CN109166099A - The steel rail defect measurement method of view-based access control model associative mechanisms - Google Patents
The steel rail defect measurement method of view-based access control model associative mechanisms Download PDFInfo
- Publication number
- CN109166099A CN109166099A CN201810797993.8A CN201810797993A CN109166099A CN 109166099 A CN109166099 A CN 109166099A CN 201810797993 A CN201810797993 A CN 201810797993A CN 109166099 A CN109166099 A CN 109166099A
- Authority
- CN
- China
- Prior art keywords
- association
- vector
- sample
- associative
- rail
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- 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
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/061—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using biological neurons, e.g. biological neurons connected to an integrated circuit
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Theoretical Computer Science (AREA)
- Molecular Biology (AREA)
- Neurology (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Microelectronics & Electronic Packaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of steel rail defect measurement methods of view-based access control model associative mechanisms.This method comprises: extracting the various dimensions feature in rail image；Association analysis is carried out by various dimensions feature of the neuron interaction associative network to shown rail image, generates association's vector；Matching mapping processing is carried out to association's vector in priori knowledge base, obtains the corresponding association's sample of association's vector；Analysis decision is carried out to association's sample using decision function, obtains association as a result, determining whether rail is defective according to association's result.The present invention is directed to steel rail defect test problems, inquire into the associative mechanisms model for establishing stratification, propose the green neuron interaction associative network with associative ability, and associative mechanisms model is constructed based on the WHAT access in mankind's visual cortex on this basis, compared with traditional steel rail defect detection method, associative mechanisms significantly improve system for the detection performance of steel rail defect.
Description
Technical field
The present invention relates to steel rail defect detection technique field more particularly to a kind of steel rail defects of view-based access control model associative mechanisms
Measurement method.
Background technique
As train running speed improves, the raising of traffic density, the raising of train operation reliability and security requirement,
The detection of steel rail defect becomes the hot spot of recent domestic research.Correlation scholar proposes much rich in constructive both at home and abroad
The shortcomings that method simultaneously achieves certain success, but there are still environmental suitability and poor robustness at present, and for severe day
Steel rail defect identification under gas (such as haze, sleet) is difficult to obtain satisfactory discrimination, this seriously constrains steel
The development of rail defect recognition technology.
A kind of steel rail defect recognition methods in the prior art is to extract steel rail defect region, the party using the method for statistics
Rail image is first divided into any piece by method, then rail image block is divided into two class of zero defect and defect with single order gray-scale statistical, from
And the region that the defects of detects rail image.The disadvantages of this method is that the precision of calculated result is relatively low.
Another kind steel rail defect recognition methods in the prior art is that the geometrical defect based on gray histogram curve extracts
Method, this method first count the grey level histogram both horizontally and vertically of Rail Surface image, extract Rail Surface image
In both horizontally and vertically gray scale all change bigger region, then defect is extracted using a kind of adaptive threshold segmentation method
Region.The disadvantages of this method are as follows: can only obtain the general shape of defect, the defect area precision of extraction is not high.
Summary of the invention
It is existing to overcome the embodiment provides a kind of steel rail defect measurement method of view-based access control model associative mechanisms
The shortcomings that technology.
To achieve the goals above, this invention takes following technical solutions.
A kind of steel rail defect measurement method of view-based access control model associative mechanisms, comprising:
Extract the various dimensions feature in rail image；
Association analysis is carried out by various dimensions feature of the neuron interaction associative network to shown rail image, generates association
Vector；
Matching mapping processing is carried out to association's vector in priori knowledge base, it is corresponding to obtain association's vector
Think sample；
Analysis decision is carried out to association's sample using decision function, obtains association as a result, according to association's result
Determine whether rail is defective.
Further, the various dimensions feature in the extraction rail image, comprising:
The rail image of defect to be identified is obtained, the V1-V4 layer in analog vision system WHAT access passes through the WHAT
V1-V4 layer in access extracts the various dimensions feature in the rail image, and it is special to generate various dimensions based on above-mentioned various dimensions feature
Vector is levied, the various dimensions feature includes: brightness, color, direction, texture and Information Entropy Features, the brightness, color, direction
Mean value, variance and corresponding High Order Moment is respectively adopted as feature vector with Information Entropy Features, the textural characteristics are adopted respectively
Use mean value, energy, consistency, inverse difference moment, contrast and correlation as feature vector, all various dimensions of the rail image
Feature vector constitutes the various dimensions set of eigenvectors of the rail image.
Further, the method further include:
Priori knowledge library is constructed, which includes: memory sample and mnemonic symbol corresponding with memory sample, institute
Stating memory sample includes flawless rail target image and defective rail target image, and the mnemonic symbol includes each
Remember the feature vector of sample.
