CN104200095B - System and method for expanding sensing channel on basis of perceiving evolutionary neural network - Google Patents
System and method for expanding sensing channel on basis of perceiving evolutionary neural network Download PDFInfo
- Publication number
- CN104200095B CN104200095B CN201410432884.8A CN201410432884A CN104200095B CN 104200095 B CN104200095 B CN 104200095B CN 201410432884 A CN201410432884 A CN 201410432884A CN 104200095 B CN104200095 B CN 104200095B
- Authority
- CN
- China
- Prior art keywords
- formula
- prototype
- represent
- signal
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Abstract
The invention provides a system and method for expanding a sensing channel on the basis of a perceiving evolutionary neural network. The system comprises a data acquisition system which is connected with a control system. A data perceiving and receiving module based on the perceiving evolutionary neural network is arranged in the control system. The system is combined with the method, so that the defects that channels for data acquisition cannot be upgraded or expanded in the prior art, application is very limited and an intellectualized acquisition function is not provided are effectively overcome.
Description
Technical field
The invention belongs to the technical field that data perception is received, and in particular to a kind of based on the biography for perceiving Evolutionary Neural Network
The expansible system and method for sense channel.
Background technology
Current image, video also have the signal of communication of some to be transferred in control system by data collecting system, this
The data acquisition modes of the communication signal data of sample are only to maintain the true colours of its original signal data, and data cannot be adopted
The passage of collection is upgraded or is extended, using very limited and do not possess intelligentized acquisition function.
The content of the invention
It is an object of the invention to provide it is a kind of based on the expansible system and method for sensing channel for perceiving Evolutionary Neural Network,
Including data collecting system, described data collecting system is connected with control system, carries in described control system and is based on
Perceive the data perception receiver module of Evolutionary Neural Network.And with reference to its method can be prevented effectively from it is of the prior art cannot logarithm
Upgraded or extended according to the passage of collection, using defect that is very limited and not possessing intelligentized acquisition function.
In order to overcome deficiency of the prior art, the invention provides a kind of based on the sensing letter for perceiving Evolutionary Neural Network
The solution of the expansible system and method in road, it is specific as follows:
A kind of expansible system of sensing channel based on perception Evolutionary Neural Network, including data collecting system 1, it is described
Data collecting system 1 is connected with control system 2, with based on the number for perceiving Evolutionary Neural Network in described control system 2
According to perception receiver module 3.
Described control system 2 perceives neuron according to original n of original neutral net, through a period of time
After perception, under conditions of increasing m new perception neuron appearance newly, arrange corresponding to original in the storage region of control system 2
The original lower dimensional space of some n perception neurons, described original lower dimensional space is Sl=Rn, and in control system 2
New higher dimensional space corresponding to m new perception neuron is set in storage region, and described new higher dimensional space is Sh=Rn +m, and the prototype obtained after perceiving and compressing is stored in set P, uses PiRepresent i-th prototype in P, described set P
Be also in the storage region of control system 2 arrange a memory space, n and m be natural number, RnRepresent n dimension real number to
Quantity space, Rn+mRepresent the real number vector space of n+m dimensions.
A kind of described method based on the expansible system of sensing channel for perceiving Evolutionary Neural Network, step is as follows:
Step 1:The image or sample of signal as video that first data collecting system carrys out collection is sent to control
In system 2, then control system 2 starts based on the expansible system module 3 of the sensing channel for perceiving Evolutionary Neural Network first most
The two sample of signal x for first collecting1And x2Each self-corresponding n is constructed respectively ties up prototype P1Prototype P is tieed up with the 2nd n2,
First n ties up prototype P1Prototype P is tieed up with the 2nd n2In being stored in set P, the sensing channel for perceiving Evolutionary Neural Network is additionally based on
3 images for sending of expansible system module or each sample of signal as video are configured to signal vector, described
A n dimension prototype P1=(x1,1,x1,2,...,x1,n), the 2nd n dimension prototypes P2=(x2,1,x2,2,...,x2,n), if at first
The sample of signal later of two sample of signal for collecting is n dimensional signals vector, then the form of described signal vector is xj
=(xj,1,xj,2,...,xj,n)∈Sl, perform based on the method for the expansible system of sensing channel for perceiving Evolutionary Neural Network
First stage, the described first stage starts to perform from step 8.In the event of the new perception neurons of m, then described letter
The form of number vector is xj=(xj,1,xj,2,...,xj,n,...,xj,n+m)∈Sh, xjRepresent the signal of j-th sample of signal to
Amount, j is natural number, xj,1Represent that first perceives input data of the neuron to the one-dimensional vector of j-th sample of signal, xj,2Table
Show that second perceives input data of the neuron to the one-dimensional vector of j-th sample of signal, xj,nRepresent n-th perception neuron
Input data to the one-dimensional vector of j-th sample of signal, xj,n+mRepresent that the n-th+m perceives neuron to j-th sample of signal
