CN107342810A - Deep learning Brilliant Eyes figure analysis method based on convolutional neural networks - Google Patents
Deep learning Brilliant Eyes figure analysis method based on convolutional neural networks Download PDFInfo
- Publication number
- CN107342810A CN107342810A CN201710534126.0A CN201710534126A CN107342810A CN 107342810 A CN107342810 A CN 107342810A CN 201710534126 A CN201710534126 A CN 201710534126A CN 107342810 A CN107342810 A CN 107342810A
- Authority
- CN
- China
- Prior art keywords
- eye pattern
- analysis
- layer
- neural networks
- convolutional neural
- 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
-
- 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
- H04B10/07—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems
- H04B10/075—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal
- H04B10/079—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal using measurements of the data signal
- H04B10/0795—Performance monitoring; Measurement of transmission parameters
- H04B10/07953—Monitoring or measuring OSNR, BER or Q
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- H04B10/07—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems
- H04B10/075—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal
- H04B10/079—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal using measurements of the data signal
- H04B10/0795—Performance monitoring; Measurement of transmission parameters
-
- 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
- H04B10/07—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems
- H04B10/075—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal
- H04B10/079—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal using measurements of the data signal
- H04B10/0795—Performance monitoring; Measurement of transmission parameters
- H04B10/07951—Monitoring or measuring chromatic dispersion or PMD
Abstract
The invention discloses a kind of deep learning Brilliant Eyes figure analysis method based on convolutional neural networks, it is related to technical field of photo communication, wherein carrying out performance evaluation to eye pattern by building simultaneously training convolutional neural networks module, comprises the following steps:Obtain eye pattern training dataset;Eye pattern is pre-processed;CNN modules are trained to carry out feature extraction;The eye pattern of required analysis is inputted to the CNN modules that training is completed after pretreatment and carries out pattern-recognition and performance evaluation;Export analysis result.Depth learning technology based on convolutional neural networks is applied in eye Diagram Analysis by the present invention, solve the problems, such as directly handle initial data in traditional eye pattern performance evaluation, manual intervention need to be carried out, the intellectuality and automation of eye pattern original image information analysis are realized using CNN, can be as the eye pattern software processing module of oscillograph and the eye Diagram Analysis module of simulation software, and then be embedded into tester and carry out intelligent signal analysis and performance monitoring.
Description
Technical field
The present invention relates to technical field of photo communication, more particularly to a kind of deep learning Brilliant Eyes based on convolutional neural networks
Figure analysis method.
Background technology
Machine learning (ML) technology provides powerful instrument to solve such as natural language processing, data mining, voice
The problem of many fields such as identification and image recognition.Meanwhile machine learning techniques have also obtained widely should in optical communication field
With having greatly facilitated the development of intelligence system.Research is concentrated mainly on to enter using different machine learning algorithms at present
In terms of row optical performance monitoring (OPM) and nonlinear impairments compensation, used machine learning algorithm includes it is expected maximum
(EM), random forest, back-propagation artificial neural network (BP-ANN), k nearest neighbor (KNN) and SVMs (SVM) etc..So
And all above-mentioned machine learning algorithms have the limitation of its algorithm in itself in the ability of feature extraction.More specifically, machine
Learning model can not directly handle the primitive form of natural data, therefore have to need considerable field before with algorithm
Speciality and engineering technical ability carry out design feature extractor, and initial data is converted into suitable internal representation or characteristic vector, and then
Subsystem can just detect the pattern of input data.Therefore, it is intended that more advanced machine learning algorithm can be developed, not only may be used
, can be with the feature needed for automatic detection directly to handle initial data.
Recently, deep learning turns into a burning hot research topic, and the purpose is to cause machine learning closer to artificial intelligence
The target of energy (AI).Deep learning is construed as the deep neural network with multiple non-linear layers, and it passes through self study
Process carries out engineer from data learning feature, rather than by human engineer.Foremost breakthrough in deep learning
One of be Google DeepMind computer program " AlphaGo ", they are hit in Trivial Pursuit Unhinged with the ability of self study first
The player of specialty is lost.In addition, as current study hotspot, deep learning is in unmanned vehicle, medical diagnosis, mood
The various application fields such as analysis achieve major progress.But as far as we know, almost there is no base but in optical communication system field
In the research work of deep learning.
Meanwhile in optical communication field, the identification of current modulation format and OSNR, CD, linear damage, nonlinear impairments etc.
The estimation technique of performance indications can not be handled directly initial data, and must artificially extract corresponding feature, it is necessary to
Substantial amounts of manual intervention.It is desirable to can carry out the intelligence of various performances point using more advanced technology using eye pattern
Analysis, without manual intervention, accomplishes accurately to measure, is handled immediately without data statistics, realizes and carries out performance evaluation using eye pattern
Intelligent and automation.
The content of the invention
It is an object of the invention to deep learning technology is applied into optical communication field, there is provided a kind of intelligence, reliable base
In the deep learning Brilliant Eyes figure analysis method of convolutional neural networks, solve directly handle original in traditional eye pattern performance evaluation
Beginning view data, the drawbacks of manual intervention need to be carried out, realize and the intelligent of performance evaluation and oneself are carried out to eye pattern original image
Dynamicization.
To reach above-mentioned purpose, the invention discloses a kind of deep learning intelligence eye Diagram Analysis based on convolutional neural networks
Method, the deep learning technology based on convolutional neural networks is applied in eye Diagram Analysis, using convolutional neural networks to eye pattern
Multiple performance analysis is carried out, the described method comprises the following steps:Step 1: the eye pattern training dataset of analysis needed for obtaining;Step
Rapid two, eye pattern image preprocessing;Step 3: training convolutional neural networks (CNN) module carries out feature extraction to eye pattern;Step
4th, the CNN modules that required analysis eye pattern input training is completed are subjected to pattern-recognition and performance evaluation;Step 5: output analysis
As a result.
Preferably, the multiple performance analyzed needed for the eye pattern is modulation format, OSNR (OSNR), dispersion
(CD), linear damage and nonlinear impairments.
Preferably, in the eye pattern training set obtaining step one, in the case of the various performance difference indexs for gathering eye pattern
Training dataset, wherein, every group of data that training data is concentrated are the specific of particular characteristic for eye pattern image and output by inputting
Indication information is to forming.
Preferably, in the eye pattern pre-treatment step two, by the colour of the training data obtained in the step 1 concentration
Eye pattern image is converted to gray level image, and obtained eye pattern gray level image is carried out into down-sampling processing.
Preferably, the training CNN modules are carried out in characteristic extraction step three, by pretreated eye in the step 2
In the CNN modules that figure input is built, after being trained process based on the training data, the CNN modules are automatically from eye pattern
Feature, and the relation between construction feature and different performance are extracted in image.
Preferably, in the CNN template patterns identification and performance evaluation step 4, the eye pattern of pretreated required analysis
Input in the CNN modules that the training is completed, CNN modules carry out pattern-recognition to the eye pattern of input, and pass through its conventional
Habit experience carries out performance evaluation to the eye pattern currently inputted.
Preferably, in the output analysis result step 5, by the packet of CNN modules output containing required analysis
Various performances, the analysis result of different performance can be obtained from output information.
Preferably, the structure of the CNN modules mainly includes:One input layer, n convolutional layer (C1, C2 ..., Cn), n
Individual pond layer (P1, P2 ..., Pn), m full articulamentums (F1, F2 ..., Fm), an output layer, wherein, the input layer it is defeated
Enter and be connected for the eye pattern image by pretreatment, input layer with convolutional layer C1;It is a1 that the convolutional layer C1, which contains k1 size,
× a1 convolution kernel, the input tomographic image obtains k1 characteristic pattern by convolutional layer C1, and then obtained characteristic pattern is transmitted
To pond layer P1;The characteristic pattern that the pond layer P1 is generated with b1 × b1 sample size to the convolutional layer C1 carries out pond,
The characteristic pattern after corresponding k1 sampling is obtained, then obtained characteristic pattern is sent to next convolutional layer C2;The n convolution
Layer and pond layer constantly extract the profound Sampling characters of image to being sequentially connected with, last pond layer Pn and entirely
Articulamentum F1 is connected, wherein, convolutional layer Ci contains the convolution kernel that ki size is ai × ai, and pond layer Pj sample size is
Bj × bj, Ci represent i-th of convolutional layer, and Pj represents j-th of pond layer;The full articulamentum F1 is last described pond layer
The one-dimensional layer that the pixel mapping of all kn characteristic patterns obtained by Pn forms, each pixel represent the one of the full articulamentum F1
Individual neuron node, all neuron nodes of F1 layers are connected entirely with next full articulamentum F2 neuron node;Through m
Individual full articulamentum is sequentially connected with, and last full articulamentum Fm is connected entirely with the output layer;The output layer exports institute
The nodal information for the eye pattern different performance that need to be analyzed.
Preferably, the nodal information of the output layer output is the binary bit sequence of L positions, wherein, N number of difference
Performance respectively with L1, L2 ..., LN positions binary bits information represent that Li positions are used to represent that the Li kinds of i-th of performance to be different
Indication information, wherein L=L1+L2+ ...+LN.
CNN eye pattern Processing Algorithm is preferably based on using as the eye pattern software processing module or simulation software of oscillograph
Eye Diagram Analysis module, and then be embedded into tester and carry out intelligent signal analysis and performance monitoring.