Further, described that the various dimensions feature of shown rail image is joined by neuron interaction associative network
Want to analyze, generate association's vector, comprising:
The green neuron interaction associative network with associative ability is constructed, all various dimensions of the rail image are special
Sign vector is input to the green neuron interaction associative network, and the green neuron interaction associative network is by the multidimensional of input
Feature vector is spent as excitation vector, and variation association analysis is carried out to the excitation vector, obtains associating vector accordingly, it is described
Green neuron interaction associative network realizes the excitation vector and the variation of merging for associating vector using genetic principle, becomes
Different mode is that segmentation intersects and segmentation makes a variation.
Further, described that matching mapping processing is carried out to association's vector in priori knowledge base, it obtains described
Associate the corresponding association's sample of vector, comprising:
Association's vector is in optimized selection using preferred function, obtains optimization association vector, the preferred function H
() such as following formula:
In formula, X is excitation vector, YiFor i-th of the association's vector generated to excitation vector X by variation association analysis；
Feature vector by optimization association vector respectively with memory sample each in priori knowledge library carries out degree of correlation matching
It calculates, obtains the maximum memory sample of the degree of correlation for associating vector with the optimization, the maximum memory sample of the degree of correlation is made
For the association's sample for associating DUAL PROBLEMS OF VECTOR MAPPING with the optimization；
If optimization association vector is ui, the feature vector for remembering sample is ri, the association vector uiWith described eigenvector
riDimension be n, then the association vector uiWith described eigenvector riBetween degree of correlation matching primitives result are as follows:
In formula, ΣijFor cosine similarity correction term,WithRespectively apart from matching degree and improvement cosine matching degree, Uik
Indicate association vector uiK tie up component, rjkIndicate feature vector riK tie up component；
The association vector uiWith described eigenvector riBetween mapping associate factor aijCalculation formula are as follows:
FaA function is represented,WithFor function FaParameter；
Factor a is associated into maximum mappingijCorresponding memory sample is as the association for associating DUAL PROBLEMS OF VECTOR MAPPING with the optimization
Sample.
Further, described that analysis decision is carried out to association's sample using decision function, association is obtained as a result, root
Determine whether rail is defective according to association's result, comprising:
For association's vector after the optimization of each category feature of rail image, obtained by degree of correlation matching mapping calculation
Association's sample is considered what the optimization association sample ballot generated by the corresponding association's sample of association's vector, will
The ballot specific gravity that the mapping association factor between optimization association's vector and association's sample is voted as this；Reservation obtains
Obtain association's sample of association's vector ballot of all categories feature, and the object as subsequent analysis decision；
Assuming that for a certain input picture, comentropy, color, direction, texture and brightness association's vector all
With same association's sample is mapped, the corresponding mapping association factor is respectively ae、ac、ad、atAnd ag, then association's sample is obtained
Voting stake vector are as follows:
V=(ae,ac,ad,at,ag)T
After carrying out preliminary screening to multiple association's samples that multiple rail images association matching obtains, further according to each association
The voting stake vector of sample carries out analysis decision to multiple association's samples, and the optimal association's sample of final output is defeated as associating
Out, determine whether rail is defective according to label information in optimal association's sample.
As can be seen from the technical scheme provided by the above-mentioned embodiment of the present invention, the embodiment of the present invention is in research human vision
In mechanism of perception on the basis of Selective attention mechanism, for steel rail defect test problems, the association's machine for establishing stratification is inquired into
Simulation.The green neuron interaction associative network with associative ability is proposed, and is based on mankind's visual cortex on this basis
In WHAT access construct associative mechanisms model, to corticocerebral association function carry out reasonable assumption be abstracted as association produce
Raw, association matching and comprehensive analysis hierarchical model.Compared with traditional steel rail defect detection method, associative mechanisms are significantly improved
Detection performance of the system for steel rail defect.
The additional aspect of the present invention and advantage will be set forth in part in the description, these will become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, required use in being described below to embodiment
Attached drawing be briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for this
For the those of ordinary skill of field, without creative efforts, it can also be obtained according to these attached drawings others
Attached drawing.
Fig. 1 is a kind of processing stream of the steel rail defect measurement method of view-based access control model associative mechanisms provided in an embodiment of the present invention
Cheng Tu；
Fig. 2 is a kind of building process schematic diagram of associative mechanisms model provided in an embodiment of the present invention；
Fig. 3 is a kind of structural schematic diagram of the associative mechanisms model of stratification provided in an embodiment of the present invention.
Fig. 4 is a kind of operation principle schematic diagram of green neuron interaction associative network provided in an embodiment of the present invention；
Fig. 5 is a kind of structural schematic diagram of green neuron interaction associative network provided in an embodiment of the present invention；
Fig. 6 be it is provided in an embodiment of the present invention it is a kind of under dim weather using associative mechanisms model to the defect of rail into
Row associates matched result schematic diagram；
Fig. 7 is a kind of schematic diagram of ROC curve provided in an embodiment of the present invention；
Fig. 8 is the schematic diagram of another ROC curve provided in an embodiment of the present invention.