One-dimensional vector input data, ShRepresent that the vector that the collected corresponding signal vector of sample of signal is collectively constituted is empty
Between, h, l are natural number, and then execution step 2 realizes that second stage, second stage are completed after performing to the operation of step 7
Method based on the expansible system of sensing channel for perceiving Evolutionary Neural Network;
Step 2:Obtain sample of signal xj=(xj,1,xj,2,...,xj,n,...,xj,n+m)∈Sh.For each Pi∈ P,
If Pi∈Sl, i.e. PiIn space SlIn, by PiWith the sample of signal x of inputjFront n dimension attributes calculate Euclidean distance, it is described
N dimension attributes be SlAttribute.If Pi∈Sh, i.e. PiIn space ShIn, by PiWith the sample of signal x of inputj(n+m)
Dimension attribute calculates Euclidean distance, and described (n+m) dimension attribute is ShAttribute, by formula (1) obtain SlIn triumph prototype
Wherein, j represents natural number,Represent PiWeight vector,Represent in SlIn Euclidean distance, then by formula
And formula (3) obtains respectively S (2)hIn triumph prototypeWith the second triumph prototype
Wherein, PhExpression belongs to ShPrototype set, i.e., it is all of to belong to ShPrototype set,Represent in ShIn
Euclidean distance, be below abbreviatedFor Pc,WithIt is abbreviated as PaAnd Pb;
Step 3:If dim is (Pc)=n, i.e. Pc∈Sl, then check whether the criterion met as shown in formula (4):
Wherein,Represent PcIn SlIn similarity threshold.If formula (4) meets, prototype PcBe activated, subsequently do as
Renewal represented by formula (5), formula (6), formula (7) and formula (8):
Wh-l=xj h-l
(7)
Wherein,Represent prototype PcIn SlIn activation number of times,Represent prototype PcWeight vector,Represent SlTo
Amount attribute, i.e.,xh-lThe attribute of m new perception neuron is represented, i.e.,
Initialization in additionWithWherein,Represent prototype PcIn ShIn activation number of times,Represent PcIn ShIn
Similarity threshold.If condition (4) is unsatisfactory for, one such as formula (9), formula (10), formula (11), formula (12) and public affairs are set up
The new prototype of formula (13):
Wnew=xj
(9)
Wherein WnewThe weight vector of newly-built prototype is represented,Represent in SlIn newly-built prototype activation number of times,Represent in ShIn newly-built prototype activation number of times,Represent in SlIn newly-built prototype similarity threshold,
Represent in ShIn newly-built prototype similarity threshold.
If dim is (Pc) ≠ n, i.e. PcIt has been ShIn prototype, if meeting the condition of formula (14):
WhereinRepresent prototype PaWeight vector, i.e. xjWith PaIt is identical, PaBe activated, it is corresponding make such as formula (15) and
The renewal of formula (16):
WhereinRepresent prototype PaIn SlIn activation number of times,Represent prototype PaIn ShIn activation number of times.
If the condition of formula (14) is unsatisfactory for, the condition for meeting formula (17) is checked whether:
WhereinRepresent prototype PaWeight vector,Represent PaIn ShIn similarity threshold,Represent prototype Pb's
Weight vector,Represent PbIn ShIn similarity threshold.If the condition of formula (17) is satisfied, to PaCarry out such as formula
(18), the renewal under formula (19), formula (20) and formula (21):
Wherein,Represent prototype PaIn SlIn weight vector,Represent prototype PaIn m newly-increased perception neuron
Weight vector;If the condition of formula (17) is unsatisfactory for, using formula (9), formula (10), formula (11), formula (12)
New prototype is set up with formula (13);
Step 4:If the condition of condition (17) is satisfied, and in prototype PaWith prototype PbBetween there is no connection, then
Set up PaAnd PbBetween connection, juxtapositionDescribedRepresent prototype PaWith prototype PbBetween connection
Age variable, described age variable represents the tolerance of the connection lifetime of connection.If PaAnd PbBetween connection deposited
As long as puttingIf prototype PiIt is thus lifted to higher dimensional space or weight vectors is updated, as formula
(22) to PiAnd PjBetween age variable be updated:
Wherein,Represent PiConnected prototype set.If the age variable of a connection is more than predetermined threshold value
Agemax, the connection is just deleted;
Step 5:If a prototype does not have neighbours, then the similarity threshold definition of the prototype is as shown in formula (23):
If there are neighbours in a prototype, then the similarity threshold definition of the prototype is as shown in formula (24):
Whenever one new signal sample x of inputjWhen, update P according to formula (23) and formula (24)a, PbAnd PcSimilitude
Threshold value;
Step 6:After λ sample signal is processed by step 1- step 5, prototype cutting is performed, and increase newly at m
The target that the prototype that there is before neuron occurs is not belonging to cutting is perceived, perceives what is increased newly after neuron occurs at m
Prototype belongs to the target of cutting, and these newly-increased prototypes are organized into set Q, is first according to formula (24) and calculates all's
Average
Wherein PhExpression belongs to space ShPrototype, | Ph| represent set PhElement number,Represent prototype PiIn Sh
In activation number of times.For each prototype Qi∈ Q, if such as formula (25)
Or formula (26)
Condition be satisfied, prototype QiTo be deleted, whereinRepresent QiNeighbours' number, λ is natural number, c
For predefined parameter, span [0,1];
Step 7:When all of prototype is mapped to ShAfter, by parameterWithCarry out as described in formula (27)
Merge, the activation number of times after being merged
Right value update formula described in formula (20) and formula (21) is changed into the more new formula of formula (28):
Formula (9), formula (10), formula (11), formula (12) and formula (13) be reduced to formula (29), formula (30) and
Shown in formula (31):
Wnew=xj
(29)
Then parameterTo be left out.For follow-up signal sample xj=(xj,1,xj,2,...,xj,n,...,xj,n+m)∈Sh
Process can refer to based on perceive Evolutionary Neural Network the expansible system of sensing channel method first stage.
