CN106599927A - Target grouping method based on fuzzy ART division - Google Patents
Target grouping method based on fuzzy ART division Download PDFInfo
- Publication number
- CN106599927A CN106599927A CN201611186577.1A CN201611186577A CN106599927A CN 106599927 A CN106599927 A CN 106599927A CN 201611186577 A CN201611186577 A CN 201611186577A CN 106599927 A CN106599927 A CN 106599927A
- Authority
- CN
- China
- Prior art keywords
- target
- current
- multiple targets
- label
- input vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/192—Recognition using electronic means using simultaneous comparisons or correlations of the image signals with a plurality of references
- G06V30/194—References adjustable by an adaptive method, e.g. learning
Abstract
The invention discloses a target grouping method based on fuzzy ART division and relates to the situation estimation technology field. The method comprises steps that 1, target position measurement data and target identification attribute data are read in; 2, a grouping target number scale is reduced through target identification attribute division to improve grouping efficiency; 3, scale difference is eliminated through division data pre-processing, target space division based on fuzzy ART is further employed to effectively filter noise interference, and grouping accuracy is improved; and 4, a grouping result is outputted. The method is advantaged in that problems of the unknown classification number and noise interference existing in a target grouping method in the prior art are mainly solved, effect, accurate, real-time and dynamic grouping can be realized for multiple formation group targets under the condition of the unknown classification number, and the method can be applied to situation estimation and command control systems.
Description
Technical field
The invention belongs to battle field situation technical field, more particularly to a kind of to be based on Fuzzy ART (Adaptive Resonance
Theory, adaptive resonance theory) divide Target cluster dividing method, can be used for battle field situation, command and control system.
Background technology
Battle state display is effective information acquiring way of the commander to real-time condition control, is provided with decision-making to develop programs
Basis and support.If will each target regard isolated individuality as, not only existence information redundancy, and densely covered target identification
The dazzling problem of information can be also caused, makes commander quickly cannot directly understand situation overview.Accordingly, it would be desirable to by recognition property and
The close target polymerization of the features such as kinematic parameter is sorted out, and is divided into several multiple targetses, with its actual formation phase for performing task
Correspondence.So, on the one hand can simplify battle state display, beneficial to commander the overall situation is controlled rapidly;On the other hand, after Target cluster dividing
As a result the essence that its task is formed into columns more can be directly embodied, action is excavated from magnanimity information and is intended to, be follow-up state
Potential analysises lay the foundation.
At present, typical Target cluster dividing method has:The methods such as fuzzy C-mean algorithm, K averages, arest neighbors and ISODATA.Wherein:
Fuzzy C-mean algorithm method and K Mean Methods, need default classification number, and the situation unknown with classification number often is faced with is not
Matching, and its classification results is stronger to dependency that preliminary classification center is chosen, in turn results in classification results stability not good enough;
Arest neighbors method, by given threshold a point group is realized, is simply easily realized and is widely used, but lacks effective threshold value
Choosing method, it is difficult to point group's problem of effective process difference metric or situation;
ISODATA methods, the dynamic point under class number unknown situation is realized by the merging to cluster result and splitting operation
Group, but it is using the distance of sample and cluster centre as point group's foundation, is suitable for solving the problems, such as spherical cluster sample point group, and for
Line style common in battle field situation is formed into columns and divides group's problem Shortcomings.
The content of the invention
Present invention aims to the deficiency in above-mentioned prior art, divided using Fuzzy ART, Jing classes select,
The increment type dynamic point group to many formation multiple targetses is realized with the step such as degree inspection and class study, it is proposed that one kind is based on fuzzy
The Target cluster dividing method that ART is divided, effectively improves point group's efficiency, accuracy rate and a stability.