Specific embodiment
Embodiments of the present invention are described below in detail, the example of the embodiment is shown in the accompanying drawings, wherein from beginning
Same or similar element or element with the same or similar functions are indicated to same or similar label eventually.Below by ginseng
The embodiment for examining attached drawing description is exemplary, and for explaining only the invention, and is not construed as limiting the claims.
Those skilled in the art of the present technique are appreciated that unless expressly stated, singular " one " used herein, " one
It is a ", " described " and "the" may also comprise plural form.It is to be further understood that being arranged used in specification of the invention
Diction " comprising " refer to that there are the feature, integer, step, operation, element and/or component, but it is not excluded that in the presence of or addition
Other one or more features, integer, step, operation, element, component and/or their group.It should be understood that when we claim member
Part is " connected " or when " coupled " to another element, it can be directly connected or coupled to other elements, or there may also be
Intermediary element.In addition, " connection " used herein or " coupling " may include being wirelessly connected or coupling.Wording used herein
"and/or" includes one or more associated any cells for listing item and all combinations.
Those skilled in the art of the present technique are appreciated that unless otherwise defined, all terms used herein (including technology art
Language and scientific term) there is meaning identical with the general understanding of those of ordinary skill in fields of the present invention.Should also
Understand, those terms such as defined in the general dictionary, which should be understood that, to be had and the meaning in the context of the prior art
The consistent meaning of justice, and unless defined as here, it will not be explained in an idealized or overly formal meaning.
In order to facilitate understanding of embodiments of the present invention, it is done by taking several specific embodiments as an example below in conjunction with attached drawing further
Explanation, and each embodiment does not constitute the restriction to the embodiment of the present invention.
Embodiment one
A kind of process flow diagram of the steel rail defect measurement method of view-based access control model associative mechanisms provided in an embodiment of the present invention
As shown in Figure 1, including following processing step:
Step S110, the various dimensions feature in rail image is extracted.
Mankind's visual cortex is mainly used for brain there are a WHAT access and identifies, associates to object.The present invention is real
It applies example and reasonable assumption, Fig. 2, which are a kind of associative mechanisms mould provided in an embodiment of the present invention, to be made that the organizational process of WHAT access
The building process schematic diagram of type, Fig. 3 are a kind of structural representation of the associative mechanisms model of stratification provided in an embodiment of the present invention
Figure.It is successively abstracted the various dimensions feature for extracting input picture in the V1-V4 layer of visual cortex, and carries out association's production in cerebral cortex
Raw, association's matching and comprehensive analysis finally determine reasonable association's output.
According to above-mentioned associative process, associative mechanisms model shown in Fig. 2 is broadly divided into feature extraction, association generates, connection
Think matching and comprehensive analysis Four processes:
Characteristic extraction procedure: the rail image of defect to be identified, the V1- in analog vision system WHAT access are obtained first
V4 layers, convolutional neural networks are similar to, every layer of neural network can all extract the different feature of picture.
The various dimensions feature in rail image is extracted by the V1-V4 layer in above-mentioned WHAT access, is based on above-mentioned various dimensions
Feature generates various dimensions feature vector.The various dimensions feature is regarded using five early stages typical in the Saliency model of Itti
Feel feature, is respectively as follows: brightness, color, direction, texture and Information Entropy Features, the brightness, color, direction and Information Entropy Features
Mean value, variance and corresponding High Order Moment is respectively adopted as feature vector, the textural characteristics be respectively adopted mean value, energy,
Consistency, inverse difference moment, contrast and correlation are constituted as feature vector, all various dimensions feature vectors of the rail image
The various dimensions set of eigenvectors of the rail image.
Priori knowledge library is constructed, which includes: memory sample and mnemonic symbol corresponding with memory sample, institute
Stating memory sample includes flawless rail target image and defective rail target image, and the mnemonic symbol includes each
Remember the feature vector of sample.
Step S120, association point is carried out to the various dimensions feature of shown rail image by neuron interaction associative network
Analysis generates association's vector.
Association's generation process is divided into two stages, and building first has the interactive associative network of the green neuron of associative ability
(Green Neurons-interactive Associative Network, GNAN), which is entire associative mechanisms
The core of model, GNAN network carry out appropriateness variation association analysis to the various dimensions feature vector of input, generate corresponding association
Vector obtains reliably associating result.For the diversity for guaranteeing associative process, biological heredity variation machine is introduced the stage in association
System simulates the variation phenomenon of thinking in associative process by the intersection and mutation process of chromosome in biological heredity, and will
Excitation ensure that the relevance of associative process as parent chromosome.The embodiment of the present invention using the input vector of GNAN network as
Excitation vector using the output vector of the GNAN network as association's vector, and realizes excitation vector and association using genetic principle
The fusion of vector makes a variation, and variation mode is segmentation intersection and segmentation variation.Segmentation intersection is carried out using binary-coded mode
It makes a variation with segmentation.One excitation vector can produce multiple corresponding association's vectors.
Step S130, matching mapping processing is carried out to association's vector in priori knowledge base, obtains association's vector pair
The association's sample answered.
The result of variation association is analyzed and processed in association's matching process, and utilizes the priori knowledge library constructed in advance
Matching treatment is carried out to association's result, to map out reliable related association's sample according to huge priori knowledge library.The mistake
Journey is divided into two stages: input the preferred stage using building preferred function to variation association it is a large amount of export optimize choosing
It selects, retains reliable variation association vector；Phase is searched in priori knowledge base using optimal association's vector in matching mapping phase
The high association's sample of pass degree, and association's factor of each pair of mapping is recorded, it is that subsequent comprehensive analysis finally determines that association's result is defeated
It lays the foundation out.
Preferred function can be measured by association's degree of correlation: association's output is always distributed across distance input on feature space
In a certain range of space of vector, and certain regularity is presented in the directional spreding of output vector.It is possible thereby to determine preferred letter
Number H () such as following formula:
In formula, X is excitation vector, YiFor i-th of the association's vector generated to excitation vector X by variation association analysis, h
It indicates a kind of function, is similar to F (x).
Match mapping phase utilize through preferred function preferably go out association's vector respectively with each memory in priori knowledge library
The feature vector of sample carries out degree of correlation matching primitives, and generates association's degree factor of each pair of mapping.For association vector ui
With feature vector ri, their dimension is n, then matching primitives result such as following formula:
In formula, ΣijFor cosine similarity correction term,WithRespectively apart from matching degree and improvement cosine matching degree, Uik
Indicate association vector uiK tie up component, rjkIndicate feature vector riK tie up component.
The association vector uiWith described eigenvector riBetween mapping associate factor aijCalculation formula are as follows::
FaA function is represented,WithFor function FaParameter；
Factor a is associated into maximum mappingijCorresponding memory sample is as the association for associating DUAL PROBLEMS OF VECTOR MAPPING with the optimization
Sample.
Step S140, analysis decision is carried out to association's sample using decision function, obtains association as a result, according to the association
As a result determine whether rail is defective.
In the comprehensive analysis stage, brain can carry out comprehensive analysis measurement to above-mentioned association's sample, so that analysis decision most closes
Association's sample of reason is as final association's result.Comprehensive analysis process in associative mechanisms model is divided into voting mechanism, analysis
Two processes of decision.Voting process are as follows: for each category feature of input picture, by association's variation and preferably generation later
Association's vector of the category feature, each association's vector can match one memory sample of mapping as connection after association matches
Think sample, this matching mapping process is the single ballot completed to association's sample, and specific gravity of voting is by matching mapping process
The mapping association factor determine, calculate directly to simplify using the ballot specific gravity voted as this of mapping association's factor.Ballot
Mechanism thinks that obtaining association's sample of association's vector ballot of all categories feature is reasonable association's sample, be can be used as subsequent
The object of analysis decision.Association's sample that unanimous vote cannot be obtained will be excluded all ballot specific gravity are all very big.
Assuming that for a certain input picture, comentropy, color, direction, texture and brightness association's Vectors matching
To same association's sample, the corresponding mapping association factor is respectively a for mappinge、ac、ad、atAnd ag, then association's sample can be obtained
Voting stake vector are as follows:
V=(ae,ac,ad,at,ag)T (4)
After carrying out preliminary screening to multiple association's samples that multiple rail images association matching obtains, model can be according to each
The voting stake vector for associating sample carries out analysis decision to multiple association's samples, and the optimal association's sample of final output is as connection
Want to export, completes associative process.
Analysis decision process can be stated with decision function:
In formula, ViThe voting stake vector of sample is associated for i-th, S is decision value.Therefore, decision is most by analysis
Association's output eventually should meet optimizing decision, as a result:
Optimal solution V*As optimal voting stake vector, corresponding optimal association's sample is final associative bond
Fruit.Include the feature and label information of image in above-mentioned optimal association's sample, the label information indicate rail it is defective or
Therefore rail zero defect may determine that in rail image according to label information in above-mentioned optimal association's sample either with or without scarce
It falls into.
Fig. 4 is a kind of operation principle schematic diagram of green neuron interaction associative network provided in an embodiment of the present invention, green
Color neuron interaction associative network (GNAN) can effectively realize association function, be part most crucial in associative mechanisms model,
Be divided into association's mapping reconciliation association two stages of mapping: input feature value is mapped to by association's mapping phase under certain rule
Higher-dimension is associated in space, completes by the heteroassociative process of Partial Feature association complete characterization, in solution association mapping phase to higher-dimension
Associate space and carry out dimensionality reduction compression de-redundancy, realizes and efficiently exported from association.Simultaneously by establishing nerve in associative process
The reciprocation of adjustment function and neuron in nerve interaction function mechanism simulation biological neural signal conductive process, makes association
Network is more in line with biological nervous system mechanism, has faster convergence rate.
Fig. 5 is a kind of structural schematic diagram of green neuron interaction associative network provided in an embodiment of the present invention, such as Fig. 5 institute
Show, GNAN network is made of two process connections, wherein input feature value is projected higher-dimension association sky by association's mapping process
Between in, realize feature appropriateness association, solution association mapping process then reasonable drawing higher-dimension association space nonlinear principal component eliminate
Redundancy generates network-evaluated output vector, simplifies following model calculating process.GNAN network includes 5 node layers altogether, can be used
Symmetrical structure, can also use unsymmetric structure, and the output vector dimension of network is equal to input vector dimension.Wherein in Fig. 5
Dotted portion represents the neural reciprocation of neighborhood of nodes in same layer, rather than the physical structure of network.
Neural interaction mechanism: in traditional neural metanetwork, neuron node is uniquely determined by threshold value and activation primitive, and
And keep permanent if neuron node parameters after threshold value and activation primitive are selected in neural metwork training and use process
It is fixed, it is clear that the working condition of this and neuron models in practical human brain is not inconsistent.Actual human brain neuron models are worked
Cheng Zhong, when neuro-transmission signal transmits between adjacent neurons node, the state of neuron node is reciprocal effect most end form
At new steady s tate.Neuro-transmission signal is inputted for some, when two adjacent neuron nodes while excitation time, two knots
State between point is mutually promoted, and nervous excitation degree increases simultaneously；When two adjacent neuron nodes simultaneously inhibit when, the two it
Between state mutually inhibit, nervous excitation degree simultaneously declines；When two inhibition of excitement one of adjacent node one, between the two
Reciprocation promote originally excited neuron node by a degree of inhibition, and the neuron node inhibited originally is emerging
Degree of putting forth energy can moderately increase.For the reciprocation of adjacent neurons node in analog neuron signal transduction process, the present invention is implemented
Example proposes the concept of green exchange neuron associative network, and the nervous excitation factor, and benefit are introduced in neuron node model
The interactive process of neuron node state is realized with nerve modulation function.
Algorithm for training network: GNAN network is divided into two stages of association's mapping process and de-mapping process, therefore in order to mention
The rate of convergence of the high entire green neuron Internet and the robustness of neural network, can use input sample collection
Principal Component Analysis extracts the independent pivot ingredient of result sample so that it is determined that higher-dimension associates the size and higher-dimension sample in space out
Value, this makes it possible to association's mapping layer is conciliate association's mapping layer as two sub-neural networks individually to be trained.Every height
When neural network is individually trained, training process has adjusts process twice: positive interactive process and back-propagation process.Forward direction interaction
Process refers to send out between neuron node adjacent during the forward conduction in nerve signal from input layer to output layer
Raw nerve reciprocation causes the excitatory state of neuron node to occur to change immediately, which follows " green " principle；Reversely
Communication process refers to the back-propagation process from output layer to input layer, guarantees the empirical risk minimization of neural network.
Embodiment two
Model verifying
Fig. 6 is to carry out the matched result of association using defect of the associative mechanisms model to rail under dim weather to illustrate
Figure.Fig. 6 (a) is original image, and Fig. 6 (b) and Fig. 6 (c) are the association's sample and association's decision value of associative mechanisms model.From Fig. 6
As can be seen that associative mechanisms model effectively can carry out association's matching to steel rail defect, this associative ability will be remarkably reinforced
To the recognition performance of steel rail defect.
Performance comparison analysis
In order to test the performance of associative mechanisms model Green neuron interaction associative network, the present invention chooses BP nerve net
Network, Hopifield neural network, Bidirectional Associative Memory Neural Networks have carried out comparative experiments.Its Green neuron interaction connection
Think the input and output layer of network and BP neural network, the node of hopifield neural network and Bidirectional Associative Memory Neural Networks
Number is the dimension of sample, and BP neural network is kept the consistency with green neuron alternating network structure, instruction by 3 hidden layers
Practicing error is 0.01.Training set and test set are chosen respectively meets actual association's sample set, and training set includes 800 samples altogether
This, training set includes 200 samples altogether.Experiment porch is Intel Duo I5 double-core CPU, dominant frequency 2.4GHZ.It tests from training
Three time, testing time, memory accuracy, association's accuracy aspects have carried out comparative analysis.Wherein remember accuracy and connection
Think that accuracy is respectively defined as memory error and associates the ratio of the error sample number less than 0.1 and test sample sum, memory
Error deltarWith association's error deltaaIt is defined as follows:
In formula, (xk,yk) it is test sample, zkFor reality output.Experimental result such as table 1.