Step 8:S is obtained by formula (32)lIn triumph prototype
Wherein, j represents natural number,Represent PiWeight vector,Represent in SlIn Euclidean distance.If met
The condition of formula (33):
WhereinRepresent prototype PaWeight vector, i.e. xjWith PaIt is identical, PaIt is activated, makees such as formula (34) accordingly
Update:
WhereinRepresent prototype PaIn SlIn activation number of times, if the condition of formula (33) is unsatisfactory for, check whether full
The condition of sufficient formula (35):
WhereinRepresent prototype PaWeight vector,Represent PaIn SlIn similarity threshold,Represent prototype Pb's
Weight vector,Represent PbIn SlIn similarity threshold.If the condition of formula (35) is satisfied, to PaCarry out such as formula
(36), the renewal under formula (37):
Wherein,Represent prototype PbIn SlIn weight vector.If the condition of formula (35) is unsatisfactory for, using such as formula
(38), formula (39) and formula (40)
Wnew=xj
(38)
Set up new prototype;
Step 9:If the condition of condition (35) is satisfied, and in prototype PaWith prototype PbBetween there is no connection, then
Set up PaAnd PbBetween connection, juxtapositionDescribedRepresent prototype PaWith prototype PbBetween age
Variable, described age variable represents the tolerance of the connection lifetime of connection;If PaAnd PbBetween connection existed, as long as
PutIf prototype PiWeight vectors be updated, as formula (41) is to PiAnd PjBetween age variable
It is updated,
Wherein,Represent PiConnected prototype set, if one connection age variable be more than predetermined threshold value
Agemax, the connection is just deleted;
Step 10:If a prototype does not have neighbours, then the similarity threshold definition of the prototype is as shown in formula (42):
If there are neighbours in a prototype, then the similarity threshold definition of the prototype is as shown in formula (43):
Whenever one new signal sample x of inputjWhen, update P according to formula (42) and formula (43)aAnd PbSimilitude threshold
Value;
Step 11:After λ sample signal is processed by step 8- step 10, prototype cutting is performed, and is first according to public affairs
Formula (44) calculates allAverage
For each prototype Pi∈ P, if such as formula (45)
Or formula (46)
Condition be satisfied, prototype PiTo be deleted, whereinRepresent PiNeighbours' number, λ is natural number, and c is predefined
Parameter, span [0,1].
By these technical characteristics, the method for the present invention overcomes in conventional method and the passage of data acquisition cannot be risen
Level or extension, using defect that is very limited and not possessing intelligentized acquisition function, and can be logical to data acquisition
Road is upgraded or is extended, and is widely used and possesses intelligentized acquisition function.
Description of the drawings
Fig. 1 is that a kind of attachment structure based on the expansible system of sensing channel for perceiving Evolutionary Neural Network of the present invention is shown
It is intended to.
Fig. 2 is the two dimensional source figure of embodiments of the invention.
Fig. 3 is the three-dimensional source figure of embodiments of the invention.
Fig. 4 is the design sketch after the present invention is performed, and the arrow left side is that the prototype for obtaining, arrow are perceived by two dimensional source diagram data
The right is the increase in new perceiving and perceive the prototype for obtaining after neuron.
Specific embodiment
The purpose of the present invention is that the efficient a kind of of development automation can based on the sensing channel for perceiving Evolutionary Neural Network
Extension system and method, are further detailed by drawings and Examples:
Based on the expansible system of sensing channel for perceiving Evolutionary Neural Network, including data collecting system 1, described data
Acquisition system 1 is connected with control system 2, with based on the data sense for perceiving Evolutionary Neural Network in described control system 2
Know receiver module 3.