Realizing the key problem in technology of the present invention is:During Target cluster dividing, first by target recognition Attribute transposition, about subtract
Divide multiple targets number scale, raising point group's efficiency eliminates different scale secondly by data prediction is divided, and then adopts based on mould
The object space of paste ART is divided, and effectively filters out noise jamming, improves point group's accuracy rate.Implementation step includes as follows:
(1) target location metric data and target recognition attribute data that the observation of current time sensor is obtained are read in, it is right
All targets enter respectively line label;Described target recognition attribute data is by red blue party identification data and type identification data set
Into;
(2) all target labels are divided according to target recognition attribute data, obtains all types of target labels of red
Collection and all types of target label collection of blue party;And initialize multiple targets and its weight vectors;
(3) respectively Fuzzy ART division is carried out to all types of target label collection of red and all types of target label collection of blue party;It is blue
The all types of target label collection in side carry out the mode of Fuzzy ART division and all types of target label collection of red carries out Fuzzy ART division
Mode it is identical;Specifically include following steps:
(3a) the current goal position metric data in all types of target label collection in red blue side is normalized and is supplemented
Coding, obtains current input vector;
(3b) class selection is carried out according to the weight vectors of current input vector and current all multiple targetses, obtains choosing class
Mark;
(3c) current input vector is calculated with the matching degree for choosing the weight vectors corresponding to category;
(3d) according to matching degree and the magnitude relationship of alarm threshold, current goal label is added to and is chosen corresponding to category
Multiple targets in and carry out class study, or using current input vector as the weight vectors of newly-increased multiple targets and by current goal
Label is added in newly-increased multiple targets;
(3e) whether carried out Target cluster dividing, if do not had if checking all targets of all types of target label collection in red blue side
Have, next target location metric data is updated to into current goal position metric data, jump to step (3a);Otherwise, perform
Step (4);
(4) whole multiple targetses are exported, whether the sensor observation data for checking subsequent time reach, if so, by lower a period of time
Quarter is updated to current time, jumps to step (1);Otherwise, process ends.
The present invention has compared to existing technology advantages below:
1) present invention is by preferentially carrying out target recognition Attribute transposition, effectively about deduction multiple targets number scale, improves a point group
Efficiency, reduces amount of calculation;
2) present invention is by normalization and supplements coding, data different scale can be eliminated, so as to ensure in various data
Or in the case of method the suitability and effectiveness;
3) present invention passes through to be divided using the object space based on Fuzzy ART, can effectively filter out noise jamming, improves and divides
Group's accuracy rate.
Description of the drawings
Fig. 1 is the overall flow figure of the present invention;
Fig. 2 is to carry out the experimental result picture of space point group to static closely form into columns with the present invention and existing method more;
Fig. 3 is global object point group's experimental result picture that with the present invention many formation multiple targetses are entered with Mobile state point group;
Fig. 4 is the regional area A Target cluster dividing experimental result pictures that with the present invention many formation multiple targetses are entered with Mobile state point group;
Fig. 5 is the regional area B Target cluster dividing experimental result pictures that with the present invention many formation multiple targetses are entered with Mobile state point group.
Specific embodiment
Below in conjunction with the accompanying drawings and embodiment the present invention will be further described in detail.
With reference to Fig. 1, the Target cluster dividing method divided based on Fuzzy ART of the present invention, following steps are specifically included:
Step 1. data are read in.
1.1) initial time k=1 is made, the target location metric data at k moment is read inWherein,Represent the k moment
T-th target position quantity direction finding amount, t is target label, and value is 1,2 ..., Nk, NkRepresent the target sum at k moment, k
Represent the moment;
1.2) the target recognition attribute data at k moment is read in, described target recognition attribute data recognizes number by red blue party
According toWith type identification dataComposition;Wherein,The red blue party recognition result of t-th target at k moment is represented,Represent the type identification result of t-th target at k moment.
Step 2. target recognition Attribute transposition.
2.1) all target labels are divided according to target recognition attribute data, obtains all types of targets in red blue side
Label set Pij:
Wherein, i is red blue party label, and j is type label, and C represents type sum.
2.2) multiple targets G is initializedij1And its weight vectors w1:
w1=(1,1,1,1); 2)
Object space of the step 3. based on Fuzzy ART is divided.
Target label collection Ps all types of to red blue side respectivelyijCarry out Fuzzy ART division;The all types of target label collection of blue party enter
The mode that row Fuzzy ART is divided is identical with the mode that all types of target label collection of red carry out Fuzzy ART division;Specifically include with
Lower step:
3.1) target label collection Ps all types of to red blue sideijInterior current goal position metric data is normalized and supplements
Coding, obtains current input vector I:
3.1.1) current goal position metric data is normalized, the measurement of normalization position is obtained:
L=(x ', y '); 3)
Wherein,Represent the position quantity direction finding amount of t-th target at k moment;X and y represent respectively the horizontal seat of target location
Mark and vertical coordinate;L represents that normalization position measures;X ' and y ' represent respectively the abscissa and vertical coordinate of target location after normalization xWithThe bound of target acquisition region transverse axis is represented respectively,yWith
The bound of the target acquisition region longitudinal axis is represented respectively;
3.1.2) normalization position is measured carries out supplementing coding, obtains current input vector:
I=(x ', y ', x 'c,y′c); 4)
Wherein, I is current input vector;x′cWith y 'cTarget location abscissa and vertical coordinate after normalization is represented respectively
Supplement coding, x 'c=1-x ', y 'c=1-y '.
3.2) according to the current input vector I and weight vectors w of current all multiple targetsesh, h=1,2 ..., Ω carry out class
Select, obtain choosing category H:
3.2.1 the class for) calculating current input vector and weight vectors selects function:
Wherein, Th(I) function is selected for class;whFor weight vectors;H is multiple targets label, and value is h=1,2 ..., Ω,
Ω represents current multiple targets number;α is selection parameter, α > 0;| | for L1 norms;∧ ships calculation for fuzzy;
3.2.2) for current input vector, the current all multiple targetses of traversal ask for maximum kind and select function, are chosen
Category:
Wherein, H is to choose category;Represent and cause class to select function Th(I) multiple targets of maximum is taken
The value of label h.
3.3) current input vector is calculated with the matching degree for choosing the weight vectors corresponding to category:
Wherein, MH(I) it is matching degree.
3.4) according to matching degree and the magnitude relationship of alarm threshold, current goal label is added to and is chosen corresponding to category
Multiple targets in and carry out class study, or using current input vector as the weight vectors of newly-increased multiple targets and by current goal
Label is added in newly-increased multiple targets:
3.4.1) if matching degree is more than or equal to alarm threshold ρ, current goal label is added to and is chosen corresponding to category
In multiple targets, execution step 3.4.2);Otherwise, it is determined whether having traveled through matching for current input vector and current all multiple targetses
Degree, if it did not, execution step 3.2), if traveled through, execution step 3.4.3);Wherein, GijH=GijH∪ t, GijHFor choosing
Multiple targets corresponding to middle category, ∪ is to take union;
3.4.2) by current input vector and after choosing the weight vectors corresponding to category to be merged, after being updated
Weight vectors, execution step 3.5);Wherein, the computing formula of the weight vectors after renewal is:
wH=β (I ∧ wH)+(1-β)wH; 8)
Wherein, β is learning rate, β ∈ [0,1].
3.4.3 current multiple targets number) is updated, using current input vector as the weight vectors of newly-increased multiple targets, incites somebody to action current
Target label is added in newly-increased multiple targets, and jumps to step 3.5);Wherein, the current multiple targets number Ω=Ω after renewal+
1, increase the weight vectors w of multiple targets newlyΩ=I, GijΩ=GijΩ∪ t, GijΩTo increase multiple targets newly.
3.5) all types of target label collection P in red blue side are checkedijAll targets whether carried out Target cluster dividing, if
No, next target location metric data is updated to into current goal position metric data, then jumps to step 3.1);Otherwise,
Execution step 4.
Step 4. grouping result is exported.
4.1) whole multiple targets G are exportedijh, i=1,2, j=1,2 ..., C, h=1,2 ..., Ω;
4.2) whether the sensor observation data for checking subsequent time reach, and if so, make k=k+1, return to step 1 carry out
Iteration;Otherwise, process ends.
The effect of the present invention can be further illustrated by following emulation experiment:
1. simulated conditions.
Simulated environment:Computer adopts Intel Core i3-2130CPU 3.4Ghz, 2GB internal memories, software to adopt
Matlab R2011a Simulation Experimental Platforms.
Simulation parameter:Selection parameter α=1 × 10-6, alarm threshold ρ=0.7, learning rate β=1.
2. emulation mode.
Method 1:The inventive method;
Method 2:Fuzzy C-mean algorithm method;
Method 3:K Mean Methods;
Method 4:ISODATA methods.
3. emulation content and result.
Emulation 1:With four kinds of methods, space point group is carried out to static closely form into columns more, as a result as shown in Fig. 2 its
In:
Fig. 2 (a) is that closely form into columns carries out the result figure of space point group more with 1 pair of static state of method;
Fig. 2 (b) is that closely form into columns carries out the result figure of space point group more with 2 pairs of static state of method;
Fig. 2 (c) is that closely form into columns carries out the result figure of space point group more with 3 pairs of static state of method;
Fig. 2 (d) is that closely form into columns carries out the result figure of space point group more with 4 pairs of static state of method.
Figure it is seen that the inventive method can preferably realize the closely point groups that form into columns more, and without the need for default class
Number, although method 2 and method 3 also can preferably realize Target cluster dividing, its classification number is both needed to be preset as actual value 2, and classifies
Number preset value directly affects the correctness of classification results, and the grouping result of mistake is necessarily obtained when number setting of classifying is wrong, because
And be difficult to meet point group's demand of number unknown situation of classifying in practice, although method 4 is without default class number, its with sample with
The distance of cluster centre is not suitable for solving the problems, such as point group that line style is formed into columns as point group's foundation, especially when target in line style formation
When minimum range is close between maximum spacing and formation, the parameter setting of the method is particularly difficult, it is difficult to obtain preferably point group's knot
Really.
Carry out 1000 times respectively to the scene in Fig. 2 to run, statistical average run time, as a result as shown in table 1.
Table 1
Can be seen that by the statistical data in table 1:Because method 2, method 3 and method 4 are required to by iterating to calculate in fact
Now divide group, more take;And combining adaptive resonance theory of the present invention and fuzzy set theory, it is possible to achieve increment type dynamic point
Group, directly obtains grouping result, and operational efficiency is higher.
To sum up can draw, the present invention is superior to method 2, method 3 and method 4 in terms of point group's accuracy rate and real-time.
Emulation 2:With method 1, many formation multiple targetses are entered with Mobile state point group, as a result as shown in Fig. 3, Fig. 4 and Fig. 5, wherein:
Fig. 3 is the global object grouping result for entering Mobile state point group with method formation multiple targets more than 1 pair;Fig. 4 is to be formed into columns 1 pair with method more
Multiple targets enters the regional area A Target cluster dividing results of Mobile state point group;Fig. 5 is to enter Mobile state with method formation multiple targets more than 1 pair
Divide the regional area B Target cluster dividing results of group.
In figure 3, dotted rectangle for ease of observation experiment result regional area, alphabetical A, the B beside regional area
Numbering corresponding to regional area, black hexagon is that each multiple targets observes original position, each multiple targets formation situation such as table 2
It is shown.The situation situation that Fig. 3 is described is that blue party aircraft is formed into columns to red platooning Fast marching more with vehicle, meets with red
Air formation is withdrawn after intercepting.In Fig. 3, Fig. 4 and Fig. 5, stain represents that the position of each target measures, in the range of solid-line rectangle frame
Multiple targets be divided and belong to same multiple targets, the numeral beside each multiple targets movement locus is the volume corresponding to multiple targets
Number.
Table 2
It is larger multiple targets formation spacing under practical situation to be can be seen that from Fig. 3, Fig. 4 and Fig. 5, but due to effect of noise,
Target distance persistently changes in the formation that observation is obtained, and the present invention can realize the efficient dynamic point group of increment type, effectively
The unknown many formation for the treatment of classification number point group's problem.