The recognition result of 1 heterogeneous networks model of table
| Title | Green neuron interaction associative network | BP neural network | Hopifield neural network | Bi-directional associative memory network |
| Training time (s) | 47 | 180 | 183 | 312 |
| Testing time (ms) | 83 | 85 | 39 | 37 |
| Remember accuracy (%) | 62 | 33 | 78 | 86 |
| Associate accuracy (%) | 84 | 31 | 63 | 74 |
As can be seen from Table 1: (1) training speed of green neuron interaction associative network is substantially better than BP neural network,
This is because accelerating the convergence rate of weight and Threshold-training before increasing after interactive to neuron；(2) Hopifield nerve
Network and bi-directional associative memory network are more biased to memory capability, and green neuron interaction associative network has best association
Ability, while keeping certain memory capability.
For further more green neuron interaction associative network, Hopifield neural network and two-way association note
Recall the performance of network (BAM), Hopifield neural network, BAM network are replaced green neuron interaction association respectively by the present invention
Network constitutes two new associative mechanisms models, and association's performance comparison of steel rail defect is carried out with the model of the embodiment of the present invention
Experiment, and (receiver operating characteristic curve, Receiver Operating Characteristics are bent by the ROC of rendering model
Line) curve is as shown in Figure 7.
From figure 7 it can be seen that the defect of (1) for rail, GNAN has best association's recognition capability, hence it is evident that is better than
Hopifield neural network and BAM；(2) when being optimal performance, the false recognition rate of GNAN only compares 5% or so
Hopifield neural network is big, but gap is little.Therefore compared with traditional neural network, green neuron interaction associative network
There is apparent advantage in terms of training speed and associative ability.
Comprehensive Experiment: in order to test associative mechanisms model for the recognition performance of steel rail defect, associative mechanisms are respectively adopted
Model and method based on Countour-Let transformation and SVM, based on gradient direction figure and sparse expression algorithm for
Steel rail defect test sample collection carries out identification experiment, and it is as shown in Figure 8 to draw ROC curve.
As can be seen from Figure 8: (1) for the defect of rail, associative mechanisms model has best association's recognition capability, reaches
To 88.6%；(2) when being optimal performance, the false recognition rate of associative mechanisms model is minimum and is maintained at 5% or so.Therefore with tradition
Steel rail defect detection method compare, associative mechanisms model steel rail defect context of detection have apparent advantage.
In conclusion the embodiment of the present invention study human visual perception mechanism in Selective attention mechanism on the basis of, needle
To steel rail defect test problems, the associative mechanisms model for establishing stratification is inquired into.Propose the green mind with associative ability
Through member interaction associative network, and associative mechanisms model is constructed based on the WHAT access in mankind's visual cortex on this basis, it is right
Corticocerebral association function carries out reasonable assumption and is abstracted as the hierarchical model that association generates, associates matching and comprehensive analysis.
Compared with traditional steel rail defect detection method, associative mechanisms significantly improve system for the detection performance of steel rail defect.
The visual perception mechanism that the embodiment of the present invention uses for reference the mankind establishes the good associative mechanisms model of robustness and uses
In the defects detection of rail.The WHAT pathways for vision that the functions such as Object identifying, association are used in mankind's visual cortex is closed
Reason is assumed to be abstracted, and the associative mechanisms model of stratification is constructed, to realize the reliable recognition of steel rail defect；Proposing has
The green neuron interaction associative network of outstanding associative ability, effectively simulates the associative process of brain, while introducing biological letter
Neural interaction mechanism in number conductive process improves the associative ability of associative network.
Those of ordinary skill in the art will appreciate that: attached drawing is the schematic diagram of one embodiment, module in attached drawing or
Process is not necessarily implemented necessary to the present invention.
As seen through the above description of the embodiments, those skilled in the art can be understood that the present invention can
It realizes by means of software and necessary general hardware platform.Based on this understanding, technical solution of the present invention essence
On in other words the part that contributes to existing technology can be embodied in the form of software products, the computer software product
It can store in storage medium, such as ROM/RAM, magnetic disk, CD, including some instructions are used so that a computer equipment
(can be personal computer, server or the network equipment etc.) executes the certain of each embodiment or embodiment of the invention
Method described in part.
All the embodiments in this specification are described in a progressive manner, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for device or
For system embodiment, since it is substantially similar to the method embodiment, so describing fairly simple, related place is referring to method
The part of embodiment illustrates.Apparatus and system embodiment described above is only schematical, wherein the conduct
The unit of separate part description may or may not be physically separated, component shown as a unit can be or
Person may not be physical unit, it can and it is in one place, or may be distributed over multiple network units.It can root
According to actual need that some or all of the modules therein is selected to achieve the purpose of the solution of this embodiment.Ordinary skill
Personnel can understand and implement without creative efforts.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
In the technical scope disclosed by the present invention, any changes or substitutions that can be easily thought of by anyone skilled in the art,
It should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with scope of protection of the claims
Subject to.
Claims (6)
1. a kind of steel rail defect measurement method of view-based access control model associative mechanisms characterized by comprising
Extract the various dimensions feature in rail image；
Association analysis is carried out to the various dimensions feature of shown rail image by neuron interaction associative network, generate association to
Amount；
Matching mapping processing is carried out to association's vector in priori knowledge base, obtains the corresponding association's sample of association's vector
This；
Analysis decision is carried out to association's sample using decision function, obtains association as a result, determining according to association's result
Whether rail is defective.
2. the method according to claim 1, wherein the various dimensions feature in the extraction rail image, packet
It includes:
The rail image of defect to be identified is obtained, the V1-V4 layer in analog vision system WHAT access passes through the WHAT access
In V1-V4 layer extract the various dimensions feature in the rail image, based on above-mentioned various dimensions feature generate various dimensions feature to
Amount, the various dimensions feature includes: brightness, color, direction, texture and Information Entropy Features, the brightness, color, direction and letter
Mean value, variance and corresponding High Order Moment is respectively adopted as feature vector in breath entropy feature, and the textural characteristics are respectively adopted
Value, energy, consistency, inverse difference moment, contrast and correlation are as feature vector, all various dimensions features of the rail image
Vector constitutes the various dimensions set of eigenvectors of the rail image.
3. according to the method described in claim 2, it is characterized in that, the method further include:
Priori knowledge library is constructed, which includes: memory sample and mnemonic symbol corresponding with memory sample, the note
Recalling sample includes flawless rail target image and defective rail target image, and the mnemonic symbol includes each memory
The feature vector of sample.
4. according to the method described in claim 3, it is characterized in that, described interact associative network to shown steel by neuron
The various dimensions feature of rail image carries out association analysis, generates association's vector, comprising:
Construct have associative ability green neuron interaction associative network, by all various dimensions features of the rail image to
Amount is input to the green neuron interaction associative network, and the green neuron interaction associative network is special by the various dimensions of input
Vector is levied as excitation vector, variation association analysis is carried out to the excitation vector, obtains associating vector, the green accordingly
Neuron interaction associative network realizes that the excitation vector associates vector and merge variation, variation side with described using genetic principle
Formula is that segmentation intersects and segmentation makes a variation.
5. according to the method described in claim 4, it is characterized in that, it is described in priori knowledge base to association's vector into
Row matching mapping processing, obtains the corresponding association's sample of association's vector, comprising:
Association's vector is in optimized selection using preferred function, obtains optimization association vector, the preferred function H ()
Such as following formula:
In formula, X is excitation vector, YiFor i-th of the association's vector generated to excitation vector X by variation association analysis；
Feature vector by optimization association vector respectively with memory sample each in priori knowledge library carries out degree of correlation matching primitives,
Obtain with it is described optimization association vector the maximum memory sample of the degree of correlation, using the maximum memory sample of the degree of correlation as with institute
State association's sample of optimization association DUAL PROBLEMS OF VECTOR MAPPING；
If optimization association vector is ui, the feature vector for remembering sample is ri, the association vector uiWith described eigenvector ri's
Dimension is n, then the association vector uiWith described eigenvector riBetween degree of correlation matching primitives result are as follows:
In formula, ΣijFor cosine similarity correction term,WithRespectively apart from matching degree and improvement cosine matching degree, UikIt indicates
Associate vector uiK tie up component, rjkIndicate feature vector riK tie up component；
The association vector uiWith described eigenvector riBetween mapping associate factor aijCalculation formula are as follows:
FaA function is represented,WithFor function FaParameter；
Factor a is associated into maximum mappingijCorresponding memory sample is as the association's sample for associating DUAL PROBLEMS OF VECTOR MAPPING with the optimization.
6. according to the method described in claim 5, it is characterized in that, described carries out association's sample using decision function
Analysis decision obtains association as a result, determining whether rail is defective according to association's result, comprising:
For association's vector after the optimization of each category feature of rail image, obtained by degree of correlation matching mapping calculation described
Associate the corresponding association's sample of vector, association's sample is considered what the optimization association sample ballot generated, it will be described
The ballot specific gravity that the mapping association factor between optimization association's vector and association's sample is voted as this；Retain and obtains institute
There are association's sample of association's vector ballot of category feature, and the object as subsequent analysis decision；
Assuming that for a certain input picture, comentropy, color, direction, texture and brightness association's vector all match and reflect
Same association's sample is penetrated, the corresponding mapping association factor is respectively ae、ac、ad、atAnd ag, then the ballot of association's sample is obtained
Weight vector are as follows:
V=(ae,ac,ad,at,ag)T
After carrying out preliminary screening to multiple association's samples that multiple rail images association matching obtains, further according to each association's sample
Voting stake vector analysis decision carried out to multiple association's samples, the optimal association's sample of final output as association's output,
Determine whether rail is defective according to label information in optimal association's sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810797993.8A CN109166099A (en) | 2018-07-19 | 2018-07-19 | The steel rail defect measurement method of view-based access control model associative mechanisms |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810797993.8A CN109166099A (en) | 2018-07-19 | 2018-07-19 | The steel rail defect measurement method of view-based access control model associative mechanisms |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109166099A true CN109166099A (en) | 2019-01-08 |
Family
ID=64897910
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810797993.8A Pending CN109166099A (en) | 2018-07-19 | 2018-07-19 | The steel rail defect measurement method of view-based access control model associative mechanisms |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109166099A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112001885A (en) * | 2020-07-14 | 2020-11-27 | 武汉理工大学 | PCB fault detection method, storage medium and system based on machine vision |
| CN112633360A (en) * | 2020-12-18 | 2021-04-09 | 中国地质大学（武汉） | Classification method based on cerebral cortex learning mode |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5777891A (en) * | 1991-05-07 | 1998-07-07 | Dapco Industries Inc | Method for real-time ultrasonic testing system |
| CN107727660A (en) * | 2017-10-13 | 2018-02-23 | 浙江树人学院 | Rail surface defects detecting system and method based on machine vision and impulse eddy current |
| CN108921846A (en) * | 2018-07-17 | 2018-11-30 | 北京航空航天大学 | A kind of rail tread defect identification method combined based on gray level image and depth image |
-
2018
- 2018-07-19 CN CN201810797993.8A patent/CN109166099A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5777891A (en) * | 1991-05-07 | 1998-07-07 | Dapco Industries Inc | Method for real-time ultrasonic testing system |
| CN107727660A (en) * | 2017-10-13 | 2018-02-23 | 浙江树人学院 | Rail surface defects detecting system and method based on machine vision and impulse eddy current |
| CN108921846A (en) * | 2018-07-17 | 2018-11-30 | 北京航空航天大学 | A kind of rail tread defect identification method combined based on gray level image and depth image |
Non-Patent Citations (1)
| Title |
|---|
| JIA, DONGYAO 等: "Recognition method based on Green Associative Mechanism for weak contrast vehicle targets", 《NEUROCOMPUTING》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112001885A (en) * | 2020-07-14 | 2020-11-27 | 武汉理工大学 | PCB fault detection method, storage medium and system based on machine vision |
| CN112633360A (en) * | 2020-12-18 | 2021-04-09 | 中国地质大学（武汉） | Classification method based on cerebral cortex learning mode |
| CN112633360B (en) * | 2020-12-18 | 2024-04-05 | 中国地质大学（武汉） | Classification method based on cerebral cortex learning mode |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108537743B (en) | Face image enhancement method based on generation countermeasure network | |
| CN113378632B (en) | Pseudo-label optimization-based unsupervised domain adaptive pedestrian re-identification method | |
| CN104866829B (en) | A kind of across age face verification method based on feature learning | |
| CN107145842B (en) | Face recognition method combining LBP characteristic graph and convolutional neural network | |
| CN106650806B (en) | A kind of cooperating type depth net model methodology for pedestrian detection | |
| CN109934195A (en) | A kind of anti-spoofing three-dimensional face identification method based on information fusion | |
| CN110502988A (en) | Group positioning and anomaly detection method in video | |
| CN111339988B (en) | Video face recognition method based on dynamic interval loss function and probability characteristic | |
| CN109711254A (en) | The image processing method and device of network are generated based on confrontation | |
| CN107463920A (en) | A kind of face identification method for eliminating partial occlusion thing and influenceing | |
| CN107194341A (en) | The many convolution neural network fusion face identification methods of Maxout and system | |
| CN109446991A (en) | Gait recognition method based on global and local Fusion Features | |
| CN105574510A (en) | Gait identification method and device | |
| CN110084149B (en) | Face verification method based on hard sample quadruple dynamic boundary loss function | |
| CN108875907B (en) | Fingerprint identification method and device based on deep learning | |
| CN108520213B (en) | Face beauty prediction method based on multi-scale depth | |
| CN110348416A (en) | Multi-task face recognition method based on multi-scale feature fusion convolutional neural network | |
| CN103605972A (en) | Non-restricted environment face verification method based on block depth neural network | |
| CN109614853A (en) | It is a kind of based on body structure divide bilinearity pedestrian identify network establishing method again | |
| CN112580445B (en) | Human body gait image visual angle conversion method based on generation of confrontation network | |
| CN106778512A (en) | Face identification method under the conditions of a kind of unrestricted based on LBP and depth school | |
| CN114511912B (en) | Cross-library micro-expression recognition method and device based on double-flow convolutional neural network | |
| CN108960142B (en) | Pedestrian re-identification method based on global feature loss function | |
| CN110969073B (en) | Facial expression recognition method based on feature fusion and BP neural network | |
| CN111881716A (en) | Pedestrian re-identification method based on multi-view-angle generation countermeasure 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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190108 |