Based on the method for the expansible system of sensing channel for perceiving Evolutionary Neural Network, step is as follows:
Step 1:The image or sample of signal as video that first data collecting system carrys out collection is sent to control
In system 2, then control system 2 starts based on the expansible system module 3 of the sensing channel for perceiving Evolutionary Neural Network first most
The two sample of signal x for first collecting1And x2Each self-corresponding n is constructed respectively ties up prototype P1Prototype P is tieed up with the 2nd n2,
First n ties up prototype P1Prototype P is tieed up with the 2nd n2In being stored in set P, the sensing channel for perceiving Evolutionary Neural Network is additionally based on
3 images for sending of expansible system module or each sample of signal as video are configured to signal vector, described
A n dimension prototype P1=(x1,1,x1,2,...,x1,n), the 2nd n dimension prototypes P2=(x2,1,x2,2,...,x2,n), if at first
The sample of signal later of two sample of signal for collecting is n dimensional signals vector, then the form of described signal vector is xj
=(xj,1,xj,2,...,xj,n)∈Sl, perform based on the method for the expansible system of sensing channel for perceiving Evolutionary Neural Network
First stage, the described first stage starts to perform from step 8.In the event of the new perception neurons of m, then described letter
The form of number vector is xj=(xj,1,xj,2,...,xj,n,...,xj,n+m)∈Sh, xjRepresent the signal of j-th sample of signal to
Amount, j is natural number, xj,1Represent that first perceives input data of the neuron to the one-dimensional vector of j-th sample of signal, xj,2Table
Show that second perceives input data of the neuron to the one-dimensional vector of j-th sample of signal, xj,nRepresent n-th perception neuron
Input data to the one-dimensional vector of j-th sample of signal, xj,n+mRepresent that the n-th+m perceives neuron to j-th sample of signal
One-dimensional vector input data, ShRepresent that the vector that the collected corresponding signal vector of sample of signal is collectively constituted is empty
Between, h, l are natural number, and then execution step 2 realizes that second stage, second stage are completed after performing to the operation of step 7
Method based on the expansible system of sensing channel for perceiving Evolutionary Neural Network;
Step 2:Obtain sample of signal xj=(xj,1,xj,2,...,xj,n,...,xj,n+m)∈Sh.For each Pi∈ P,
If Pi∈Sl, i.e. PiIn space SlIn, by PiWith the sample of signal x of inputjFront n dimension attributes calculate Euclidean distance, it is described
N dimension attributes be SlAttribute.If Pi∈Sh, i.e. PiIn space ShIn, by PiWith the sample of signal x of inputj(n+m)
Dimension attribute calculates Euclidean distance, and described (n+m) dimension attribute is ShAttribute, by formula (1) obtain SlIn triumph prototype
Wherein, j represents natural number,Represent PiWeight vector,Represent in SlIn Euclidean distance, then by formula
And formula (3) obtains respectively S (2)hIn triumph prototypeWith the second triumph prototype
Wherein, PhExpression belongs to ShPrototype set, i.e., it is all of to belong to ShPrototype set,Represent in ShIn
Euclidean distance, be below abbreviatedFor Pc,WithIt is abbreviated as PaAnd Pb;
Step 3:If dim is (Pc)=n, i.e. Pc∈Sl, then check whether the criterion met as shown in formula (4):
Wherein,Represent PcIn SlIn similarity threshold.If formula (4) meets, prototype PcBe activated, subsequently do as
Renewal represented by formula (5), formula (6), formula (7) and formula (8):
Wh-l=xj h-l
(7)
Wherein,Represent prototype PcIn SlIn activation number of times,Represent prototype PcWeight vector,Represent SlTo
Amount attribute, i.e.,xh-lThe attribute of m new perception neuron is represented, i.e.,
Initialization in additionWithWherein,Represent prototype PcIn ShIn activation number of times,Represent PcIn ShIn
Similarity threshold.If condition (4) is unsatisfactory for, one such as formula (9), formula (10), formula (11), formula (12) and public affairs are set up
The new prototype of formula (13):
Wnew=xj
(9)
Wherein WnewThe weight vector of newly-built prototype is represented,Represent in SlIn newly-built prototype activation number of times,Represent in ShIn newly-built prototype activation number of times,Represent in SlIn newly-built prototype similarity threshold,Represent in ShIn newly-built prototype similarity threshold.
If dim is (Pc) ≠ n, i.e. PcIt has been ShIn prototype, if meeting the condition of formula (14):
WhereinRepresent prototype PaWeight vector, i.e. xjWith PaIt is identical, PaBe activated, it is corresponding make such as formula (15) and
The renewal of formula (16):
WhereinRepresent prototype PaIn SlIn activation number of times,Represent prototype PaIn ShIn activation number of times.