Carry out 1000 operations, statistical average run time respectively to the scene in Fig. 3, the run time of the present invention is
0.0039s, meets requirement of real-time.
To sum up can draw, the present invention the unknown many formation multiple targetses of number of classifying under practical situation can be realized effectively,
Accurately and in real time dynamic divides group.
Claims (9)
1. the Target cluster dividing method for being divided based on Fuzzy ART, it is characterised in that comprise the following steps:
(1) target location metric data and target recognition attribute data that the observation of current time sensor is obtained are read in, to all
Target enters respectively line label;Described target recognition attribute data is made up of red blue party identification data and type identification data;
(2) all target labels are divided according to target recognition attribute data, obtain all types of target label collection of red and
The all types of target label collection of blue party;Meanwhile, initialize multiple targets and its weight vectors;
(3) respectively Fuzzy ART division is carried out to all types of target label collection of red and all types of target label collection of blue party;Blue party is each
Type target label set carries out the mode of Fuzzy ART division and all types of target label collection of red and carries out the side of Fuzzy ART division
Formula is identical;Specifically include following steps:
(3a) it is normalized and supplements volume to the current goal position metric data in all types of target label collection in red blue side
Code, obtains current input vector;
(3b) class selection is carried out according to the weight vectors of current input vector and current all multiple targetses, obtains choosing category;
(3c) current input vector is calculated with the matching degree for choosing the weight vectors corresponding to category;
(3d) according to matching degree and the magnitude relationship of alarm threshold, current goal label is added to the group chosen corresponding to category
In target and carry out class study, or using current input vector as the weight vectors of newly-increased multiple targets and by current goal label
In being added to newly-increased multiple targets;
(3e) whether carried out Target cluster dividing, if it did not, will if checking all targets of all types of target label collection in red blue side
Next target location metric data is updated to current goal position metric data, jumps to step (3a);Otherwise, execution step
(4);
(4) whole multiple targetses are exported, whether the sensor observation data for checking subsequent time reach, if so, by subsequent time more
It is newly current time, jumps to step (1);Otherwise, process ends.
2. the Target cluster dividing method divided based on Fuzzy ART according to claim 1, it is characterised in that step (2) is described
The all types of target label collection of red and all types of target label collection of blue party, specially:
Wherein, PijFor all types of target label collection in red blue side, i is red blue side's label, and j is category label, and value is 1,2 ...,
C, C represent type sum;T is target label, and value is 1,2 ..., Nk, NkThe target sum at k moment is represented, k represents the moment;Represent the red blue party identification data of t-th target at k moment;Represent the type identification data of t-th target at k moment.
3. it is according to claim 2 based on Fuzzy ART divide Target cluster dividing method, it is characterised in that the step
(3a), following steps are specifically included:
(3a1) current goal position metric data is normalized, obtains the measurement of normalization position:
Wherein,Represent the position quantity direction finding amount of t-th target at k moment;X and y represent respectively target location abscissa and
Vertical coordinate;L represents that normalization position measures;X ' and y ' represent respectively the abscissa and vertical coordinate of target location after normalization X andRepresent the bound of target acquisition region transverse axis respectively, y and
The bound of the target acquisition region longitudinal axis is represented respectively;
(3a2) normalization position is measured carries out supplementing coding, obtains current input vector:
I=(x ', y ', x 'c,y′c);
Wherein, I is current input vector;x′cWith y 'cThe supplement of target location abscissa and vertical coordinate after normalization is represented respectively
Coding, x 'c=1-x ', y 'c=1-y '.
4. it is according to claim 3 based on Fuzzy ART divide Target cluster dividing method, it is characterised in that the step
(3b) following steps are specifically included:
(3b1) class for calculating current input vector and weight vectors selects function:
Wherein, Th(I) function is selected for class;whFor weight vectors;H is multiple targets label, and value is h=1,2 ..., Ω, Ω table
Show current multiple targets number;α is selection parameter, α > 0;| | for L1 norms;∧ ships calculation for fuzzy;
(3b2) for current input vector, the current all multiple targetses of traversal ask for maximum kind and select function, obtain choosing category:
Wherein, H is to choose category;Represent and cause class to select function Th(I) the multiple targets label h of maximum is taken
Value.
5. it is according to claim 4 based on Fuzzy ART divide Target cluster dividing method, it is characterised in that the step
(3c) computing formula of matching degree is in:
Wherein, MH(I) it is matching degree.
6. it is according to claim 5 based on Fuzzy ART divide Target cluster dividing method, it is characterised in that the step
(3d) it is specially:
If (3d1) matching degree is more than or equal to alarm threshold, current goal label is added to the multiple targets chosen corresponding to category
In, execution step (3d2);Otherwise, it is determined whether the matching degree of current input vector and current all multiple targetses has been traveled through, if
No, execution step (3b), if traveled through, execution step (3d3);Wherein, GijH=GijH∪ t, GijHTo choose category institute
Corresponding multiple targets, ∪ is to take union;
(3d2) by current input vector and after choosing the weight vectors corresponding to category to be merged, the weight after being updated
Vector, execution step (3e);
(3d3) update current multiple targets number, using current input vector as newly-increased multiple targets weight vectors, by current goal mark
Number it is added in newly-increased multiple targets, and jumps to step (3e);Wherein, GijΩ=GijΩ∪ t, GijΩTo increase multiple targets newly.
7. it is according to claim 6 based on Fuzzy ART divide Target cluster dividing method, it is characterised in that after the renewal
The computing formula of weight vectors be:
wH=β (I ∧ wH)+(1-β)wH;
Wherein, β is learning rate, β ∈ [0,1].
8. it is according to claim 6 based on Fuzzy ART divide Target cluster dividing method, it is characterised in that after the renewal
Current multiple targets number Ω=Ω+1, increase newly multiple targets weight vectors wΩ=I.
9. the Target cluster dividing method divided based on Fuzzy ART according to claim 2, it is characterised in that step (2) is described
Initialization multiple targets and its weight vectors, specially:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611186577.1A CN106599927B (en) | 2016-12-20 | 2016-12-20 | The Target cluster dividing method divided based on Fuzzy ART |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611186577.1A CN106599927B (en) | 2016-12-20 | 2016-12-20 | The Target cluster dividing method divided based on Fuzzy ART |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106599927A true CN106599927A (en) | 2017-04-26 |
CN106599927B CN106599927B (en) | 2019-05-17 |
Family
ID=58600453
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611186577.1A Active CN106599927B (en) | 2016-12-20 | 2016-12-20 | The Target cluster dividing method divided based on Fuzzy ART |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106599927B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109766905A (en) * | 2018-09-28 | 2019-05-17 | 中国人民解放军空军工程大学 | Target cluster dividing method based on Self-Organizing Feature Maps |
CN109993182A (en) * | 2017-12-29 | 2019-07-09 | 中移(杭州)信息技术有限公司 | A kind of mode identification method and device based on Fuzzy ART |
CN114567512A (en) * | 2022-04-26 | 2022-05-31 | 深圳市永达电子信息股份有限公司 | Network intrusion detection method, device and terminal based on improved ART2 |
CN117633563A (en) * | 2024-01-24 | 2024-03-01 | 中国电子科技集团公司第十四研究所 | Multi-target top-down hierarchical grouping method based on OPTICS algorithm |
CN117633563B (en) * | 2024-01-24 | 2024-05-10 | 中国电子科技集团公司第十四研究所 | Multi-target top-down hierarchical grouping method based on OPTICS algorithm |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663430B (en) * | 2012-04-17 | 2013-11-27 | 清华大学 | Object clustering method in situation assessment |
US20150205325A1 (en) * | 2014-01-20 | 2015-07-23 | Canon Kabushiki Kaisha | Apparatus for controlling clock signals to be supplied to a plurality of processing modules and control method thereof |
-
2016
- 2016-12-20 CN CN201611186577.1A patent/CN106599927B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663430B (en) * | 2012-04-17 | 2013-11-27 | 清华大学 | Object clustering method in situation assessment |
US20150205325A1 (en) * | 2014-01-20 | 2015-07-23 | Canon Kabushiki Kaisha | Apparatus for controlling clock signals to be supplied to a plurality of processing modules and control method thereof |
Non-Patent Citations (2)
Title |
---|
KESKIN G A等: "The Fuzzy ART algorithm:A categorization meth for supplier evaluation and selection", 《EXPERT SYSTEMS WITH APPLICATIONS》 * |
段同乐等: "二叉树多分类SVM在目标分群中的应用", 《专题技术与工程应用》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109993182A (en) * | 2017-12-29 | 2019-07-09 | 中移(杭州)信息技术有限公司 | A kind of mode identification method and device based on Fuzzy ART |
CN109993182B (en) * | 2017-12-29 | 2021-08-17 | 中移(杭州)信息技术有限公司 | Pattern recognition method and device based on Fuzzy ART |
CN109766905A (en) * | 2018-09-28 | 2019-05-17 | 中国人民解放军空军工程大学 | Target cluster dividing method based on Self-Organizing Feature Maps |
CN109766905B (en) * | 2018-09-28 | 2023-02-28 | 中国人民解放军空军工程大学 | Target grouping method based on self-organizing feature mapping network |
CN114567512A (en) * | 2022-04-26 | 2022-05-31 | 深圳市永达电子信息股份有限公司 | Network intrusion detection method, device and terminal based on improved ART2 |
CN117633563A (en) * | 2024-01-24 | 2024-03-01 | 中国电子科技集团公司第十四研究所 | Multi-target top-down hierarchical grouping method based on OPTICS algorithm |
CN117633563B (en) * | 2024-01-24 | 2024-05-10 | 中国电子科技集团公司第十四研究所 | Multi-target top-down hierarchical grouping method based on OPTICS algorithm |
Also Published As
Publication number | Publication date |
---|---|
CN106599927B (en) | 2019-05-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107871119B (en) | Target detection method based on target space knowledge and two-stage prediction learning | |
CN110287932B (en) | Road blocking information extraction method based on deep learning image semantic segmentation | |
CN109784203B (en) | Method for inspecting contraband in weak supervision X-ray image based on layered propagation and activation | |
CN102810161B (en) | Method for detecting pedestrians in crowding scene | |
CN104574445B (en) | A kind of method for tracking target | |
CN111079602A (en) | Vehicle fine granularity identification method and device based on multi-scale regional feature constraint | |
CN102110227B (en) | Compound method for classifying multiresolution remote sensing images based on context | |
CN105260738A (en) | Method and system for detecting change of high-resolution remote sensing image based on active learning | |
CN106688011A (en) | Method and system for multi-class object detection | |
CN105335986A (en) | Characteristic matching and MeanShift algorithm-based target tracking method | |
US9626761B2 (en) | Sampling method and image processing apparatus of CS-RANSAC for estimating homography | |
CN108492298A (en) | Based on the multispectral image change detecting method for generating confrontation network | |
CN110491132A (en) | Vehicle based on video frame picture analyzing, which is disobeyed, stops detection method and device | |
CN111832615A (en) | Sample expansion method and system based on foreground and background feature fusion | |
CN109740609A (en) | A kind of gauge detection method and device | |
CN104182985A (en) | Remote sensing image change detection method | |
CN110472572A (en) | The quick identification and classification method of naval target under a kind of complex environment | |
CN106599927A (en) | Target grouping method based on fuzzy ART division | |
CN103745233B (en) | The hyperspectral image classification method migrated based on spatial information | |
CN104318559A (en) | Quick feature point detecting method for video image matching | |
CN103839257A (en) | Method for detecting changes of SAR images of generalized Gaussian K&I | |
CN112102322A (en) | Fault identification method based on multi-mode U-Net | |
CN103886606B (en) | SAR image segmentation method based on joint generalized gamma distribution parameters | |
CN113393679A (en) | Regional traffic guidance method and system based on traffic intersection traffic flow identification and statistics | |
CN105205816A (en) | Method for extracting high-resolution SAR image building zone through multi-feature weighted fusion |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |