to process more complicated identification problem.
Content of the invention
For above technical problem, the invention discloses a kind of foundation side of the model of vibration for being applied to optical fiber sensing system
Method, carries out attribute character analysis according to the different vibration signal that fiber-optic vibration safety pre-warning system is collected, and sets up corresponding
Characteristic model, by neural network recognization algorithm, Recognition of Vibration Sources is carried out to the vibration signal for collecting, identification is substantially increased
Accuracy, with higher discrimination.
In this regard, the technical scheme is that：
A kind of method for building up of the model of vibration for being applied to optical fiber sensing system, comprises the following steps：
Step S1：Vibration signal is adjusted to by optical fiber interference signal output by fiber-optic vibration device, using the signal of output
Intensity, signal period combine the time for producing signal, the distance for producing signal, with the distance of fiber position and optical fiber starting point as the
One coordinate axess, time set up signal condition figure for the second coordinate axess；Wherein, if the signal condition figure is used for characterizing the optical fiber
Information of the dry fiber position in several temporal signal values；
Step S2：These data are manually marked label, by people by the signal condition figure under the different Vibration Conditions of collection
The data message of work mark label and signal condition figure is input to convolutional neural networks model and carries out image recognition and CNN training,
Obtain the model of vibration for being applied to optical fiber sensing system；
Wherein, the image recognition of the convolutional neural networks model and CNN training are comprised the following steps：
First stage, propagation stage forward, detailed process are as follows：
Discuss that multi-class problem, common c classes are total to N number of training sample using square error cost function, error term is：
In formula (1),The vector of the kth dimension of the corresponding label of n-th sample is represented,Represent that n-th sample is corresponding
K-th output of network output；Total classes of the c for representative sample；For multi-class problem, output is typically organized as " one-of-c "
Form, that is, the output node for there was only the corresponding class of the input is just output as, the position of other classes or node are 0 or negative
Number, depending on the activation primitive of your output layer；It is exactly -1 that sigmoid is exactly 0, tanh；
It is the summation of the error of each training sample based on the error in whole training sets, for n-th sample
Error, is expressed as：
Then according to BP rules calculation cost function E with regard to each weights of network partial derivative；Represent current with l
Layer, the output of current layer can be expressed as：
Xl=f (ul),with,ul=Wlxl-1+bl(3)
In formula, X represents output, and u represents input, and w represents weights, and b represents biasing, and wherein which of network l represent
Layer；
Second stage, in the back-propagation stage, detailed process is as follows：
The sensitivity of the base of the error that back propagation is returned, i.e. each neuron is：
The sensitivity of back propagation l layers is：
δl=(Wl+1)Tδl+1οf'(ul) (5)
In formula (5), T represents matrix transpose, and W represents weights, and u represents input, is neuron signal in the technical program
Input, " ο " represents each element multiplication；
The sensitivity of the neuron of output layer is：
δL=f ' (uL)ο(yn-tn). (6)
Finally, right value update is carried out with delta rules to each neuron；Stated with the form of vector, for l
Layer, error for the input that the derivative that this layer each weights are combined as matrix is this layer with the sensitivity of this layer is：
In formula (7), wherein l represents the number of plies of neutral net, and it is vector which layer X representatives is；W is the weights of l layers,
η is learning rate, and δ represents sensitivity；For each weights has a specific learning rate.
Wherein, the convolutional neural networks model includes convolutional layer and pond layer, and wherein, C layers are characterized extract layer, each
The input of neuron is connected with the local receptor field of preceding layer, and extracts the feature of the local, once the local feature is extracted
Afterwards, it and the position relationship between other features are also decided therewith；S layers are Feature Mapping layers, each computation layer of network by
Multiple Feature Mapping compositions, each Feature Mapping is a plane, and in plane, the weights of all neurons are equal.Feature Mapping is tied
Activation primitive of the sigmoid functions that structure adopts influence function core little as convolutional network so that Feature Mapping has displacement not
Degeneration.
Input picture by and three trainable wave filter and can biasing put and carry out convolution, produce three in C1 layers after convolution
Individual Feature Mapping figure, then in Feature Mapping figure, per group of four pixels are sued for peace again, and weighted value, biasing are put, by one
Sigmoid functions obtain the Feature Mapping figure of three S2 layers.These mapping graphs entered filtering again and obtained C3 layers.This hierarchical structure
S4 is produced as S2 again.Finally, these pixel values are rasterized, and connect into a vector and be input to traditional nerve net
Network, is exported.