If the condition of formula (14) is unsatisfactory for, the condition for meeting formula (17) is checked whether:
WhereinRepresent prototype PaWeight vector,Represent PaIn ShIn similarity threshold,Represent prototype Pb's
Weight vector,Represent PbIn ShIn similarity threshold.If the condition of formula (17) is satisfied, to PaCarry out such as formula
(18), the renewal under formula (19), formula (20) and formula (21):
Wherein,Represent prototype PaIn SlIn weight vector,Represent prototype PaIn m newly-increased perception neuron
Weight vector;If the condition of formula (17) is unsatisfactory for, using formula (9), formula (10), formula (11), formula (12)
New prototype is set up with formula (13);
Step 4:If the condition of condition (17) is satisfied, and in prototype PaWith prototype PbBetween there is no connection, then
Set up PaAnd PbBetween connection, juxtapositionDescribedRepresent prototype PaWith prototype PbBetween connection
Age variable, described age variable represents the tolerance of the connection lifetime of connection.If PaAnd PbBetween connection deposited
As long as puttingIf prototype PiIt is thus lifted to higher dimensional space or weight vectors is updated, as formula
(22) to PiAnd PjBetween age variable be updated:
Wherein,Represent PiConnected prototype set.If the age variable of a connection is more than predetermined threshold value
Agemax, the connection is just deleted;
Step 5:If a prototype does not have neighbours, then the similarity threshold definition of the prototype is as shown in formula (23):
If there are neighbours in a prototype, then the similarity threshold definition of the prototype is as shown in formula (24):
Whenever one new signal sample x of inputjWhen, update P according to formula (23) and formula (24)a, PbAnd PcSimilitude
Threshold value;
Step 6:After λ sample signal is processed by step 1- step 5, prototype cutting is performed, and increase newly at m
The target that the prototype that there is before neuron occurs is not belonging to cutting is perceived, perceives what is increased newly after neuron occurs at m
Prototype belongs to the target of cutting, and these newly-increased prototypes are organized into set Q, is first according to formula (24) and calculates all's
Average
Wherein PhExpression belongs to space ShPrototype, | Ph| represent set PhElement number,Represent prototype PiIn Sh
In activation number of times.For each prototype Qi∈ Q, if such as formula (25)
Or formula (26)
Condition be satisfied, prototype QiTo be deleted, whereinRepresent QiNeighbours' number, λ is natural number, c
For predefined parameter, span [0,1];
Step 7:When all of prototype is mapped to ShAfter, by parameterWithCarry out as described in formula (27)
Merge, the activation number of times after being merged
Right value update formula described in formula (20) and formula (21) is changed into the more new formula of formula (28):
Formula (9), formula (10), formula (11), formula (12) and formula (13) be reduced to formula (29), formula (30) and
Shown in formula (31):
Wnew=xj
(29)
Then parameterTo be left out.For follow-up signal sample xj=(xj,1,xj,2,...,xj,n,...,xj,n+m)∈Sh
Process can refer to based on perceive Evolutionary Neural Network the expansible system of sensing channel method first stage.
Step 8:S is obtained by formula (32)lIn triumph prototype
Wherein, j represents natural number,Represent PiWeight vector,Represent in SlIn Euclidean distance.If met
The condition of formula (33):
WhereinRepresent prototype PaWeight vector, i.e. xjWith PaIt is identical, PaIt is activated, makees such as formula (34) accordingly
Update:
WhereinRepresent prototype PaIn SlIn activation number of times, if the condition of formula (33) is unsatisfactory for, check whether full
The condition of sufficient formula (35):
WhereinRepresent prototype PaWeight vector,Represent PaIn SlIn similarity threshold,Represent prototype Pb's
Weight vector,Represent PbIn SlIn similarity threshold.If the condition of formula (35) is satisfied, to PaCarry out such as formula
(36), the renewal under formula (37):
Wherein,Represent prototype PbIn SlIn weight vector.If the condition of formula (35) is unsatisfactory for, using such as formula
(38), formula (39) and formula (40)
Wnew=xj
(38)
Set up new prototype;
Step 9:If the condition of condition (35) is satisfied, and in prototype PaWith prototype PbBetween there is no connection, then
Set up PaAnd PbBetween connection, juxtapositionDescribedShow prototype PaWith prototype PbBetween age become
Amount, described age variable represents the tolerance of the connection lifetime of connection;If PaAnd PbBetween connection existed, as long as puttingIf prototype PiWeight vectors be updated, as formula (41) is to PiAnd PjBetween age variable enter
Row updates,
Wherein,Represent PiConnected prototype set, if one connection age variable be more than predetermined threshold value
Agemax, the connection is just deleted;
Step 10:If a prototype does not have neighbours, then the similarity threshold definition of the prototype is as shown in formula (42):
If there are neighbours in a prototype, then the similarity threshold definition of the prototype is as shown in formula (43):
Whenever one new signal sample x of inputjWhen, update P according to formula (42) and formula (43)aAnd PbSimilitude threshold
Value;
Step 11:After λ sample signal is processed by step 8- step 10, prototype cutting is performed, and is first according to public affairs
Formula (44) calculates allAverage
For each prototype Pi∈ P, if such as formula (45)
Or formula (46)
Condition be satisfied, prototype PiTo be deleted, whereinRepresent PiNeighbours' number, λ is natural number, and c is predefined
Parameter, span [0,1].
Using the method for the present invention, as shown in Figure 2, Figure 3 and Figure 4, carry out according to the present invention from the two dimensional source diagram data of Fig. 2
The step of first carry out the perception of first stage, the prototype for obtaining after increasing new perception neuron, passes through as shown in arrow left part
Second stage just can be obtained such as the prototype of arrow right part, the prototype on arrow both sides respectively with the source figure and three-dimensional source figure phase of two dimension
Relatively, can be appreciated that reduction reproduction effect is very true to nature, and this method is capable of full-automatic realization, without the need for artificial participation,
It is truly realized intelligentized operation.
