CN104504233B - A kind of abnormality recognition method based on multi-C vector entropy stochastical sampling - Google Patents
A kind of abnormality recognition method based on multi-C vector entropy stochastical sampling Download PDFInfo
- Publication number
- CN104504233B CN104504233B CN201410646085.0A CN201410646085A CN104504233B CN 104504233 B CN104504233 B CN 104504233B CN 201410646085 A CN201410646085 A CN 201410646085A CN 104504233 B CN104504233 B CN 104504233B
- Authority
- CN
- China
- Prior art keywords
- point
- sample
- vector
- sample point
- entropy
- 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.)
- Expired - Fee Related
Links
Abstract
The invention provides a kind of abnormality recognition method based on multi-C vector entropy stochastical sampling, the method is comprised the following steps:I, the sample point selection sampled point from sample space Ω, generate subsample space ω;II, the multi-C vector entropy for determining the sample point;III, repeat the above steps, determine the fusion results of the sample point multi-C vector entropy;IV, the intensity of anomaly for determining the sample point;V, determine abnormity point.The method is by merging the sample point of stochastical sampling, solve the problems such as sample size that faces of anomalous identification in large-scale data is big, dimension is high, the method can not only reduce the time complexity of anomalous identification, the accuracy of recognition effect be improved, also with stronger autgmentability.
Description
Technical field
It is in particular to a kind of to be based on multi-C vector entropy stochastical sampling the present invention relates to a kind of method in anomalous identification field
Abnormality recognition method.
Background technology
Anomalous identification refers to be found out from one group of related data with away from overall acnode or abnormity point, these abnormity points
It is the point for being both not belonging to cluster or be not belonging to ambient noise, they are often as produced by entirely different mechanism.Currently, it is abnormal
Identification has been widely used in telecommunication fraud, credit card abuse, loan and has examined as a kind of important data mining and analysis method
Batch, the analysis of drug research, medical analysis, consumer behaviour, weather forecast, financial field client segmentation, network invasion monitoring etc.
Field.
In the prior art, the method for anomalous identification is mainly included based on the abnormality recognition method for counting, based on the different of distance
Normal recognition methods and the abnormality recognition method based on density and four kinds of the abnormality recognition method based on cluster, introduce separately below.
(1) abnormality recognition method based on statistics
Statistical method is the method based on model, as one model of data creation, and according to object fitting model
Situation have assessing them and much may meet the model.From the eighties in 20th century, Identifying Outliers are in field of statistics
In be widely studied, usual user is modeled with certain statistical distribution to data point, then with assume model, according to point
It is distributed to determine whether exception.Such as, in statistics, it is assumed that data set Normal Distribution, those are inclined between average
Difference meets or exceeds 3 times of data objects of standard deviation can be referred to as exceptional value.According to this law, can derive a series of
Abnormality recognition method based on statistics.
Abnormality recognition method based on statistics often has the following disadvantages:First, the premise based on statistical method is necessary
Know what distribution data set obeys, if estimating that mistake has resulted in heavytailed distribution, influences recognition result;Secondly, the method
Single variable can only be recognized, i.e., identification can only be confined to single index every time, it is impossible to be analyzed with reference to multi objective, therefore cannot
Analysis high dimensional data.
(2) abnormality recognition method based on distance.
Abnormality recognition method based on distance is thought, if an object is away from most of point, then it is exactly abnormal.
This method than statistical method more typically, be easier to use because determine data set point to the distance between than determine it
Statistical distribution be easier.One intensity of anomaly score of object can generally be given by the distance of the K arest neighbors to it.Should
Method is generally more sensitive to the selection of arest neighbors number K, if K is too small, abnormal score may be inaccurate;If K is too big,
Then normal point is likely to be identified as abnormity point.The K average distance of arest neighbors can generally be chosen as abnormal score.
Abnormality recognition method based on statistics often has the following disadvantages:First, the time complexity of the method is more in Ο
(n2), it is difficult to suitable for large data sets;Secondly, selection of the method to parameter is more sensitive, easily influences final identification knot
Really;Simultaneously as the method is using global threshold value, therefore the data set with different densities region can not be processed.