The beneficial effects of the present invention are:The present invention solves the drawbacks of traditional eye Diagram Analysis, will be based on convolutional Neural net
The deep learning technology of network is applied in eye Diagram Analysis, and multiple performance analysis, application are carried out to eye pattern using convolutional neural networks
The present invention can be directly handled eye pattern raw image data, without carrying out feature extraction by manual intervention, realize eye
The intellectuality and automation of figure performance evaluation, and then can be as the eye pattern software processing module of oscillograph or the eye of simulation software
Map analysis module, it is embedded into tester and carries out intelligent signal analysis and performance monitoring.
Brief description of the drawings
Fig. 1 shows the flow chart of the deep learning Brilliant Eyes figure analysis method of the invention based on convolutional neural networks;
Fig. 2 shows the deep learning intelligence eye Diagram Analysis structure based on convolutional neural networks of one embodiment of the invention
Schematic diagram;
Fig. 3 shows the different modulating form of one embodiment of the invention collection and different OSNR part eye pattern image;
Fig. 4 shows OSNR estimated under the different modulating form of one embodiment of the invention accuracy schematic diagram;
Fig. 5 shows under the different modulating form of one embodiment of the invention CNN with other machines learning algorithm for eye
The contrast schematic diagram of figure performance evaluation accuracy.
Embodiment
With reference to the accompanying drawings and examples, the embodiment of the present invention is described in further detail.Implement below
Example is used to illustrate the present invention, but is not limited to protection scope of the present invention.
As shown in figure 1, the deep learning Brilliant Eyes figure analysis method proposed by the present invention based on convolutional neural networks, by base
It is applied in the deep learning technology of convolutional neural networks in eye Diagram Analysis, a variety of property is carried out to eye pattern using convolutional neural networks
It can analyze, comprise the following steps:Step 1: the eye pattern training dataset of analysis needed for obtaining;Step 2: eye pattern image is located in advance
Reason;Step 3: training convolutional neural networks (CNN) module carries out feature extraction to eye pattern;Step 4: required analysis eye pattern input
The CNN modules that training is completed carry out pattern-recognition and performance evaluation;Step 5: output analysis result.
In the present embodiment, the eye pattern performance to be analyzed is modulation format and OSNR.
In the acquisition eye pattern training dataset step 1, base is established based on VPI Transmission Maker 9.0
This analogue system, the optical signal of four kinds of different modulating forms is generated by pseudo-random binary sequence, be respectively:4PAM, RZ-
DPSK, NRZ-OOK, RZ-OOK.Four kinds of modulation formats are all based on direct detection mode, and the message reflection of transmission is in signal
In amplitude, it is suitable for follow-up eye Diagram Analysis.Erbium-doped fiber amplifier (EDFA) is used in analogue system by the spontaneous hair of amplification
Penetrate (ASE) noise to be added in optical signal, and under 1dB step-length, be adjusted to OSNR using variable optical attenuator (VOA)
10 to 25dB.In order to simulate real optical signal as far as possible, dispersion (CD) emulator is added in system so that simulate generation
Eye pattern can more reflect real situation.For the optical signal of four kinds of different modulating forms in the present embodiment, 4PAM, NRZ-OOK and
RZ-OOK signals directly detect by photoelectric detector (PD), and RZ-DPSK signals with delay interferometer (DI) then by combining
Photoelectric detector (BPD) is balanced to be detected.After sampling is synchronized, obtain comprising four kinds of signal strength informations
Data signal.In order to obtain visual effect more true to nature, the present embodiment uses eye pattern generation module special in oscillograph, will
The data signal received is converted to corresponding eye pattern image.
Based on the analogue system, the different OSNR values of the every kind of modulation format generation 16 of the present embodiment regulation ()
Eye pattern image, " jpg " form that 100 pixel sizes are 900 × 1200 is collected to each OSNR values of every kind of modulation format
Eye pattern image, here, using each OSNR values of every kind of modulation format and its corresponding eye pattern image as one group of training data,
Therefore whole training data set includes 6400 (1600 × 4) group training data altogether.
In the eye pattern image preprocessing step 2, in order to reduce amount of calculation and enhancing generalization ability, it will be received in step 1
The eye pattern image collected causes original coloured image to be converted to gray level image after greyscale transformation, and causes original by down-sampling
The pixel size of beginning eye pattern is down to 28 × 28, is finally input to the training dataset after processing in the CNN modules established.Such as
Shown in Fig. 3, different eye patterns can show different modulation formats, and if observed eye pattern visually be entered
Row carefully analyzes, and it is it can also be seen that eye pattern and the first approximation relation of OSNR values.