Further, since the neuron on a mapping face shares weights, thus the number of network freedom parameter is reduced, dropped
The complexity that low network parameter is selected.Followed by one use of each feature extraction layer (C- layers) in convolutional neural networks
To seek the computation layer (S- layers) of local average and second extraction, this distinctive structure of feature extraction twice makes network in identification
There is higher distortion tolerance to input sample.
There is convolutional neural networks model powerful learning capacity and efficient feature representation ability, more important advantage to be
Information is successively extracted from Pixel-level initial data to abstract semantic concept, this causes it extracting the global characteristics of image and upper
There is in terms of context information prominent advantage.
This technical scheme, using based on convolutional neural networks (Convolutional Neural Networks, CNN)
Carrying out model foundation, the recognizer with higher efficiency improves discrimination to image recognition algorithm.First, shaken using optical fiber
The effective informations such as dynamic signal, fiber-optic vibration intensity, distance, time, generate fiber-optic signal state diagram, preferably thermodynamic chart, finally
By the image recognition algorithm based on CNN, the training of model is carried out to the picture of substantial amounts of thermodynamic chart, obtain being applied to optical fiber biography
The model of vibration of sensing system.New signal condition figure is identified classifying using the model that obtains of training, preferably thermodynamic chart is just
Oscillatory type can be obtained.The technical program proposes the optical fiber early warning identification based on optical fiber vibration sensing signal thermodynamic chart and calculates
Method, carries out attribute character analysis according to the different vibration signal that fiber-optic vibration safety pre-warning system is collected, and sets up corresponding
Vibration performance model, by neural network recognization algorithm, Recognition of Vibration Sources is carried out to the vibration signal for collecting, is substantially increased
The accuracy of identification.
As a further improvement on the present invention, the distance with fiber position and optical fiber starting point as the first coordinate axess, when
Between set up signal condition figure for the second coordinate axess, including：
If it the first coordinate axess, time is that the second coordinate axess, fiber position exist with the distance of fiber position and optical fiber starting point to be
A dry temporal signal value sets up signal condition figure for three axes.
As a further improvement on the present invention, the signal condition figure is thermodynamic chart, wherein, pixel in the thermodynamic chart
Be worth the information for the optical fiber several fiber positions being characterized in several temporal signal values.
As a further improvement on the present invention, to thermodynamic chart by color from shallow represent to deep shockproofness by little to
Greatly.
The invention also discloses a kind of determination method of the model of vibration of optical fiber sensing system, comprises the following steps：
Step A：The reception optical fiber signal of telecommunication, in the acquisition fiber optic telecommunications number, several fiber positions corresponding are at several
Between signal value；Wherein, the fiber optic telecommunications number are converted to by the optical signal of fiber reflection；
Step B：Thermodynamic chart is set up as permanent coordinate axess, time for axis of ordinates with distance of the fiber position with optical fiber starting point；
Pixel value in the thermodynamic chart is used for characterizing the letter of the optical fiber several fiber positions in several temporal signal values
Breath；
Step C：Pretreatment is carried out to thermodynamic chart, then according to claim 1 obtain described be applied to Fibre Optical Sensor system
The model of vibration of system carries out control identification, obtains current model of vibration classification.Wherein, the thermodynamic chart to producing carries out pre- place
Reason, filters out a part of noise, is conducive to recognizing.
As a further improvement on the present invention, in step A, fiber-optic vibration threshold values is set, if fiber optic telecommunications number exceed light
Fine vibration threshold values, then do not carry out subsequent step.
As a further improvement on the present invention, the fiber position and the optical fiber starting point are calculated using L=(c*T/n)/2
The distance between；Wherein, c is light speed in a vacuum, and T is that time of the fiber position in the fiber optic telecommunications number is long
Degree, n is the refractive index of the optical fiber.
Compared with prior art, beneficial effects of the present invention are：
Using technical scheme, by using the suitable convolutional neural networks mould of convolutional neural networks algorithms selection
Type, is identified to the signal condition figure of the fiber-optic vibration signal generation for obtaining, and sets up model of vibration, and is produced using historical data
Raw thermodynamic chart is trained to model, is adjusted ginseng, and new thermodynamic chart is identified with the model for training, compared to biography
The DTW algorithms of system increase significantly on discrimination, and identification accuracy is good, and identification ability is strong and the suitability is wide.Experimental result
Show, the method can be very good identification vibration, fully meet safety-security area and demand is recognized to intrusion behavior.
Description of the drawings
Fig. 1 is a kind of flow chart of the method for building up of the model of vibration for being applied to optical fiber sensing system of the present invention.
Fig. 2 is the structural representation of convolutional neural networks model of the present invention.
Specific embodiment
Below in conjunction with the accompanying drawings, the preferably embodiment of the present invention is described in further detail.
A kind of method for building up of the model of vibration for being applied to optical fiber sensing system, comprises the following steps：
Step S1：Vibration signal is adjusted to optical fiber interference signal by fiber-optic vibration device to export, the reception optical fiber signal of telecommunication,
If being not above vibration threshold values set in advance, using output signal intensity, the signal period combine produce signal when
Between, produce the distance of signal, with the distance of fiber position and optical fiber starting point as axis of abscissas, the time set up heating power for axis of ordinates
Figure；Pixel value in the thermodynamic chart is used for characterizing the optical fiber several fiber positions in several temporal signal values
To deep, information, thermodynamic chart represent that shockproofness is ascending from shallow by color.
Step S2：These data are manually marked label, manually will be marked by the thermodynamic chart under the different Vibration Conditions of collection
The data message of note label and signal condition figure is input to convolutional neural networks model and carries out image recognition and CNN training, obtains
The model of vibration for being applied to optical fiber sensing system.Wherein, the label of the artificial mark is preferably hand digging, machinery and digs
Pick, animal are passed through, machinery is passed through.
Wherein, the image recognition of the convolutional neural networks model and CNN training are comprised the following steps：
First stage, propagation stage forward, detailed process are as follows：
Discuss that multi-class problem, common c classes are total to N number of training sample using square error cost function, error term is：
In formula (1),The vector of the kth dimension of the corresponding label of n-th sample is represented,Represent that n-th sample is corresponding
K-th output of network output；Total classes of the c for representative sample；For multi-class problem, output is typically organized as " one-of-c "
Form, that is, the output node for there was only the corresponding class of the input is just output as, the position of other classes or node are 0 or negative
Number, depending on the activation primitive of your output layer；It is exactly -1 that sigmoid is exactly 0, tanh；
Because the error in whole training sets is the summation of the error of each training sample, we first examine here
Consider for the BP of a sample.For the error of n-th sample, it is expressed as：
Then according to BP rules calculation cost function E with regard to each weights of network partial derivative；Represent current with l
Layer, the output of current layer can be expressed as：
Xl=f (ul),with,ul=Wlxl-1+bl(3)
In formula (3), X represents output, and u represents input, and w represents weights, and b represents biasing, and wherein which of network l represent
Layer；
Output activation primitive f (.) can have many kinds, usually sigmoid functions or hyperbolic tangent function.
Sigmoid is by output squeezing to [0,1], so last output meansigma methodss typically tend to 0.If so training number by us
It is 1 according to zero-mean and variance is normalized to, convergence can be increased during gradient declines.For normalized data set
For, hyperbolic tangent function is also good selection.
Second stage, in the back-propagation stage, detailed process is as follows：
The error that back propagation is returned can regard the sensitivity sensitivities of the base of each neuron as, wherein
The meaning of sensitivity is exactly that our base b changes are how many, and error can change how many, that is, rate of change of the error to base, also
It is derivative, the sensitivity of the base of the error that back propagation is returned, i.e. each neuron is as shown in following formula (4).Wherein, second
Individual equal sign is obtained according to the chain rule of derivation.
BecauseSoThat is the sensitivity of bias basesWith error E to one
Individual node fully enters the derivative of uIt is equal.This derivative is exactly to allow high-rise error back propagation to come to the god of bottom
Pen.
The sensitivity of back propagation l layers is：
δl=(Wl+1)Tδl+1οf'(ul) (5)
In formula (5), T represents matrix transpose, and W represents weights, and u represents neuron signal input, and " ο " represents each element
It is multiplied；
The sensitivity of the neuron of output layer is：
δL=f ' (uL)ο(yn-tn). (6)
Finally, right value update is carried out with delta (i.e. δ) rule to each neuron.It is exactly specifically that one is given
Fixed neuron, obtains its input, is then zoomed in and out with the delta (i.e. δ) of this neuron.Form table with vector
State and be exactly, for l layers, error is the input of this layer (equal to upper for the derivative of each weights of layer (being combined as matrix)
One layer of output) multiplication cross with the sensitivity of this layer δ of each neuron (this layer be combined into a vectorial form).Then
To partial derivative be multiplied by the weights of neuron that negative learning rate is exactly this layer and have updated：
In formula (7), wherein l represents the number of plies of neutral net, and it is vector which layer X representatives is, w is the weights of l layers,
η is learning rate, δ represent before sensitivity for the more new-standard cement of bias bases is similar；In fact, for each weights
(W)ijThere is a specific learning rate ηIj.
Implement process as shown in Figure 1.In FIG, vibration signal is adjusted to by fiber optic interferometric by fiber-optic vibration device
Signal output；Using the signal intensity of output, the signal period combines the time for producing signal, and the distance generation for producing signal is obtained
Thermodynamic chart, represents shockproofness size by color from shallow to deep, and heating power diagram data under collection different situations, manually to this
A little data are labeled, and are trained to the convolutional neural networks model for designing using these data inputs, and finally giving needs
The identification model that wants.
Depth network has powerful learning capacity and efficient feature representation ability, and more important advantage is from Pixel-level
Initial data successively extracts information to abstract semantic concept, and this causes it extracting the global characteristics and contextual information of image
Aspect has prominent advantage, and wherein convolutional layer and pond layer is the important component part of convolutional neural networks.
Input picture by and three trainable wave filter and can biasing put and carry out convolution, produce three in C1 layers after convolution
Individual Feature Mapping figure, then in Feature Mapping figure, per group of four pixels are sued for peace again, and weighted value, biasing are put, by one
Sigmoid functions obtain the Feature Mapping figure of three S2 layers.These mapping graphs entered filtering again and obtained C3 layers.This hierarchical structure
S4 is produced as S2 again.Finally, these pixel values are rasterized, and connect into a vector and be input to traditional nerve net
Network, is exported, and network structure Fig. 2 is as follows.
In fig. 2, usually, C layers are characterized extract layer, the local receptor field phase of the input of each neuron and preceding layer
Connect, and extract the feature of the local, after the local feature is extracted, it and the position relationship between other features are also true therewith
Decide；S layers are Feature Mapping layers, and each computation layer of network is made up of multiple Feature Mapping, and each Feature Mapping is one
Plane, in plane, the weights of all neurons are equal.Feature Mapping structure is using the little sigmoid function conducts of influence function core
The activation primitive of convolutional network so that Feature Mapping has shift invariant.
Further, since the neuron on a mapping face shares weights, thus the number of network freedom parameter is reduced, dropped
The complexity that low network parameter is selected.Followed by one use of each feature extraction layer (C- layers) in convolutional neural networks
To seek the computation layer (S- layers) of local average and second extraction, this distinctive structure of feature extraction twice makes network in identification
There is higher distortion tolerance to input sample.
For the performance of authentication image recognizer, after the completion of image identification system design, generated using gathered data
Thermodynamic chart is tested, and sets up the model of vibration for being applied to optical fiber sensing system using said method, has vibration signal again
When, the determination method to the model of vibration of the vibration signal is comprised the following steps：
Step A：The reception optical fiber signal of telecommunication, in the acquisition fiber optic telecommunications number, several fiber positions corresponding are at several
Between signal value；Fiber-optic vibration threshold values is set, if fiber optic telecommunications number do not carry out subsequent step beyond fiber-optic vibration threshold values.
Wherein, the fiber optic telecommunications number are converted to by the optical signal of fiber reflection；
Step B：Thermodynamic chart is set up as permanent coordinate axess, time for axis of ordinates with distance of the fiber position with optical fiber starting point；
Pixel value in the thermodynamic chart is used for characterizing the letter of the optical fiber several fiber positions in several temporal signal values
Breath；Wherein, the distance between the fiber position and described optical fiber starting point are calculated using L=(c*T/n)/2；Wherein, c exists for light
Speed in vacuum, T are time span of the fiber position in the fiber optic telecommunications number, and n is the refractive index of the optical fiber.
Step C：Pretreatment is carried out to thermodynamic chart, then according to claim 1 obtain described be applied to Fibre Optical Sensor system
The model of vibration of system carries out control identification, obtains current model of vibration classification.
As our Fibre Optical Sensor is arranged on soil depths for detecting top vibration source classification, therefore we select
Big machinery excavation, hand digging, automobile are passed through, strike the typical vibration source classification of four kinds of well and the mankind or animal through this normal
The violate-action signal that sees has carried out neural network filter test.When the vibration difference that different behaviors causes, generation
Interference signal is also different, and after processing through FPGA, the frequency variation signal waveform of generation and amplitude also have significantly difference.Therefore,
We enter row mode knowledge using the convolutional neural networks image recognition algorithm of host computer to the frequency variation signal after FPGA process
Do not model, and carried out prognostic experiment, test result indicate that more than 95% has all been reached to the predictablity rate for vibrating behavior, and
Traditional DTW algorithms are compared, experimental result such as table 1：
1 experimental result relative analyses of table
The above, only presently preferred embodiments of the present invention not makees any pro forma restriction to the present invention；All
The those of ordinary skill of the industry can described in by specification and the above and swimmingly implement the present invention；But, all familiar
Professional and technical personnel is made using disclosed above technology contents in the range of without departing from technical solution of the present invention
A little change, modification with develop equivalent variations, be the present invention Equivalent embodiments；Meanwhile, all realities according to the present invention
The change of any equivalent variations that matter technology is made to above example, modification and differentiation etc., still fall within the technology of the present invention
Within the protection domain of scheme.
Claims (7)
1. a kind of method for building up of the model of vibration for being applied to optical fiber sensing system, it is characterised in that comprise the following steps：
Step S1：By fiber-optic vibration device by vibration signal be adjusted to optical fiber interference signal output, using output signal intensity,
Signal period combines the time for producing signal, the distance for producing signal, is sat with distance of the fiber position with optical fiber starting point as first
Parameter, time set up signal condition figure for the second coordinate axess；Wherein, the signal condition figure be used for characterize the optical fiber several
Information of the fiber position in several temporal signal values；
Step S2：These data are manually marked label, manually will be marked by the signal condition figure under the different Vibration Conditions of collection
The data message of note label and signal condition figure is input to convolutional neural networks model and carries out image recognition and CNN training, obtains
The model of vibration for being applied to optical fiber sensing system；
Wherein, the image recognition of the convolutional neural networks model and CNN training are comprised the following steps：
First stage, propagation stage forward, detailed process are as follows：
Discuss that multi-class problem, common c classes are total to N number of training sample using square error cost function, error term is：
In formula (1),The vector of the kth dimension of the corresponding label of n-th sample is represented,Represent the corresponding network of n-th sample
K-th output of output；Total classes of the c for representative sample；For multi-class problem, output is typically organized as the shape of " one-of-c "
Formula, that is, the output node for there was only the corresponding class of the input is just output as, the position of other classes or node are 0 or negative,
Depend on the activation primitive of your output layer；It is exactly -1 that sigmoid is exactly 0, tanh；
It is the summation of the error of each training sample based on the error in whole training sets, for the error of n-th sample,
It is expressed as：
Then according to BP rules calculation cost function E with regard to each weights of network partial derivative；Current layer is represented with l, when
The output of front layer can be expressed as：
Xl=f (ul),with,ul=Wlxl-1+bl(3)
In formula, X represents output, and u represents input, and w represents weights, and b represents biasing, and wherein l represents which layer of network；
Second stage, in the back-propagation stage, detailed process is as follows：
The sensitivity of the base of the error that back propagation is returned, i.e. each neuron is：
The sensitivity of back propagation l layers is：
In formula (5), T represents matrix transpose, and W represents weights, and u represents input,Represent each element multiplication；The god of output layer
Through the sensitivity of unit it is：
Finally, right value update is carried out with delta rules to each neuron；Stated with the form of vector, for l layers, missed
Difference for the input that the derivative that this layer each weights are combined as matrix is this layer with the sensitivity of this layer is：
In formula (7), wherein l represents the number of plies of neutral net, and it is vector which layer X representatives is；W is the weights of l layers, and η is
Learning rate, δ represent sensitivity；For each weights has a specific learning rate.
2. the method for building up of the model of vibration for being applied to optical fiber sensing system according to claim 1, it is characterised in that：Institute
State the distance with fiber position with optical fiber starting point signal condition figure is set up as the first coordinate axess, time for the second coordinate axess, including：
With the distance of fiber position and optical fiber starting point be the first coordinate axess, time be the second coordinate axess, fiber position at several
Temporal signal value sets up signal condition figure for three axes.
3. the method for building up of the model of vibration for being applied to optical fiber sensing system according to claim 1, it is characterised in that：Institute
It is thermodynamic chart to state signal condition figure, and wherein, pixel value in the thermodynamic chart is used for characterizing the optical fiber several fiber positions
Information in several temporal signal values.
4. the method for building up of the model of vibration for being applied to optical fiber sensing system according to claim 1, it is characterised in that：Heat
Try hard to represent that shockproofness ascending from shallow to deep by color.
5. a kind of determination method of the model of vibration of optical fiber sensing system, it is characterised in that comprise the following steps：
Step A：The reception optical fiber signal of telecommunication, obtains
Signal value；Wherein, the fiber optic telecommunications number are converted to by the optical signal of fiber reflection；
Step B：Thermodynamic chart is set up as permanent coordinate axess, time for axis of ordinates with distance of the fiber position with optical fiber starting point；Described
Pixel value in thermodynamic chart is used for characterizing the information of the optical fiber several fiber positions in several temporal signal values；
Step C：Pretreatment is carried out to thermodynamic chart, then according to claim 1 obtain described be applied to optical fiber sensing system
Model of vibration carries out control identification, obtains current model of vibration classification.
6. the determination method of the model of vibration of optical fiber sensing system according to claim 5, it is characterised in that：In step A,
Fiber-optic vibration threshold values is set, if fiber optic telecommunications number do not carry out subsequent step beyond fiber-optic vibration threshold values.
7. the determination method of the model of vibration of optical fiber sensing system according to claim 5, it is characterised in that：Using L=
(c*T/n)/2 calculate the distance between the fiber position and described optical fiber starting point；Wherein, c is light speed in a vacuum, T
For time span of the fiber position in the fiber optic telecommunications number, n is the refractive index of the optical fiber.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610907115.8A CN106503642B (en) | 2016-10-18 | 2016-10-18 | A kind of model of vibration method for building up applied to optical fiber sensing system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610907115.8A CN106503642B (en) | 2016-10-18 | 2016-10-18 | A kind of model of vibration method for building up applied to optical fiber sensing system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106503642A true CN106503642A (en) | 2017-03-15 |
| CN106503642B CN106503642B (en) | 2019-09-20 |
Family
ID=58293680
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610907115.8A Active CN106503642B (en) | 2016-10-18 | 2016-10-18 | A kind of model of vibration method for building up applied to optical fiber sensing system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106503642B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108871760A (en) * | 2018-06-07 | 2018-11-23 | 广东石油化工学院 | A kind of high-efficient gear method of fault pattern recognition |
| CN108932480A (en) * | 2018-06-08 | 2018-12-04 | 电子科技大学 | The study of distributing optical fiber sensing signal characteristic and classification method based on 1D-CNN |
| CN109598342A (en) * | 2018-11-23 | 2019-04-09 | 中国运载火箭技术研究院 | A kind of decision networks model is from game training method and system |
| CN110084986A (en) * | 2019-04-29 | 2019-08-02 | 西人马（厦门）科技有限公司 | A kind of circumference safety protection method and device |
| CN110135283A (en) * | 2019-04-25 | 2019-08-16 | 上海大学 | The signal recognition method of optical fiber perimeter defence system based on FastDTW algorithm |
| CN110363216A (en) * | 2019-05-31 | 2019-10-22 | 上海波汇软件有限公司 | A kind of implementation method applied to on-line training DNN model in DAS system |
| CN110852187A (en) * | 2019-10-22 | 2020-02-28 | 华侨大学 | Method and system for identifying perimeter intrusion event |
| CN111160106A (en) * | 2019-12-03 | 2020-05-15 | 上海微波技术研究所（中国电子科技集团公司第五十研究所） | Method and system for extracting and classifying optical fiber vibration signal features based on GPU |
| CN111538246A (en) * | 2020-07-08 | 2020-08-14 | 浙江浙能天然气运行有限公司 | System and method for estimating interference of mechanical equipment on distributed optical fiber sensor |
| CN112364836A (en) * | 2020-12-07 | 2021-02-12 | 无锡科晟光子科技有限公司 | Vibration optical fiber signal classification method based on full convolution neural network |
| CN112539772A (en) * | 2020-11-02 | 2021-03-23 | 上海大学 | Positioning method of Sagnac distributed optical fiber sensing system based on convolutional neural network integrated learning |
| CN114415018A (en) * | 2022-01-19 | 2022-04-29 | 山东超晟光电科技有限公司 | Self-learning grating interference spectrum analysis technology for motor fault early warning |
| CN114684217A (en) * | 2022-03-16 | 2022-07-01 | 武汉理工大学 | Rail transit health monitoring system and method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110307935A1 (en) * | 2010-06-09 | 2011-12-15 | Verizon Patent And Licensing Inc. | Video content delivery optimization over mobile wireless networks |
| CN105698916A (en) * | 2016-03-01 | 2016-06-22 | 深圳艾瑞斯通技术有限公司 | Optical fiber vibration model determination method and optical fiber early warning device and system |
| CN105787439A (en) * | 2016-02-04 | 2016-07-20 | 广州新节奏智能科技有限公司 | Depth image human body joint positioning method based on convolution nerve network |
| CN105933063A (en) * | 2016-06-20 | 2016-09-07 | 深圳艾瑞斯通技术有限公司 | Optical fiber signal processing method and device, and optical fiber sensing system |
| CN105973451A (en) * | 2016-05-09 | 2016-09-28 | 深圳艾瑞斯通技术有限公司 | Optical fiber vibration model determination method and device |
-
2016
- 2016-10-18 CN CN201610907115.8A patent/CN106503642B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110307935A1 (en) * | 2010-06-09 | 2011-12-15 | Verizon Patent And Licensing Inc. | Video content delivery optimization over mobile wireless networks |
| CN105787439A (en) * | 2016-02-04 | 2016-07-20 | 广州新节奏智能科技有限公司 | Depth image human body joint positioning method based on convolution nerve network |
| CN105698916A (en) * | 2016-03-01 | 2016-06-22 | 深圳艾瑞斯通技术有限公司 | Optical fiber vibration model determination method and optical fiber early warning device and system |
| CN105973451A (en) * | 2016-05-09 | 2016-09-28 | 深圳艾瑞斯通技术有限公司 | Optical fiber vibration model determination method and device |
| CN105933063A (en) * | 2016-06-20 | 2016-09-07 | 深圳艾瑞斯通技术有限公司 | Optical fiber signal processing method and device, and optical fiber sensing system |
Non-Patent Citations (1)
| Title |
|---|
| 陈帮 等: "光纤光栅振动传感器的振动理论分析", 《武汉理工大学学报（信息与管理工程版）》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108871760A (en) * | 2018-06-07 | 2018-11-23 | 广东石油化工学院 | A kind of high-efficient gear method of fault pattern recognition |
| CN108932480B (en) * | 2018-06-08 | 2022-03-15 | 电子科技大学 | Distributed optical fiber sensing signal feature learning and classifying method based on 1D-CNN |
| CN108932480A (en) * | 2018-06-08 | 2018-12-04 | 电子科技大学 | The study of distributing optical fiber sensing signal characteristic and classification method based on 1D-CNN |
| CN109598342A (en) * | 2018-11-23 | 2019-04-09 | 中国运载火箭技术研究院 | A kind of decision networks model is from game training method and system |
| CN110135283A (en) * | 2019-04-25 | 2019-08-16 | 上海大学 | The signal recognition method of optical fiber perimeter defence system based on FastDTW algorithm |
| CN110084986A (en) * | 2019-04-29 | 2019-08-02 | 西人马（厦门）科技有限公司 | A kind of circumference safety protection method and device |
| CN110363216A (en) * | 2019-05-31 | 2019-10-22 | 上海波汇软件有限公司 | A kind of implementation method applied to on-line training DNN model in DAS system |
| CN110852187A (en) * | 2019-10-22 | 2020-02-28 | 华侨大学 | Method and system for identifying perimeter intrusion event |
| CN110852187B (en) * | 2019-10-22 | 2023-04-07 | 华侨大学 | Method and system for identifying perimeter intrusion event |
| CN111160106A (en) * | 2019-12-03 | 2020-05-15 | 上海微波技术研究所（中国电子科技集团公司第五十研究所） | Method and system for extracting and classifying optical fiber vibration signal features based on GPU |
| CN111160106B (en) * | 2019-12-03 | 2023-12-12 | 上海微波技术研究所（中国电子科技集团公司第五十研究所） | GPU-based optical fiber vibration signal feature extraction and classification method and system |
| CN111538246A (en) * | 2020-07-08 | 2020-08-14 | 浙江浙能天然气运行有限公司 | System and method for estimating interference of mechanical equipment on distributed optical fiber sensor |
| CN112539772A (en) * | 2020-11-02 | 2021-03-23 | 上海大学 | Positioning method of Sagnac distributed optical fiber sensing system based on convolutional neural network integrated learning |
| CN112539772B (en) * | 2020-11-02 | 2023-04-07 | 上海大学 | Positioning method of Sagnac distributed optical fiber sensing system based on convolutional neural network integrated learning |
| CN112364836A (en) * | 2020-12-07 | 2021-02-12 | 无锡科晟光子科技有限公司 | Vibration optical fiber signal classification method based on full convolution neural network |
| CN114415018A (en) * | 2022-01-19 | 2022-04-29 | 山东超晟光电科技有限公司 | Self-learning grating interference spectrum analysis technology for motor fault early warning |
| CN114684217A (en) * | 2022-03-16 | 2022-07-01 | 武汉理工大学 | Rail transit health monitoring system and method |
| CN114684217B (en) * | 2022-03-16 | 2024-03-01 | 武汉理工大学 | Rail transit health monitoring system and method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106503642B (en) | 2019-09-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106503642A (en) | A kind of model of vibration method for building up for being applied to optical fiber sensing system | |
| CN108932480A (en) | The study of distributing optical fiber sensing signal characteristic and classification method based on 1D-CNN | |
| CN110363151A (en) | Based on the controllable radar target detection method of binary channels convolutional neural networks false-alarm | |
| CN110570613A (en) | Fence vibration intrusion positioning and mode identification method based on distributed optical fiber system | |
| CN104008399B (en) | The recognition methodss of the gauge pointer shake based on support vector machine in a kind of instrument detection | |
| CN106407990A (en) | Bionic target identification system based on event driving | |
| CN109086773A (en) | Fault plane recognition methods based on full convolutional neural networks | |
| CN109583322A (en) | A kind of recognition of face depth network training method and system | |
| CN108090447A (en) | Hyperspectral image classification method and device under double branch's deep structures | |
| CN107154048A (en) | The remote sensing image segmentation method and device of a kind of Pulse-coupled Neural Network Model | |
| CN104729667A (en) | Method for recognizing disturbance type in a distributed optical fiber vibration sensing system | |
| CN105512622B (en) | A kind of visible remote sensing image sea land dividing method based on figure segmentation and supervised learning | |
| CN104680554A (en) | SURF-based compression tracing method and system | |
| CN114510960A (en) | Method for recognizing distributed optical fiber sensor system mode | |
| CN108388828A (en) | A kind of seashore wetland land cover pattern information extracting method of comprehensive multi- source Remote Sensing Data data | |
| Ge et al. | Coarse-to-fine foraminifera image segmentation through 3D and deep features | |
| Buchanan et al. | Deep convolutional neural networks for detecting dolphin echolocation clicks | |
| CN105930793A (en) | Human body detection method based on SAE characteristic visual learning | |
| Monigari et al. | Plant leaf disease prediction | |
| CN110390949A (en) | Acoustic Object intelligent identification Method based on big data | |
| Ranjith et al. | An IoT based Monitoring System to Detect Animal in the Railway Track using Deep Learning Neural Network | |
| CN106529600A (en) | SVM-based recognition method of building angular points in high-resolution optical image | |
| Zhang et al. | A precise apple leaf diseases detection using BCTNet under unconstrained environments | |
| CN115422962A (en) | Gesture and gesture recognition method and device based on millimeter wave radar and deep learning algorithm | |
| CN110738113B (en) | Object detection method based on adjacent scale feature filtering and transferring |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | 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 |