(3) abnormality recognition method based on density.
From for the viewpoint based on density, abnormity point is the object in density regions.One intensity of anomaly of object
Usual score is the inverse of the data collection density.Identifying Outliers based on density and the close phase of anomalous identification based on distance
Close, because density is generally defined with the distance of arest neighbors, a kind of method of conventional definition density is that definition density is to K
The inverse of the average distance of individual arest neighbors.Distinguishingly, when packet contains the region of different densities, they can not correctly be recognized
Abnormity point, therefore subsequently extend local density's detection technique to judge abnormity point again,
Abnormality recognition method based on density often has the following disadvantages:First, as the method based on distance, the party
The time complexity of method is higher, and the treatment to large data sets is very difficult;Secondly, selection of the method to parameter is more sensitive,
Also final recognition result is easily influenceed.
(4) abnormality recognition method based on cluster
If an object does not belong to by force any cluster, then the object is defined as the abnormity point based on cluster.Using poly-
Class technology abnormity point, generally using the method for abandoning the tuftlet away from other clusters, this method can also be with other any clusters
Technology is used together, however it is necessary that between most tuftlet size and tuftlet and other clusters distance threshold value, therefore, it is to cluster number
Selection is extremely sensitive.If less cluster is also height cohesion, then the abnormality recognition method based on cluster will be unable to detection
Go out this abnormity point.But this method can be using based on noting abnormalities linearly and close to the clustering technique of linear complexity
Point, therefore time complexity is relatively low.
Abnormality recognition method based on cluster often has the following disadvantages:First, the quality pair of the cluster that clustering algorithm is produced
The quality influence of the abnormity point that the algorithm is produced is very big;Secondly, the abnormal point set and their score of generation may be very
Rely on the number and the existence of data abnormal point of cluster used.These can all increase the difficulty of anomalous identification.
To sum up, it can be seen that:Based on statistics anomalous identification application be mainly limited to scientific data statistics, this be mainly because
To must know in advance that the range of application of the distribution characteristics which limits it of data.Abnormality recognition method based on distance is with base
Compared in the abnormality recognition method of statistics, it is not necessary to which user possesses any domain knowledge.And, distance exception is closer to
The essential reason that abnormity point is formed.Anomalous identification based on density is a kind of extension of the abnormality recognition method based on distance, pin
Identification to local outlier is more efficient.Local anomaly identification then more conforms to real-life true application.Based on poly-
The anomalous identification technology of class tends to rely on the clustering result quality and time loss of the clustering algorithm of itself.
However, with the increase of sample data volume, bigger challenge, above-mentioned four kinds abnormal knowledges are also proposed to anomalous identification
Other method existence time expense is big, and the shortcomings of being limited is applied in higher dimensional space.Abnormality recognition method of the prior art is for small
Sample data set, time complexity is more in Ο (n2) or Ο (n3);And under large-scale data, this time overhead is likely difficult to connect
Receive.In addition, the increase of dimension also bring it is more and more diluter in another space of " dimension disaster " problem, i.e. data occupied by it
Dredge, the distance between sample point is almost equal, cause much to lose meaning based on distance and the parameter based on density.Cause
This, it is desirable to provide a kind of efficient, accurate Identifying Outliers method.
The content of the invention
To overcome above-mentioned the deficiencies in the prior art, the present invention to provide a kind of abnormal knowledge based on multi-C vector entropy stochastical sampling
Other method.
Realize solution that above-mentioned purpose used for:
A kind of abnormality recognition method based on multi-C vector entropy stochastical sampling, it is theed improvement is that:Methods described includes
Following steps:
I, the sample point selection sampled point from sample space Ω, generate subsample space ω;
II, the multi-C vector entropy for determining the sample point;
III, repeat the above steps, determine the fusion results of the sample point multi-C vector entropy;
IV, the intensity of anomaly for determining the sample point;
V, determine abnormity point.
Further, in the step I, the number N of the sample point in the sample space is determined;
Determine in the sample point with the method for samplingThe individual sampled point, institute is generated according to the sampled point
State subsample space ω.
Further, the method for sampling is stochastical sampling method.
Further, comprised the following steps in the step II:
The vector that S201, the sample point for determining the sample space Ω are constituted with the sampled point of the subsample space ω;
S202, determine each sample point to the multi-C vector entropy between each sampled point.
Further, each sample point to the multi-C vector entropy Φ between each sampled point is determined as the following formula
(A):
In formula, A is any sample point;Φ (A) is the multi-C vector entropy of sample point A;
It is vectorVectorial entropy;
It is any vector with the sample point A as origin and with each sampled point as terminal,
It is vectorMould;
D is the dimension of the sample point;
It is vectorJth dimension attribute, if property value be negative, take absolute value calculating multi-C vector entropy;
N is the number of sampled point in the subsample space ω,N is sample point in the sample space Ω
Number.
Further, comprised the following steps in the step III:
S301, number of repetition are K times, and K multi-C vector entropy is obtained for each sample point;The scope of K is 10≤K
≤20;
S302, the K multi-C vector entropy is merged using average value convergence strategy, determine the multi-C vector of the sample point
The average value of entropy as the sample point fusion value.
Further, K step I is repeated, the K subsample space ω is obtained and is met claimed below:
The union of the K sub- sample space ω is the complete or collected works of the sample space Ω.
Further, in the step IV, the fusion value of the sample point is sorted, according to the fusion value determines
The intensity of anomaly of sample point in sample space Ω.
Further, the score value of the fusion value is higher, and the intensity of anomaly of the sample point is higher, otherwise intensity of anomaly is got over
It is low.
Further, in the step V, sample described in the threshold decision of the multi-C vector entropy fusion value according to the sample point
This point is abnormity point or normal point;
If the multi-C vector entropy fusion value is more than or equal to threshold value, judge that the sample point is abnormity point, otherwise for normal
Point.
Compared with prior art, the invention has the advantages that:
1st, the method that the present invention is provided is based on stochastic sampling strategy, generation multiple subsamples space such that it is able to reduce different
The other time complexity of general knowledge.
2nd, the method that the present invention is provided is by building the multi-C vector between sample point and sampled point, and by calculating multidimensional
The entropy of vector is distributed to carry out anomalous identification, solves the problems, such as the dimension disaster of higher dimensional space.
3rd, the calculating multi-C vector entropy process of stochastic sampling strategy is relatively independent in the method that the present invention is provided, and is conducive to increasing
The autgmentability of strong this method.
4th, the method that the present invention is provided utilizes fusion type strategy, is calculated by the multi-C vector entropy for combining multiple stochastical sampling
As a result, the average value for calculating multiple multi-C vector entropys characterizes intensity of anomaly, increases the diversity that sample multi-C vector entropy is calculated.
5th, the method that the present invention is provided can provide the quantized value of all sample point intensity of anomalys, and according to its low degree high
Sequence, is conducive to increasing the discrimination of each sample point intensity of anomaly.
Brief description of the drawings
Fig. 1 is the abnormality recognition method flow chart based on multi-C vector entropy stochastical sampling in the present embodiment;
Fig. 2 be the present embodiment in from sample space determine subsample space schematic diagram;
Fig. 3 is the schematic diagram of fusion sample point multi-C vector entropy in the present embodiment;
Fig. 4 be the present embodiment in normal point be distributed in ellipsoidal surfaces, abnormity point obey volume more than spheroid normal distribution
Schematic diagram;
Fig. 5 is during normal point is distributed in spheroid in the present embodiment, abnormity point obeys normal distribution of the volume more than spheroid
Schematic diagram;
Fig. 6 is the distribution situation schematic diagram of the multi-C vector entropy that all-pair is answered in Fig. 4 for being given in the present embodiment
Fig. 7 is the distribution situation schematic diagram of the multi-C vector entropy that all-pair is answered in Fig. 5 for being given in the present embodiment.
Specific embodiment
Specific embodiment of the invention is described in further detail below in conjunction with the accompanying drawings.
To realize concentrating various dimensions, large-scale data efficient, the accurate identification of abnormal data, the present embodiment provides one
Plant the abnormality recognition method based on multi-C vector entropy stochastical sampling.