The above, is only presently preferred embodiments of the present invention, and any pro forma restriction is not made to the present invention, though
So the present invention is disclosed above with preferred embodiment, but is not limited to the present invention, any to be familiar with this professional technology people
Member, in the range of without departing from technical solution of the present invention, when making a little change or modification using the technology contents of the disclosure above
For the Equivalent embodiments of equivalent variations, as long as be without departing from technical solution of the present invention content, according to the technical spirit of the present invention,
Within the spirit and principles in the present invention, any simple modification, equivalent and improvement that above example is made etc., still
Belong within the protection domain of technical solution of the present invention.
Claims (1)
1. it is a kind of based on perceive Evolutionary Neural Network the expansible system of sensing channel method, it is characterised in that adopt including data
Collecting system, described data collecting system is connected with control system, with refreshing based on evolution is perceived in described control system
The data perception receiver module of Jing networks;Described control system perceives nerve according to original n of original neutral net
Unit, after the perception of a period of time, under conditions of increasing m new perception neuron appearance newly, in the memory block of control system
Original lower dimensional space corresponding to original n perception neuron is set in domain, and described original lower dimensional space is Sl=
Rn, and the new higher dimensional space corresponding to m new perception neuron is set in the storage region of control system, described is new
Higher dimensional space be Sh=Rn+m, and the prototype obtained after perceiving and compressing is stored in set P, uses PiRepresent i-th in P
Prototype, described set P be also in the storage region of control system arrange a memory space, n and m be natural number, Rn
Represent the real number vector space of n dimensions, Rn+mRepresent the real number vector space of n+m dimensions;
Step is as follows:
Step 1:The image or sample of signal as video that first data collecting system carrys out collection is sent to control system
In, then control system starts based on perceiving the expansible system module of sensing channel of Evolutionary Neural Network first collecting at first
Two sample of signal x1And x2Each self-corresponding n is constructed respectively ties up prototype P1Prototype P is tieed up with the 2nd n2, n dimension originals
Type P1Prototype P is tieed up with the 2nd n2In being stored in set P, the expansible system of sensing channel for perceiving Evolutionary Neural Network is additionally based on
System module is configured to signal vector, described n dimensions each sample of signal as the image or video for sending
Prototype P1=(x1,1,x1,2,...,x1,n), the 2nd n dimension prototypes P2=(x2,1,x2,2,...,x2,n), if at first collected
The sample of signal later of two sample of signal is n dimensional signals vector, then the form of described signal vector is xj=(xj,1,
xj,2,...,xj,n)∈Sl, perform the first rank of the method based on the expansible system of sensing channel for perceiving Evolutionary Neural Network
Section, described first stage starts to perform from step 8, in the event of m new perception neuron, then described signal vector
Form be xj=(xj,1,xj,2,...,xj,n,...,xj,n+m)∈Sh, xjThe signal vector of j-th sample of signal is represented, j is certainly
So count, xj,1Represent that first perceives input data of the neuron to the one-dimensional vector of j-th sample of signal, xj,2Represent second
Perceive input data of the neuron to the one-dimensional vector of j-th sample of signal, xj,nRepresent n-th and perceive neuron to j-th letter
The input data of the one-dimensional vector of number sample, xj,n+mRepresent that the n-th+m perceives one-dimensional vector of the neuron to j-th sample of signal
Input data, ShThe vector space that the collected corresponding signal vector of sample of signal is collectively constituted is represented, h, l are certainly
So count, then execution step 2 realizes that second stage, second stage complete to be evolved based on perception after performing to the operation of step 7
The method of the expansible system of sensing channel of neutral net;
Step 2:Obtain sample of signal xj=(xj,1,xj,2,...,xj,n,...,xj,n+m)∈Sh, for each Pi∈ P, if
Pi∈Sl, i.e. PiIn space SlIn, by PiWith the sample of signal x of inputjFront n dimension attributes calculate Euclidean distance, described n dimensions
Attribute is SlAttribute, if Pi∈Sh, i.e. PiIn space ShIn, by PiWith the sample of signal x of inputj(n+m) dimension category
Property calculate Euclidean distance, described (n+m) dimension attribute is ShAttribute, by formula (1) obtain SlIn triumph prototype
Wherein, j represents natural number,Represent PiWeight vector,Represent in SlIn Euclidean distance, then by formula (2) and
Formula (3) obtains respectively ShIn triumph prototypeWith the second triumph prototype
Wherein, PhExpression belongs to ShPrototype set, i.e., it is all of to belong to ShPrototype set,Represent in ShIn Europe
Family name's distance, is below abbreviatedFor Pc,WithIt is abbreviated as PaAnd Pb;
Step 3:If dim is (Pc)=n, i.e. Pc∈Sl, then check whether the criterion met as shown in formula (4):
Wherein,Represent PcIn SlIn similarity threshold, if formula (4) meet, prototype PcIt is activated, subsequently does such as formula
(5), the renewal represented by formula (6), formula (7) and formula (8):
Wherein,Represent prototype PcIn SlIn activation number of times,Represent prototype PcWeight vector,Represent SlVector
Attribute, i.e.,xh-lThe attribute of m new perception neuron is represented, i.e.,
Initialization in additionWithWherein,Represent prototype PcIn ShIn activation number of times,Represent PcIn ShIn
Similarity threshold, if condition (4) is unsatisfactory for, sets up one such as formula (9), formula (10), formula (11), formula (12) and public affairs
The new prototype of formula (13):
Wnew=xj (9)
Wherein WnewThe weight vector of newly-built prototype is represented,Represent in SlIn newly-built prototype activation number of times,