The training CNN modules are carried out in characteristic extraction step three, wherein the eye pattern training dataset of input CNN modules,
Each of which eye pattern image corresponds with a label vector being made up of 20 bits, and first 4 of label vector represent not
Same modulation format (4PAM:0001、RZ-DPSK:0010、NRZ-OOK:0100、RZ-OOK:1000), latter 16 represent difference
OSNR values (10dB:0000000000000001、11dB:0000000000000010 ..., 25dB:
1000000000000000).In described training process, CNN modules gradually extract the validity feature of input eye pattern image.Together
When, in order to minimize the error between desired tag vector sum reality output label vector, CNN modules are used by backpropagation
The method that gradient declines progressively adjusts the parameter of its kernel.
Fig. 2 represents the intelligent eye Diagram Analysis structural representation based on convolutional neural networks of a specific embodiment of the invention
Figure, the structure of the CNN modules mainly include following components:One input layer, two convolutional layers (C1, C2), two ponds
Change layer (P1, P2), full articulamentum (F1), an output layer.28 × 28 eye pattern images by pretreatment are as input layer
CNN modules are inputted, are connected with convolutional layer C1;The eye pattern image of input passes through the volume containing the convolution kernel that 6 sizes are 5 × 5
Lamination C1, it is 24 × 24 characteristic patterns to obtain 6 sizes, and then obtained characteristic pattern is sent into pond layer P1;Pond layer P1 is with 2
× 2 sample size carries out maximum pond to 6 characteristic patterns, obtains the feature after the sampling that corresponding 6 sizes are 12 × 12
Figure, and then obtained characteristic pattern is sent to convolutional layer C2;Convolutional layer C2 contains the convolution kernel that 12 sizes are 5 × 5, pond layer
6 characteristic patterns obtained by P1 obtain the characteristic pattern that 12 sizes are 8 × 8 through convolutional layer C2, and then obtained characteristic pattern is transmitted
To pond layer P2;12 sizes that pond layer P2 is equally generated using 2 × 2 sample size to convolutional layer C2 are entered as 4 × 4 characteristic patterns
The maximum pond of row, the characteristic pattern after corresponding 12 samplings is obtained, obtained characteristic pattern is then sent to full articulamentum F1;Pond
The pixel for changing all characteristic patterns obtained by layer P2 is mapped as one-dimensional full articulamentum F1, and each pixel represents full articulamentum F1's
One neuron node, full articulamentum F1 each neuron node are connected entirely with output layer;Last output layer output institute
The nodal information for the eye pattern performance that need to be analyzed.
Wherein, convolutional layer is the core component of CNN modules.Parameter in this layer is made up of one group of convolution kernel, and they have
Less local receptor field, but extend to the entire depth of eye pattern image.During propagating forward, each convolution
Core carries out convolution with the pixel on the width and height of eye pattern image, exports a two-dimentional plane, it is referred to as from the volume
The characteristic pattern of product karyogenesis.Different from the classical convolution in mathematics, the operation in CNN is discrete convolution, can be counted as square
Battle array is multiplied.Convolution kernel can be looked at as property detector, and by convolution kernel, CNN modules can be from the image learning of input
The feature exclusive to its, while in order to build a significantly more efficient model, multiple convolution kernels are generally required to detect multiple spies
Sign, to produce multiple characteristic patterns in convolutional layer.After the feature extraction by convolutional layer, pond layer can will be semantically similar
Feature be merged into corresponding one, typical pond mode is to calculate the maximum of local unit block in a characteristic pattern, is entered
The sub-sampling of row characteristic pattern.Each sub-sample unit 2 × 2 unit area from convolution characteristic pattern obtains defeated in the present embodiment
Enter, and the numerical value using the maximum of these inputs as Chi Huahou, and then form the characteristic pattern behind pond.
In the CNN template patterns identification and performance evaluation step 4,4 kinds of different modulatings of pretreated required analysis
There is scope to be for form, every kind of modulation formatOSNR values (using 1dB as step-length) eye pattern image be input to it is above-mentioned
To train in the CNN modules completed, CNN modules carry out pattern-recognition to the eye pattern under the different situations of input, and by training rank
Section learning experience form and OSNR performance evaluation are modulated to the eye pattern of input, by analysis result with the bit of 20 to
The form output of amount.
In the output analysis result step 5, from 20 bit vectors of CNN modules output, its first 4 available
The modulation format information of analyzed eye pattern, latter 16 can obtain corresponding OSNR values.
To show the accuracy of institute's extracting method analysis of the present invention, Fig. 4 is shown under different modulating form difference iterations
Estimated accuracy of the CNN modules to OSNR.Obviously, the accuracy of four kinds of modulation formats is with the increase of CNN module iterationses
And increase.The CNN modules that different iterationses are trained have different performance recognition capabilities.In the present embodiment, iteration is worked as
When number is more than 31, CNN modules reach 100% to its corresponding OSNR precision estimated under four kinds of modulation formats, that is, divide
The performance inerrancy result of analysis.
Meanwhile be the advantage for proving the present invention, by CNN and other four kinds of famous machine learning algorithms, i.e. decision tree,
KNN, BP-ANN and SVM are compared.Each algorithm for OSNR under different modulating form estimated accuracy in histogram
Form is shown in Fig. 5, and CNN has obvious advantage for other four kinds of algorithms as seen from the figure.Wherein, decision Tree algorithms processing speed
Fast and the requirement very little to internal memory, these advantages also cause its estimated accuracy relatively low simultaneously;KNN algorithms generally have in low dimensional
There is good estimated accuracy, but very big deviation may be produced on high-dimensional;SVM algorithm uses in estimated accuracy and internal memory
On be respectively provided with very big advantage, it only needs seldom supporting vector, but it is substantially a binary classifier, so
Value in face of multiple OSNR just needs multiple SVM classifiers to be handled;Although BP neural network is also to be sent out from neutral net
Exhibition, but it lacks the ability of feature extraction, it is necessary to which a large amount of training datas can be only achieved preferable effect, and be easily trapped into
Local minimum and over-fitting.Compared with algorithm above, CNN is constructed to the sensitive relatively low of input data variance
Network is more powerful, can largely avoid over-fitting, and can automatically extract the feature of input data, especially
It is that have extraordinary effect on image procossing, simultaneously as the advantage such as local receptor field, weight distribution, sub-sampling,
CNN can realize optimal accuracy with appropriate calculating cost.
To sum up, the deep learning technology based on convolutional neural networks is applied to eye Diagram Analysis by method proposed by the invention
In, the eye pattern software processing module of oscillograph or the eye Diagram Analysis module of simulation software can be effective as, and then be embedded into
Intelligent signal analysis and performance monitoring are carried out in tester, realizes the automation and intellectuality of eye Diagram Analysis.
Above example is merely to illustrate the present invention, and and protection scope of the present invention is any limitation as, led for correlation
The technical staff in domain, the present invention can have various modifications and variations.Within the spirit and principles of the invention, that is made is any
Modification, equivalent substitution, improvement etc., should be included in the scope of the protection.
Claims (10)
1. a kind of deep learning Brilliant Eyes figure analysis method based on convolutional neural networks, it is characterised in that convolution god will be based on
Deep learning technology through network is applied in eye Diagram Analysis, and multiple performance analysis is carried out to eye pattern using convolutional neural networks,
It the described method comprises the following steps:
Step 1: the eye pattern training dataset of analysis needed for obtaining;
Step 2: eye pattern image preprocessing;
Step 3: training convolutional neural networks (CNN) module carries out feature extraction to eye pattern;
Step 4: the CNN modules that required analysis eye pattern input training is completed carry out pattern-recognition and performance evaluation;
Step 5: output analysis result.
2. the deep learning Brilliant Eyes figure analysis method according to claim 1 based on convolutional neural networks, its feature exist
In, the multiple performance analyzed needed for the eye pattern be modulation format, OSNR (OSNR), dispersion (CD), linear damage and
Nonlinear impairments.
3. the deep learning Brilliant Eyes figure analysis method according to claim 1 based on convolutional neural networks, its feature exist
In, in the eye pattern training set obtaining step one, the training dataset in the case of the various performance difference indexs of eye pattern is gathered, its
In, training data concentrate every group of data by input for eye pattern image and output be particular characteristic specific indexes information to structure
Into.
4. the deep learning Brilliant Eyes figure analysis method according to claim 1 based on convolutional neural networks, its feature exist
In in the eye pattern pre-treatment step two, the colored eye pattern image that the training data obtained in the step 1 is concentrated is changed
For gray level image, and obtained eye pattern gray level image is subjected to down-sampling processing.
5. the deep learning Brilliant Eyes figure analysis method according to claim 1 based on convolutional neural networks, its feature exist
In the training CNN modules are carried out in characteristic extraction step three, and pretreated eye pattern input in the step 2 is built
CNN modules in, after being trained process based on the training data, the CNN modules are extracted special from eye pattern image automatically
Sign, and the relation between construction feature and different performance.
6. the deep learning Brilliant Eyes figure analysis method according to claim 1 based on convolutional neural networks, its feature exist
In in the CNN template patterns identification and performance evaluation step 4, the eye pattern of pretreated required analysis inputs the training
In the CNN modules of completion, CNN modules carry out pattern-recognition to the eye pattern of input, and by its conventional learning experience to current
The eye pattern of input carries out performance evaluation.
7. the deep learning Brilliant Eyes figure analysis method according to claim 1 based on convolutional neural networks, its feature exist
In described to export in analysis result step 5, various performances of the packet containing required analysis exported by the CNN modules can
The analysis result of different performance is obtained from output information.
8. the deep learning Brilliant Eyes figure analysis method according to claim 1 based on convolutional neural networks, its feature exist
In the structure of the CNN modules mainly includes:One input layer, n convolutional layer (C1, C2 ..., Cn), n pond layer (P1,
P2 ..., Pn), the full articulamentum of m (F1, F2 ..., Fm), an output layer;
Wherein, the input of the input layer is to be connected by the eye pattern image of pretreatment, input layer with convolutional layer C1;
The convolutional layer C1 contains k1Individual size is a1×a1Convolution kernel, it is described input tomographic image obtain k by convolutional layer C11It is individual
Characteristic pattern, and then obtained characteristic pattern is sent to pond layer P1;
The pond layer P1 is with b1×b1The characteristic pattern that is generated to the convolutional layer C1 of sample size carry out pond, obtain corresponding
K1Characteristic pattern after individual sampling, then obtained characteristic pattern is sent to next convolutional layer C2;
The n convolutional layer and pond layer constantly extract the profound Sampling characters of image to being sequentially connected with, last
Individual pond layer Pn is connected with full articulamentum F1, wherein, convolutional layer Ci contains kiIndividual size is ai×aiConvolution kernel, pond layer
Pj sample size is bj×bj, Ci i-th of convolutional layer of expression, Pj j-th of pond layer of expression;
The full articulamentum F1 is all k obtained by last pond layer PnnWhat the pixel mapping of individual characteristic pattern formed
One-dimensional layer, each pixel represent a neuron node of the full articulamentum F1, all neuron nodes of F1 layers with it is next
Individual full articulamentum F2 neuron node is connected entirely;
It is sequentially connected with through m full articulamentums, last full articulamentum Fm is connected entirely with the output layer;
The nodal information of the eye pattern different performance of analysis needed for the output layer output.
9. the deep learning Brilliant Eyes figure analysis method according to claim 8 based on convolutional neural networks, its feature exist
In, the nodal information of output layer output is the binary bit sequence of L positions, wherein, N number of different performance respectively with
L1、L2、…、LNPosition binary bits information represents, LiPosition is used for the L for representing i-th of performanceiThe different indication information of kind, its
Middle L=L1+L2+…+LN。
10. the deep learning Brilliant Eyes figure analysis method according to claim 1 based on convolutional neural networks, its feature exist
In the eye pattern Processing Algorithm based on CNN is using as the eye Diagram Analysis mould of the eye pattern software processing module of oscillograph or simulation software
Block, and then be embedded into tester and carry out intelligent signal analysis and performance monitoring.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710534126.0A CN107342810B (en) | 2017-07-03 | 2017-07-03 | Deep learning Brilliant Eyes figure analysis method based on convolutional neural networks |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710534126.0A CN107342810B (en) | 2017-07-03 | 2017-07-03 | Deep learning Brilliant Eyes figure analysis method based on convolutional neural networks |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107342810A true CN107342810A (en) | 2017-11-10 |
CN107342810B CN107342810B (en) | 2019-11-19 |
Family
ID=60218952
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710534126.0A Active CN107342810B (en) | 2017-07-03 | 2017-07-03 | Deep learning Brilliant Eyes figure analysis method based on convolutional neural networks |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107342810B (en) |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108446631A (en) * | 2018-03-20 | 2018-08-24 | 北京邮电大学 | The smart frequency spectrum figure analysis method of deep learning based on convolutional neural networks |
CN108957125A (en) * | 2018-03-20 | 2018-12-07 | 北京邮电大学 | Smart frequency spectrum figure analysis method based on machine learning |
CN109120563A (en) * | 2018-08-06 | 2019-01-01 | 电子科技大学 | A kind of Modulation Identification method based on Artificial neural network ensemble |
CN109217923A (en) * | 2018-09-28 | 2019-01-15 | 北京科技大学 | A kind of joint optical information networks and rate, modulation format recognition methods and system |
CN109547102A (en) * | 2018-12-17 | 2019-03-29 | 北京邮电大学 | A kind of optical information networks method, apparatus, electronic equipment and readable storage medium storing program for executing |
CN109768944A (en) * | 2018-12-29 | 2019-05-17 | 苏州联讯仪器有限公司 | A kind of signal modulation identification of code type method based on convolutional neural networks |
CN109905167A (en) * | 2019-02-25 | 2019-06-18 | 苏州工业园区新国大研究院 | A kind of optical communication system method for analyzing performance based on convolutional neural networks |
CN111157551A (en) * | 2018-11-07 | 2020-05-15 | 浦项工科大学校产学协力团 | Method for analyzing perovskite structure using machine learning |
CN111863104A (en) * | 2020-07-29 | 2020-10-30 | 展讯通信(上海)有限公司 | Eye pattern determination model training method, eye pattern determination device, eye pattern determination apparatus, and medium |
CN111860852A (en) * | 2019-04-30 | 2020-10-30 | 百度时代网络技术(北京)有限公司 | Method, device and system for processing data |
CN111934755A (en) * | 2020-07-08 | 2020-11-13 | 国网宁夏电力有限公司电力科学研究院 | SDN controller and optical signal-to-noise ratio prediction method of optical communication equipment |
CN112115760A (en) * | 2019-06-20 | 2020-12-22 | 和硕联合科技股份有限公司 | Object detection system and object detection method |
CN112836422A (en) * | 2020-12-31 | 2021-05-25 | 电子科技大学 | Interference and convolution neural network mixed scheme measuring method |
CN113141214A (en) * | 2021-04-06 | 2021-07-20 | 中山大学 | Deep learning-based underwater optical communication misalignment robust blind receiver design method |
US11907090B2 (en) | 2021-08-12 | 2024-02-20 | Tektronix, Inc. | Machine learning for taps to accelerate TDECQ and other measurements |
US11923895B2 (en) | 2021-03-24 | 2024-03-05 | Tektronix, Inc. | Optical transmitter tuning using machine learning and reference parameters |
US11923896B2 (en) | 2021-03-24 | 2024-03-05 | Tektronix, Inc. | Optical transceiver tuning using machine learning |
US11940889B2 (en) | 2021-08-12 | 2024-03-26 | Tektronix, Inc. | Combined TDECQ measurement and transmitter tuning using machine learning |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105205453A (en) * | 2015-08-28 | 2015-12-30 | 中国科学院自动化研究所 | Depth-auto-encoder-based human eye detection and positioning method |
CN106326874A (en) * | 2016-08-30 | 2017-01-11 | 天津中科智能识别产业技术研究院有限公司 | Method and device for recognizing iris in human eye images |
CN106650688A (en) * | 2016-12-30 | 2017-05-10 | 公安海警学院 | Eye feature detection method, device and recognition system based on convolutional neural network |
GB2545661A (en) * | 2015-12-21 | 2017-06-28 | Nokia Technologies Oy | A method for analysing media content |
-
2017
- 2017-07-03 CN CN201710534126.0A patent/CN107342810B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105205453A (en) * | 2015-08-28 | 2015-12-30 | 中国科学院自动化研究所 | Depth-auto-encoder-based human eye detection and positioning method |
GB2545661A (en) * | 2015-12-21 | 2017-06-28 | Nokia Technologies Oy | A method for analysing media content |
CN106326874A (en) * | 2016-08-30 | 2017-01-11 | 天津中科智能识别产业技术研究院有限公司 | Method and device for recognizing iris in human eye images |
CN106650688A (en) * | 2016-12-30 | 2017-05-10 | 公安海警学院 | Eye feature detection method, device and recognition system based on convolutional neural network |
Non-Patent Citations (1)
Title |
---|
赖俊森: "基于眼图重构和人工神经网络的光性能监测", 《光电子·激光》 * |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108957125A (en) * | 2018-03-20 | 2018-12-07 | 北京邮电大学 | Smart frequency spectrum figure analysis method based on machine learning |
CN108446631A (en) * | 2018-03-20 | 2018-08-24 | 北京邮电大学 | The smart frequency spectrum figure analysis method of deep learning based on convolutional neural networks |
CN109120563B (en) * | 2018-08-06 | 2020-12-29 | 电子科技大学 | Modulation recognition method based on neural network integration |
CN109120563A (en) * | 2018-08-06 | 2019-01-01 | 电子科技大学 | A kind of Modulation Identification method based on Artificial neural network ensemble |
CN109217923A (en) * | 2018-09-28 | 2019-01-15 | 北京科技大学 | A kind of joint optical information networks and rate, modulation format recognition methods and system |
CN111157551A (en) * | 2018-11-07 | 2020-05-15 | 浦项工科大学校产学协力团 | Method for analyzing perovskite structure using machine learning |
CN109547102A (en) * | 2018-12-17 | 2019-03-29 | 北京邮电大学 | A kind of optical information networks method, apparatus, electronic equipment and readable storage medium storing program for executing |
CN109768944A (en) * | 2018-12-29 | 2019-05-17 | 苏州联讯仪器有限公司 | A kind of signal modulation identification of code type method based on convolutional neural networks |
CN109905167A (en) * | 2019-02-25 | 2019-06-18 | 苏州工业园区新国大研究院 | A kind of optical communication system method for analyzing performance based on convolutional neural networks |
CN111860852A (en) * | 2019-04-30 | 2020-10-30 | 百度时代网络技术(北京)有限公司 | Method, device and system for processing data |
CN112115760A (en) * | 2019-06-20 | 2020-12-22 | 和硕联合科技股份有限公司 | Object detection system and object detection method |
CN112115760B (en) * | 2019-06-20 | 2024-02-13 | 和硕联合科技股份有限公司 | Object detection system and object detection method |
TWI738009B (en) * | 2019-06-20 | 2021-09-01 | 和碩聯合科技股份有限公司 | Object detection system and object detection method |
US11195083B2 (en) | 2019-06-20 | 2021-12-07 | Pegatron Corporation | Object detection system and object detection method |
CN111934755B (en) * | 2020-07-08 | 2022-03-25 | 国网宁夏电力有限公司电力科学研究院 | SDN controller and optical signal-to-noise ratio prediction method of optical communication equipment |
CN111934755A (en) * | 2020-07-08 | 2020-11-13 | 国网宁夏电力有限公司电力科学研究院 | SDN controller and optical signal-to-noise ratio prediction method of optical communication equipment |
CN111863104A (en) * | 2020-07-29 | 2020-10-30 | 展讯通信(上海)有限公司 | Eye pattern determination model training method, eye pattern determination device, eye pattern determination apparatus, and medium |
CN111863104B (en) * | 2020-07-29 | 2023-05-09 | 展讯通信(上海)有限公司 | Eye diagram judgment model training method, eye diagram judgment device, eye diagram judgment equipment and medium |
CN112836422B (en) * | 2020-12-31 | 2022-03-18 | 电子科技大学 | Interference and convolution neural network mixed scheme measuring method |
CN112836422A (en) * | 2020-12-31 | 2021-05-25 | 电子科技大学 | Interference and convolution neural network mixed scheme measuring method |
US11923895B2 (en) | 2021-03-24 | 2024-03-05 | Tektronix, Inc. | Optical transmitter tuning using machine learning and reference parameters |
US11923896B2 (en) | 2021-03-24 | 2024-03-05 | Tektronix, Inc. | Optical transceiver tuning using machine learning |
CN113141214A (en) * | 2021-04-06 | 2021-07-20 | 中山大学 | Deep learning-based underwater optical communication misalignment robust blind receiver design method |
US11907090B2 (en) | 2021-08-12 | 2024-02-20 | Tektronix, Inc. | Machine learning for taps to accelerate TDECQ and other measurements |
US11940889B2 (en) | 2021-08-12 | 2024-03-26 | Tektronix, Inc. | Combined TDECQ measurement and transmitter tuning using machine learning |
Also Published As
Publication number | Publication date |
---|---|
CN107342810B (en) | 2019-11-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107342810B (en) | Deep learning Brilliant Eyes figure analysis method based on convolutional neural networks | |
CN107342962A (en) | Deep learning intelligence Analysis On Constellation Map method based on convolutional neural networks | |
Wang et al. | Modulation format recognition and OSNR estimation using CNN-based deep learning | |
CN105046277B (en) | Robust mechanism study method of the feature significance in image quality evaluation | |
CN108427921A (en) | A kind of face identification method based on convolutional neural networks | |
CN107944386B (en) | Visual scene recognition methods based on convolutional neural networks | |
CN109063569A (en) | A kind of semantic class change detecting method based on remote sensing image | |
CN109344759A (en) | A kind of relatives' recognition methods based on angle loss neural network | |
CN104834905A (en) | Facial image identification simulation system and method | |
CN106778910A (en) | Deep learning system and method based on local training | |
CN110490242A (en) | Training method, eye fundus image classification method and the relevant device of image classification network | |
CN106682702A (en) | Deep learning method and system | |
CN107423727A (en) | Face complex expression recognition methods based on neutral net | |
CN106997373A (en) | A kind of link prediction method based on depth confidence network | |
CN109190521A (en) | A kind of construction method of the human face recognition model of knowledge based purification and application | |
Lv et al. | Joint OSNR monitoring and modulation format identification on signal amplitude histograms using convolutional neural network | |
Capece et al. | Implementation of a coin recognition system for mobile devices with deep learning | |
CN113435254A (en) | Sentinel second image-based farmland deep learning extraction method | |
CN113989217A (en) | Human eye diopter detection method based on deep learning | |
CN110289987A (en) | Multi-agent system network resilience appraisal procedure based on representative learning | |
CN115965819A (en) | Lightweight pest identification method based on Transformer structure | |
Wang et al. | Identification of growing points of cotton main stem based on convolutional neural network | |
Huang et al. | Application of Data Augmentation and Migration Learning in Identification of Diseases and Pests in Tea Trees | |
CN110189361A (en) | A kind of multi-channel feature and the target following preferentially updated parallel | |
Gong et al. | Research on high resolution remote sensing image classification based on convolution neural network |
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 |