As shown in figure 1, the abnormality recognition method flow chart based on multi-C vector entropy stochastical sampling in the present embodiment;The method
The intensity of anomaly of all sample points in sample space can be recognized.The method is comprised the following steps:
Step one, in the sample point of sample space choose sampled point.
Step 2, the multi-C vector entropy for determining sample point in the sample space.
Step 3, repeat the above steps one and step 2 K times, determines the fusion results of the sample point multi-C vector entropy.
Step 4, the intensity of anomaly for determining the sample point.
In step one, in sample space Ω, including N number of sample point, chosen from sample point with the method for samplingIt is individual
Sampled point, generates subsample space ω.
The method of sampling selects stochastical sampling method in the present embodiment.
In step 2, the multi-C vector entropy of the sampled point is determined, can realize in two steps.
First, the vector that the sample point of the sample space Ω is constituted with the sampled point of the subsample space ω is determinedThen, it is determined that multi-C vector entropy of all sample points to each sampled point.
Determine the multi-C vector entropy of all sample points, comprise the following steps:
Determine any sample point A, determine in sample space sample point A to the multi-C vector that vector is constituted between sampled point
Entropy Φ (A):
In formula,It is any vector with each sample point as origin and with each sampled point as terminal, and
It is vector field homoemorphism;
N is the number of sampled point in the subsample space,N is the sample point number of sample space;;
D is the dimension of sample point in the sample space;
It is vectorJth dimension attribute, if property value be negative, take its absolute value calculate multi-C vector entropy;
It is vectorMulti-C vector entropy.
In step 3, repeat the above steps one and two K time, determine the fusion results of the sample point multi-C vector entropy.Tool
Body is comprised the following steps:
First, return to step one and step 2, calculate the multi-C vector entropy of sample point, judge whether to reach K times, if reaching
Into step 4, otherwise return to step one continues to calculate.By the above method, for each sample point, K can be obtained individual many
Dimensional vector entropy;The scope of K is 10≤K≤20.
Repeat step one K times, expression can obtain K sub- sample space, it is necessary to explanation, obtains the K subsample
Space ω meets claimed below:
The union of the K sub- sample space ω is the complete or collected works of the sample space Ω.
Then, the K multi-C vector entropy is merged using average value convergence strategy, determines the multi-C vector of the sample point
The average value of entropy as the sample point fusion value.
In step 4, the intensity of anomaly of sample point in the sample space is determined.
The method is:The fusion value of the multi-C vector entropy of each sample point is ranked up, according to the fusion value
Determine the intensity of anomaly of sampled point in the sample space.
The score value of fusion value is higher, and the intensity of anomaly of the sample point is higher, conversely, the score value of fusion value is lower, sample
The intensity of anomaly of point is lower.
The intensity of anomaly of all sample points in sample space is determined by above-mentioned steps one to step 4, by step 5,
Abnormity point is determined according to intensity of anomaly.Method is:Described in the threshold decision of the multi-C vector entropy fusion value according to the sample point
Sample point is abnormity point or normal point.
If the multi-C vector entropy fusion value is more than or equal to threshold value, judge that the sample point is abnormity point, if conversely, described
Multi-C vector entropy fusion value is less than threshold value, judges that the sample point is normal point.
For the threshold value for determining multi-C vector entropy fusion value, there is provided two methods:
1st, according to the requirement of Identifying Outliers in different method or system, the threshold of the multi-C vector entropy of sample point is determined
Value.
It should be noted that due to the requirement of different method or system, threshold value selects different values.In general, it is accurate
Exactness requirement is higher, and threshold value value is smaller.
2nd, the multi-C vector entropy fusion value according to all sample points in sample space further determine that the judgement of abnormity point according to
According to method is:Overall average is further calculated to all fusion values, in this, as basis for estimation.
Application Example one
Assuming that in a sample space, data set is D={ a1,a2,...,a30, sample size N is 30, and attribute dimension d is
5.Determine abnormity point with the method for the present invention.Comprise the following steps that:
Step one, sampled point is determined in sample space.
It is sampled point number in stochastical sampling point, i.e. subsample space to randomly select the sample point in 10% data setGeneration subsample space, is expressed as D*={ s1,s2,s3, s1,s2,s3Respectively three sampled points.
Setting s1={ 2,3,1,2,3 }, s2={ 4,3,2,2,1 }, s3={ 3,2,1,1,2 }.
Step 2, the multi-C vector entropy for determining the sample point.Specifically may include following two step:
S201, determine respectively in the sample space in each sample point and the subsample space between each sampled point to
Amount.
In the present embodiment, it is first determined any sample point A={ 3,1,2,1,2 } in sample space, it is determined that being with sample point A
Origin, with sampled point s1={ 2,3,1,2,3 } are the vector of terminal
Then, it is determined that sample point A to s2={ 4,3,2,2,1 } and s3The vector of={ 3,2,1,1,2 };
Finally, all sample points and all sampled points in the space of subsample in sample space can determine that according to the above method
Vector;
S202, each sample point to the vector distribution relation Φ between each sampled point is determined respectively, the vector point
Cloth relation as the sample point to the sampled point multi-C vector entropy.
With above-mentioned vectorAs a example by, shown determination sample point A to sampled point s as the following formula1Multi-C vector
Entropy:
It should be noted that(wherein, j=1,2 ..., 5) represent vectorJth dimension attribute, if property value is negative
Number, then take its absolute value and calculate multi-C vector entropy.
It is computed, determines multi-C vector entropy Φ (A)=2.158 of sample point.
Step 3, repeat the above steps, determine the fusion results of the sample point multi-C vector entropy.Specifically include following step
Suddenly:
First, repeat step one and step 2 K times, 10≤K≤20, K takes 10 in the present embodiment.
It is repeated 10 times, expression can obtain 10 sub- sample spaces, 10 subsample space ω meet claimed below:
The union of 10 sub- sample space ω is the complete or collected works of the sample space Ω.
Then, 10 multi-C vector entropys are merged using average value convergence strategy, determine the multidimensional of the sample point to
The average value of entropy is measured as the fusion value of the sample point.
Assuming that for sample point A={ 3,1,2,1,2 }, the multi-C vector entropy of 10 calculating be respectively 3,1,2,1,2,3,
4,3,2,2 }, then 10 average values of multi-C vector entropy for calculating the sample point are 2.3.Above-mentioned average value characterizes sample point A
Intensity of anomaly.
Step 4, the intensity of anomaly for determining the sample point.
The fusion value of the multi-C vector entropy of each sample point is ranked up, the sample is determined according to the fusion value
The intensity of anomaly of sampled point in this space.
The score value of fusion value is higher, and the intensity of anomaly of the sample point is higher, conversely, the score value of fusion value is lower, sample
The intensity of anomaly of point is lower.
Step 5, determine abnormity point.
The threshold value of the multi-C vector entropy fusion value according to the sample point determines the abnormity point of sample space.If the sample
The multi-C vector entropy fusion value of point is more than or equal to the threshold value, then judge that the sample point is abnormity point;Otherwise it is judged to just
Chang Dian.
For example, the multi-C vector entropy fusion value of 3 points is respectively 10,5,2, and if the threshold value that user sets is 6, only
Any is abnormal;If the threshold value of user's setting is the 3, the 1st, 2 points be abnormal;If with setting threshold value be 1, a little
All it is abnormal.
Application Example two
As shown in Figure 4,5, Fig. 4,5 are respectively normal point and are distributed in ellipsoidal surfaces, abnormity point obedience volume more than spheroid
During normal distribution schematic diagram and normal point are distributed in spheroid, abnormity point obey normal distribution schematic diagram of the volume more than spheroid.
Fig. 4,5 are simulated in three-dimensional space under different scenes, the distribution situation of elliposoidal data set, plus sige in figure
Normal point is represented, it is square to represent abnormity point, altogether comprising normal point 200, abnormity point 20, totally 220 data points.
Identifying Outliers are carried out with the method for the present invention.First, in the corresponding elliposoidal data set of every kind of scene, with
Machine chooses 10% point, i.e., 22 points generate corresponding subsample space.
Then, respectively for every kind of scene, all sample points in the scene are traveled through, calculates the multi-C vector of each sample point
Entropy.Finally, the distribution situation of multi-C vector entropy under every kind of scene is drawn.
As shown in Figure 6,7, Fig. 6 gives the distribution situation of the multi-C vector entropy that all-pair in Fig. 4 is answered, and Fig. 7 gives
The distribution situation of the multi-C vector entropy that all-pair is answered in Fig. 5.As can be seen that abnormity point multi-C vector entropy from above-mentioned two figure
Value apparently higher than the multi-C vector entropy of normal point value.
Finally it should be noted that:Above example is merely to illustrate the technical scheme of the application rather than to its protection domain
Limitation, although being described in detail to the application with reference to above-described embodiment, those of ordinary skill in the art should
Understand:Those skilled in the art read still can be to applying after the application specific embodiment carry out a variety of changes, modification or
Person's equivalent, but these changes, modification or equivalent, are applying within pending claims.
Claims (7)
1. a kind of abnormality recognition method based on multi-C vector entropy stochastical sampling, it is characterised in that:Methods described includes following step
Suddenly:
I, the sample point selection sampled point from sample space Ω, generate subsample space ω;
II, the multi-C vector entropy for determining the sample point;
Comprised the following steps in the step II:
The vector that S201, the sample point for determining the sample space Ω are constituted with the sampled point of the subsample space ω;
S202, determine each sample point to the multi-C vector entropy between each sampled point;
Each sample point to the multi-C vector entropy Φ (A) between each sampled point is determined as the following formula:
In formula, A is any sample point;Φ (A) is the multi-C vector entropy of sample point A;
It is vectorVectorial entropy;
It is any vector with the sample point A as origin and with each sampled point as terminal,
It is vectorMould;
D is the dimension of the sample point;
It is vectorJth dimension attribute, if property value be negative, take absolute value calculating multi-C vector entropy;
N is the number of sampled point in the subsample space ω,N be in the sample space Ω sample point
Number;
III, repeat the above steps, determine the fusion results of the sample point multi-C vector entropy;
Comprised the following steps in the step III:
S301, number of repetition are K times, and K multi-C vector entropy is obtained for each sample point;The scope of K be 10≤K≤
20;
S302, the K multi-C vector entropy is merged using average value convergence strategy, determine the multi-C vector entropy of the sample point
Average value as the sample point fusion value;
IV, the intensity of anomaly for determining the sample point;
V, determine abnormity point.
2. a kind of abnormality recognition method based on multi-C vector entropy stochastical sampling as claimed in claim 1, it is characterised in that:Institute
State in step I, determine the number N of the sample point in the sample space;
Determine in the sample point with the method for samplingThe individual sampled point, the son is generated according to the sampled point
Sample space ω.
3. a kind of abnormality recognition method based on multi-C vector entropy stochastical sampling as claimed in claim 2, it is characterised in that:Institute
The method of sampling is stated for stochastical sampling method.
4. a kind of abnormality recognition method based on multi-C vector entropy stochastical sampling as claimed in claim 1, it is characterised in that:Weight
Multiple K step I, obtains the K subsample space ω and meets claimed below:
The union of the K sub- sample space ω is the complete or collected works of the sample space Ω.
5. a kind of abnormality recognition method based on multi-C vector entropy stochastical sampling as claimed in claim 1, it is characterised in that:Institute
In stating step IV, the fusion value of the sample point is sorted, sample point in the sample space Ω is determined according to the fusion value
Intensity of anomaly.
6. a kind of abnormality recognition method based on multi-C vector entropy stochastical sampling as claimed in claim 5, it is characterised in that:Institute
The score value for stating fusion value is higher, and the intensity of anomaly of the sample point is higher, otherwise intensity of anomaly is lower.
7. a kind of abnormality recognition method based on multi-C vector entropy stochastical sampling as claimed in claim 1, it is characterised in that:Institute
State in step V, sample point described in the threshold decision of the multi-C vector entropy fusion value according to the sample point is abnormity point or normal
Point;
If the multi-C vector entropy fusion value is more than or equal to threshold value, judge that the sample point is abnormity point, otherwise be normal point.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410646085.0A CN104504233B (en) | 2014-11-14 | 2014-11-14 | A kind of abnormality recognition method based on multi-C vector entropy stochastical sampling |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410646085.0A CN104504233B (en) | 2014-11-14 | 2014-11-14 | A kind of abnormality recognition method based on multi-C vector entropy stochastical sampling |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104504233A CN104504233A (en) | 2015-04-08 |
CN104504233B true CN104504233B (en) | 2017-06-06 |
Family
ID=52945630
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410646085.0A Expired - Fee Related CN104504233B (en) | 2014-11-14 | 2014-11-14 | A kind of abnormality recognition method based on multi-C vector entropy stochastical sampling |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104504233B (en) |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TWM391701U (en) * | 2009-10-30 | 2010-11-01 | Wei-Lieh Hsu | A traffic detection system |
KR20130136251A (en) * | 2012-06-04 | 2013-12-12 | 한국전자통신연구원 | Method and apparatus for situation recognition using object energy function |
CN103281293A (en) * | 2013-03-22 | 2013-09-04 | 南京江宁台湾农民创业园发展有限公司 | Network flow rate abnormity detection method based on multi-dimension layering relative entropy |
CN103441982A (en) * | 2013-06-24 | 2013-12-11 | 杭州师范大学 | Intrusion alarm analyzing method based on relative entropy |
CN103577835B (en) * | 2013-08-02 | 2016-08-10 | 中国科学技术大学苏州研究院 | The method using the multidimensional characteristic vectors detection hidden channel of IP ID |
CN104123544B (en) * | 2014-07-23 | 2018-03-13 | 通号通信信息集团有限公司 | Anomaly detection method and system based on video analysis |
-
2014
- 2014-11-14 CN CN201410646085.0A patent/CN104504233B/en not_active Expired - Fee Related
Also Published As
Publication number | Publication date |
---|---|
CN104504233A (en) | 2015-04-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103473786B (en) | Gray level image segmentation method based on multi-objective fuzzy clustering | |
CN106599230A (en) | Method and system for evaluating distributed data mining model | |
CN104239862B (en) | A kind of face identification method | |
CN104063719A (en) | Method and device for pedestrian detection based on depth convolutional network | |
CN110322453A (en) | 3D point cloud semantic segmentation method based on position attention and auxiliary network | |
CN108256482A (en) | A kind of face age estimation method that Distributed learning is carried out based on convolutional neural networks | |
CN107038292A (en) | A kind of many output of wind electric field correlation modeling methods based on adaptive multivariable nonparametric probability | |
CN103888541B (en) | Method and system for discovering cells fused with topology potential and spectral clustering | |
CN111950515B (en) | Semantic feature pyramid network-based small face detection method | |
CN108090515B (en) | Data fusion-based environment grade evaluation method | |
CN107391670A (en) | A kind of mixing recommendation method for merging collaborative filtering and user property filtering | |
CN110874577B (en) | Automatic verification method of certificate photo based on deep learning | |
CN108171119B (en) | SAR image change detection method based on residual error network | |
CN101369316A (en) | Image characteristics extraction method based on global and local structure amalgamation | |
CN106407908A (en) | Training model generation method and human face detection method and device | |
CN110472634A (en) | Change detecting method based on multiple dimensioned depth characteristic difference converged network | |
CN110533116A (en) | Based on the adaptive set of Euclidean distance at unbalanced data classification method | |
CN112085072A (en) | Cross-modal retrieval method of sketch retrieval three-dimensional model based on space-time characteristic information | |
CN103903238A (en) | Method for fusing significant structure and relevant structure of characteristics of image | |
CN103902798B (en) | Data preprocessing method | |
KR20210100592A (en) | Face recognition technology based on heuristic Gaussian cloud transformation | |
CN102194106B (en) | Human face recognition method used in door access system | |
CN110472671A (en) | Based on multistage oil-immersed transformer fault data preprocess method | |
CN114821299B (en) | Remote sensing image change detection method | |
CN104899604A (en) | Feature-level fusion method based on data set merging |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170606 Termination date: 20191114 |