Represent in ShIn newly-built prototype activation number of times,Represent in SlIn newly-built prototype similarity threshold,Table
Show in ShIn newly-built prototype similarity threshold;
If dim is (Pc) ≠ n, i.e. PcIt has been ShIn prototype, if meeting the condition of formula (14):
WhereinRepresent prototype PaWeight vector, i.e. xjWith PaIt is identical, PaIt is activated, it is corresponding to make such as formula (15) and formula
(16) renewal:
WhereinRepresent prototype PaIn SlIn activation number of times,Represent prototype PaIn ShIn activation number of times;
If the condition of formula (14) is unsatisfactory for, the condition for meeting formula (17) is checked whether:
WhereinRepresent prototype PaWeight vector,Represent PaIn ShIn similarity threshold,Represent prototype PbWeights to
Amount,Represent PbIn ShIn similarity threshold, if the condition of formula (17) is satisfied, to PaCarry out such as formula (18), formula
(19), the renewal under formula (20) and formula (21):
Wherein,Represent prototype PaIn SlIn weight vector,Represent prototype PaIn the power of m newly-increased perception neuron
Value vector;If the condition of formula (17) is unsatisfactory for, using formula (9), formula (10), formula (11), formula (12) and public affairs
Formula (13) sets up new prototype;
Step 4:If the condition of condition (17) is satisfied, and in prototype PaWith prototype PbBetween there is no connection, then set up
PaAnd PbBetween connection, juxtapositionDescribedRepresent prototype PaWith prototype PbBetween connection year
Age variable, described age variable represents the tolerance of the connection lifetime of connection, if PaAnd PbBetween connection existed, only
PutIf prototype PiIt is thus lifted to higher dimensional space or weight vectors is updated, as formula (22)
To PiAnd PjBetween age variable be updated:
Wherein,Represent PiConnected prototype set, if one connection age variable be more than predetermined threshold value
Agemax, the connection is just deleted;
Step 5:If a prototype does not have neighbours, then the similarity threshold definition of the prototype is as shown in formula (23):
If there are neighbours in a prototype, then the similarity threshold definition of the prototype is as shown in formula (24):
Whenever one new signal sample x of inputjWhen, update P according to formula (23) and formula (24)a, PbAnd PcSimilarity threshold;
Step 6:After λ sample signal is processed by step 1- step 5, prototype cutting is performed, and in m newly-increased perception
The prototype that neuron there is before occurring is not belonging to the target of cutting, and the prototype increased newly after neuron occurs is perceived at m
Belong to the target of cutting, these newly-increased prototypes are organized into set Q, be first according to formula (24) and calculate allAverage
Wherein PhExpression belongs to space ShPrototype, | Ph| represent set PhElement number,Represent prototype PiIn ShIn
Activation number of times, for each prototype Qi∈ Q, if such as formula (25)
Or formula (26)
Condition be satisfied, prototype QiTo be deleted, wherein Qi∈ Q,Represent QiNeighbours' number, λ is natural number, and c is predetermined
The parameter of justice, span [0,1];
Step 7:When all of prototype is mapped to ShAfter, by parameterWithThe merging as described in formula (27) is carried out,
Activation number of times after being merged
Right value update formula described in formula (20) and formula (21) is changed into the more new formula of formula (28):
Formula (9), formula (10), formula (11), formula (12) and formula (13) are reduced to formula (29), formula (30) and formula
(31) shown in:
Wnew=xj (29)
Then parameterTo be left out, for follow-up signal sample xj=(xj,1,xj,2,...,xj,n,...,xj,n+m)∈ShPlace
Reason can refer to the first stage of the method based on the expansible system of sensing channel for perceiving Evolutionary Neural Network,
Step 8:S is obtained by formula (32)lIn triumph prototype
Wherein, j represents natural number,Represent PiWeight vector,Represent in SlIn Euclidean distance, if meeting formula
(33) condition:
WhereinRepresent prototype PaWeight vector, i.e. xjWith PaIt is identical, PaIt is activated, the corresponding renewal made such as formula (34):
WhereinRepresent prototype PaIn SlIn activation number of times, if the condition of formula (33) is unsatisfactory for, check whether that satisfaction is public
The condition of formula (35):
WhereinRepresent prototype PaWeight vector,Represent PaIn SlIn similarity threshold,Represent prototype PbWeights to
Amount,Represent PbIn SlIn similarity threshold, if the condition of formula (35) is satisfied, to PaCarry out such as formula (36), formula
(37) renewal under:
Wherein,Represent prototype PbIn SlIn weight vector, if the condition of formula (35) is unsatisfactory for, using as formula (38),
Formula (39) and formula (40)
Wnew=xj (38)
Set up new prototype;
Step 9:If the condition of condition (35) is satisfied, and in prototype PaWith prototype PbBetween there is no connection, then set up
PaAnd PbBetween connection, juxtapositionDescribedRepresent prototype PaWith prototype PbBetween age become
Amount, described age variable represents the tolerance of the connection lifetime of connection;If PaAnd PbBetween connection existed, as long as puttingIf prototype PiWeight vectors be updated, as formula (41) is to PiAnd PjBetween age variable enter
Row updates,
Wherein,Represent PiConnected prototype set, if one connection age variable be more than predetermined threshold value
Agemax, the connection is just deleted;
Step 10:If a prototype does not have neighbours, then the similarity threshold definition of the prototype is as shown in formula (42):
If there are neighbours in a prototype, then the similarity threshold definition of the prototype is as shown in formula (43):
Whenever one new signal sample x of inputjWhen, update P according to formula (42) and formula (43)aAnd PbSimilarity threshold;
Step 11:After λ sample signal is processed by step 8- step 10, prototype cutting is performed, and is first according to formula
(44) calculate allAverage
For each prototype Pi∈ P, if such as formula (45)
Or formula (46)
Condition be satisfied, prototype PiTo be deleted, whereinRepresent PiNeighbours' number, λ is natural number, and c is predefined ginseng
Number, span [0,1].
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410432884.8A CN104200095B (en) | 2014-08-28 | 2014-08-28 | System and method for expanding sensing channel on basis of perceiving evolutionary neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410432884.8A CN104200095B (en) | 2014-08-28 | 2014-08-28 | System and method for expanding sensing channel on basis of perceiving evolutionary neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104200095A CN104200095A (en) | 2014-12-10 |
CN104200095B true CN104200095B (en) | 2017-05-10 |
Family
ID=52085388
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410432884.8A Active CN104200095B (en) | 2014-08-28 | 2014-08-28 | System and method for expanding sensing channel on basis of perceiving evolutionary neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104200095B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8148689B1 (en) * | 2008-07-24 | 2012-04-03 | Braunheim Stephen T | Detection of distant substances |
CN103618576A (en) * | 2013-08-29 | 2014-03-05 | 上海永畅信息科技有限公司 | Low complexity modeling based multidimensional channel perception system of internet of vehicles |
CN103929210A (en) * | 2014-04-25 | 2014-07-16 | 重庆邮电大学 | Hard decision decoding method based on genetic algorithm and neural network |
-
2014
- 2014-08-28 CN CN201410432884.8A patent/CN104200095B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8148689B1 (en) * | 2008-07-24 | 2012-04-03 | Braunheim Stephen T | Detection of distant substances |
CN103618576A (en) * | 2013-08-29 | 2014-03-05 | 上海永畅信息科技有限公司 | Low complexity modeling based multidimensional channel perception system of internet of vehicles |
CN103929210A (en) * | 2014-04-25 | 2014-07-16 | 重庆邮电大学 | Hard decision decoding method based on genetic algorithm and neural network |
Also Published As
Publication number | Publication date |
---|---|
CN104200095A (en) | 2014-12-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110728330A (en) | Object identification method, device, equipment and storage medium based on artificial intelligence | |
DE112018002572T5 (en) | METHODS, SYSTEMS AND DEVICES FOR OPTIMIZING THE PIPELINE DESIGN | |
EP2191233A1 (en) | Apparatus and method for updating map data | |
DE102012218966A1 (en) | Method and system for identifying original data generated by things in the Internet of Things | |
CN109104731B (en) | Method and device for building cell scene category division model and computer equipment | |
CN109242013A (en) | A kind of data mask method, device, electronic equipment and storage medium | |
CN110163213A (en) | Remote sensing image segmentation method based on disparity map and multiple dimensioned depth network model | |
CN111382305B (en) | Video deduplication method, video deduplication device, computer equipment and storage medium | |
CN106446223A (en) | Map data processing method and device | |
CN109783757A (en) | Render method and device, the system, storage medium, electronic device of webpage | |
CN109242813A (en) | Image interfusion method and device based on potential low-rank representation and NSST | |
CN106683174A (en) | 3D reconstruction method, apparatus of binocular visual system, and binocular visual system | |
CN103763565A (en) | Anaglyph coding method based on three-dimensional self-organizing mapping | |
CN112817563B (en) | Target attribute configuration information determining method, computer device, and storage medium | |
Shao et al. | Toward domain transfer for no-reference quality prediction of asymmetrically distorted stereoscopic images | |
CN105005852A (en) | Image analysis based intelligent monitoring system for dormitory environment | |
CN113392731A (en) | Modulated signal classification method and system based on graph neural network | |
CN104238885A (en) | Automatic layout and content displaying method and device suitable for multi-equipment content displaying | |
CN104200095B (en) | System and method for expanding sensing channel on basis of perceiving evolutionary neural network | |
CN105550644A (en) | Skin state analysis method and skin state analysis device based on single-sided mirror | |
DE112019006526T5 (en) | Computing device | |
CN104243956A (en) | Three-dimensional image visual saliency map extraction method | |
CN106210711A (en) | A kind of without with reference to stereo image quality evaluation methodology | |
CN106960432B (en) | A kind of no reference stereo image quality evaluation method | |
Jayageetha et al. | Medical image quality assessment using CSO based deep neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |