CN106022368B - A method of the increment track abnormality detection based on increment core principle component analysis - Google Patents

A method of the increment track abnormality detection based on increment core principle component analysis Download PDF

Info

Publication number
CN106022368B
CN106022368B CN201610325491.6A CN201610325491A CN106022368B CN 106022368 B CN106022368 B CN 106022368B CN 201610325491 A CN201610325491 A CN 201610325491A CN 106022368 B CN106022368 B CN 106022368B
Authority
CN
China
Prior art keywords
nucleus lesion
track
increment
data
nucleus
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
Application number
CN201610325491.6A
Other languages
Chinese (zh)
Other versions
CN106022368A (en
Inventor
张磊
樊庆富
刘磊军
鲍苏宁
张国兴
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China University of Mining and Technology CUMT
Original Assignee
China University of Mining and Technology CUMT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by China University of Mining and Technology CUMT filed Critical China University of Mining and Technology CUMT
Priority to CN201610325491.6A priority Critical patent/CN106022368B/en
Publication of CN106022368A publication Critical patent/CN106022368A/en
Application granted granted Critical
Publication of CN106022368B publication Critical patent/CN106022368B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting

Abstract

A method of the increment track abnormality detection based on increment core principle component analysis belongs to the method for increment track abnormality detection.This method: the initialization of progress model first calculates, and carries out initial nucleus lesion calculating using traditional Batch KPCA, whenever having M item to increase track data arrival newly, is first standardized to this M track data；Then the nucleus lesion of newly-increased data is calculated using Batch KPCA；The average reconstruction error of newly-increased data and training data is calculated separately, if the two error is greater than given threshold values, subsequent nucleus lesion segmentation-merging method is executed, updates nucleus lesion；Then updated nucleus lesion is projected, extracts principal component；Finally unsupervised learning and abnormality detection are carried out using one-class support vector machine.Advantage: this method is better than traditional core principle component analysis method, reduces computational complexity, improves the efficiency of track abnormality detection.

Description

A method of the increment track abnormality detection based on increment core principle component analysis
Technical field
It is especially a kind of based on increment core principle component analysis the present invention relates to a kind of method of increment track abnormality detection The method of increment track abnormality detection.
Background technique
Track data includes the various features such as geographical position coordinates, speed, direction, is considered as a kind of high dimensional data.Core Principal component analysis is a kind of nonlinear principal component analysis track method for detecting abnormality, by Nonlinear Mapping by track data from Original data space is mapped in high-dimensional feature space, is then carried out in high-dimensional feature space with linear principal component analysis special Sign is extracted.But computation complexity of the core principle component analysis when carrying out nuclear matrix feature decomposition is O (N3), is seriously affected big Application on scale data collection.Incremental learning mode is introduced to reduce time complexity be the key that improve such method.
Summary of the invention
The invention aims to provide a kind of method of increment track abnormality detection based on increment core principle component analysis, Solve the problems, such as that the computation complexity of existing core principle component analysis method is high.
The object of the present invention is achieved like this: this method:
The initialization for carrying out model first calculates, and carries out initial nucleus lesion calculating using traditional Batch KPCA, Whenever having M item to increase track data arrival newly, first this M track data is standardized；
Then the nucleus lesion of newly-increased data is calculated using Batch KPCA；
The average reconstruction error of newly-increased data and training data is calculated separately, if the two error is greater than given threshold values, Subsequent nucleus lesion segmentation-merging method is executed, nucleus lesion is updated；
Then updated nucleus lesion is projected, extracts principal component；
Finally unsupervised learning and abnormality detection are carried out using one-class support vector machine；
Specific step is as follows for this method:
(1) it is somebody's turn to do the increment track method for detecting abnormality based on increment core principle component analysis, it is necessary first to set sliding window Size P and the trace number M that updates every time；P represents the size for the nucleus lesion for needing to update every time, nucleus lesion Size immobilizes during algorithm executes；M represents the size of each increment；
(2) the initial nucleus lesion model and meter of sliding data window are then calculated using traditional Batch KPCA Calculate its average reconstruction errorThe newly-increased track data vector of circulation batch processing later；Handling newly-increased track data When vector, its nucleus lesion model is first constructed, calculates its average reconstruction error
(3) its average reconstruction error ratio ε between sliding window nucleus lesion is then calculatedratio；Work as εratioIt is high When given threshold values v, sliding window nucleus lesion is updated using nucleus lesion framing shadowgraphy, first uses core feature Space dividing method removes M earliest track data feature vector from sliding data window, reduces nucleus lesion；Then M newly-increased track data vector is merged into sliding window nucleus lesion using nucleus lesion merging method；
(4) projection of sliding window nucleus lesion is calculated simultaneously, is sought principal component and is calculated its eigenspace projection；
(5) unsupervised learning and abnormality detection, record finally are carried out using principal component of the one-class support vector machine to extraction The abnormal track detected；After having detected, needs to recalculate sliding window nucleus lesion and be averaged reconstruction errorTrack data is increased newly next time to handle.
Beneficial effect as the above scheme is adopted should the increment track abnormality detection based on increment core principle component analysis Method updates nucleus lesion data model using nucleus lesion framing shadowgraphy.Maintain the cunning of a fixed size Dynamic data window first removes earliest from sliding data window nucleus lesion model whenever having the newly-increased track of M item to arrive M track data, then M newly-increased track data is merged into nucleus lesion；Only need to calculate the core feature of M track Space incrementally updates nucleus lesion on the basis of original sliding data window nucleus lesion, avoids each update When will recalculate the deficiency of nucleus lesion, the computation complexity for solving existing core principle component analysis method high is asked Topic, has reached the purpose of the present invention.
Advantage: this method is better than traditional core principle component analysis method, reduces computational complexity, improves track exception The efficiency of detection.
Detailed description of the invention
Fig. 1 is that the present invention is based on the flow charts of the increment track method for detecting abnormality frame of increment core principle component analysis.
Fig. 2 is the contrast effect figure of the present invention with the increment track method for detecting abnormality based on traditional KPCA.
Fig. 3 is abnormality detection effect picture of the present invention.
Fig. 4 is abnormality detection effect picture of the present invention.
Specific embodiment
Below in conjunction with the accompanying drawings and embodiment the invention will be further described.
Embodiment 1: the initialization of progress model first calculates, and it is empty to carry out incipient nucleus feature using traditional Batch KPCA Between calculate, whenever have M item increase newly track data arrive when, first this M track data is standardized；Then Batch is used KPCA calculates the nucleus lesion of newly-increased data.The average reconstruction error of newly-increased data and training data is calculated separately, if two Person's error is greater than given threshold values, then executes subsequent nucleus lesion segmentation-merging method, updates nucleus lesion；Then right Updated nucleus lesion is projected, and principal component is extracted；Finally unsupervised learning is carried out using one-class support vector machine And abnormality detection.
It is shown in Figure 1, a kind of increment track method for detecting abnormality based on increment core principle component analysis, including following step It is rapid:
(1) initial track data, track standardization；
(2) the sliding data window for determining a fixed size, calculates initial nucleus lesion and reconstruction error；
(3) the average reconstruction error ratio between newly-increased data and sliding window is calculated, is given if the two error is greater than Threshold values then executes nucleus lesion segmentation-merging method, updates nucleus lesion；
(4) it calculates sliding window nucleus lesion after updating to project, extracts principal component；
(5) unsupervised learning and abnormality detection are carried out using one-class support vector machine；
The specific method is as follows:
The increment track method for detecting abnormality based on increment core principle component analysis uses Min-max method standard first Change every track, the trace number M for setting the size P of sliding window and updating every time.It is special that P represents the core for needing to update every time The size in space is levied, nucleus lesion size immobilizes during algorithm executes；M represents the size of each increment；
Then initial nucleus lesion model and the calculating of sliding data window are calculated using traditional Batch KPCA Its average reconstruction errorThe input vector t of one n dimension is mapped as the vector φ (t) of l dimension by kernel function；Weight Building error ε is exactlyWith it in the squared-distance in nucleus lesion between projection, whereinMapping after centralization to It measures φ (t)；The newly-increased track data vector of circulation batch processing later；When handling newly-increased track data vector, it is first constructed Nucleus lesion model calculates its average reconstruction errorAverage reconstruction error ratio εratioIt is M newly-increased track number According to average reconstruction error and sliding data window in ratio between training data ensemble average reconstruction error.Specific formula for calculation For
Then its average reconstruction error ratio ε between sliding window nucleus lesion is calculatedratio；Work as εratioIt is higher than When given threshold values v, sliding window nucleus lesion is updated using nucleus lesion framing shadowgraphy, it is first empty using core feature Between dividing method remove M earliest track data feature vector from sliding data window, reduce nucleus lesion, core feature Space dividing method is the feature space dividing method based on original input space, and coring is carried out to it and obtains being suitable for increment core The nucleus lesion dividing method of principal component analysis；Then the M track data that will be increased newly using nucleus lesion merging method Vector is merged into sliding window nucleus lesion, and nucleus lesion merges i.e. after nucleus lesion is divided, from sliding number According to dividing in window nucleus lesion and obtain nucleus lesion model Ω=(U, the Φ being made of residual trackx, α, Λ, N), The nucleus lesion model that M newly-increased track is constituted is Θ=(V, Φy, β, Δ, M), merging Ω and Θ obtains updated Nucleus lesion model Q=(W, Φz,τ,∏,P)。
The projection of sliding window nucleus lesion is calculated simultaneously, is sought principal component and is calculated its eigenspace projection；
Unsupervised learning and abnormality detection finally are carried out using principal component of the one-class support vector machine to extraction, traverses track Set judges abnormal track using decision function, records the abnormal track detected for every track sample in data set, And corresponding track label is set；After having detected, needs to recalculate sliding window nucleus lesion and be averaged reconstruction errorTrack data is increased newly next time to handle.
Embodiment 2: the present invention is compared with the increment track Outlier Detection Algorithm based on tradition KPCA (Batch KPCA)；
In order to verify effectiveness of the invention, 7270 in Atlantic hurricane data between nineteen ninety to 2006 are chosen Tracing point amounts to 221 tracks and is verified as experimental data set.Incremental track as can be seen from Figure 2 based on traditional KPCA The execution time of mark method for detecting abnormality is quickling increase with the increase of sliding data window.And based on nucleus lesion point The execution time for the increment track Outlier Detection Algorithm for cutting-merging can also increase with the increase of sliding data window, still Amplitude is not big.And in the case where identical sliding data window size, based on nucleus lesion segmentation-merging incremental track Mark Outlier Detection Algorithm is less than the calculating time used in the increment track method for detecting abnormality based on traditional KPCA, and with The increase of data window is slided, this species diversity can be increasing.
Embodiment 3: abnormality detection effect picture of the invention；
The testing result of the method for the present invention the 4th iteration and the 6th iteration on Atlantic hurricane data set is chosen, is such as schemed Shown in 3 and Fig. 4.Thicker lines represent the abnormal track detected in figure, and thinner lines represent normal trace.It can from figure To find out, abnormality detection effect is fine, and the track of many behavior abnormalities all is checked out out.And from the result of iteration twice As can be seen that being continuously added with newly-increased track, some new abnormal tracks are detected, and since nucleus lesion exists It constantly updates, some previous abnormal tracks are ignored.

Claims (1)

1. a kind of method of the increment track abnormality detection based on increment core principle component analysis, it is characterized in that:
The initialization for carrying out model first calculates, and carries out initial nucleus lesion calculating using traditional Batch KPCA, whenever When having M item to increase track data arrival newly, first this M track data is standardized；
Then the nucleus lesion of newly-increased data is calculated using Batch KPCA；
The average reconstruction error of newly-increased data and training data is calculated separately, if the two error is greater than given threshold values, is executed Subsequent nucleus lesion segmentation-merging method updates nucleus lesion；
Then updated nucleus lesion is projected, extracts principal component；
Finally unsupervised learning and abnormality detection are carried out using one-class support vector machine；
Specific step is as follows:
(1) it is somebody's turn to do the increment track method for detecting abnormality based on increment core principle component analysis, it is necessary first to set the big of sliding window The small P and trace number M updated every time；P represents the size for the nucleus lesion for needing to update every time, nucleus lesion size It immobilizes during algorithm executes；Trace number M represents the size of each increment；
(2) the initial nucleus lesion model of sliding data window then is calculated using traditional Batch KPCA and calculate it Average reconstruction errorThe newly-increased track data vector of circulation batch processing later；Handling newly-increased track data vector When, its nucleus lesion model is first constructed, its average reconstruction error is calculated
(3) its average reconstruction error ratio ε between sliding window nucleus lesion is then calculatedratio；Work as εratioHigher than giving When determining threshold values v, sliding window nucleus lesion is updated using nucleus lesion framing shadowgraphy, first uses nucleus lesion Dividing method removes M track data feature vector of earliest trace number from sliding data window, reduces nucleus lesion； Then newly-increased M track data vector of trace number is merged by sliding window core feature using nucleus lesion merging method In space；
(4) projection of sliding window nucleus lesion is calculated simultaneously, is sought principal component and is calculated its eigenspace projection；
(5) unsupervised learning and abnormality detection, record detection finally are carried out using principal component of the one-class support vector machine to extraction The abnormal track arrived；After having detected, needs to recalculate sliding window nucleus lesion and be averaged reconstruction errorWith Just it handles and increases track data newly next time.
CN201610325491.6A 2016-05-17 2016-05-17 A method of the increment track abnormality detection based on increment core principle component analysis Active CN106022368B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610325491.6A CN106022368B (en) 2016-05-17 2016-05-17 A method of the increment track abnormality detection based on increment core principle component analysis

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610325491.6A CN106022368B (en) 2016-05-17 2016-05-17 A method of the increment track abnormality detection based on increment core principle component analysis

Publications (2)

Publication Number Publication Date
CN106022368A CN106022368A (en) 2016-10-12
CN106022368B true CN106022368B (en) 2019-04-05

Family

ID=57097096

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610325491.6A Active CN106022368B (en) 2016-05-17 2016-05-17 A method of the increment track abnormality detection based on increment core principle component analysis

Country Status (1)

Country Link
CN (1) CN106022368B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109165550B (en) * 2018-07-13 2021-06-29 首都师范大学 Multi-modal surgery track rapid segmentation method based on unsupervised deep learning
CN109726737B (en) * 2018-11-27 2020-11-10 武汉极意网络科技有限公司 Track-based abnormal behavior detection method and device
CN111666316B (en) * 2020-06-19 2023-09-15 南京大学 Isolation distribution core construction method, abnormal data detection method and device

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103020643A (en) * 2012-11-30 2013-04-03 武汉大学 Classification method based on kernel feature extraction early prediction multivariate time series category
CN104915671A (en) * 2015-06-23 2015-09-16 中国矿业大学 FGAK (Fast Global Alignment Kernels) based abnormal trajectory detection method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103020643A (en) * 2012-11-30 2013-04-03 武汉大学 Classification method based on kernel feature extraction early prediction multivariate time series category
CN104915671A (en) * 2015-06-23 2015-09-16 中国矿业大学 FGAK (Fast Global Alignment Kernels) based abnormal trajectory detection method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Incremental Kernel Principal Component Analysis;Tat-Jun Chin and David Suter;《IEEE TRANSACTIONS ON IMAGE PROCESSING》;20070731;第16卷(第6期);第I部分，第IV部分 *
基于核主成分分析的异常轨迹检测方法;鲍苏宁 等;《计算机应用技术》;20140710;第34卷(第7期);摘要，第1.1节，第1.2节，第1.3节，第1.4节 *

Also Published As

Publication number Publication date
CN106022368A (en) 2016-10-12

Similar Documents

Publication Publication Date Title
CN105069413B (en) A kind of human posture's recognition methods based on depth convolutional neural networks
CN107316067B (en) A kind of aerial hand-written character recognition method based on inertial sensor
Li et al. Fabric defect detection based on biological vision modeling
CN103810288B (en) Method for carrying out community detection on heterogeneous social network on basis of clustering algorithm
CN103544483B (en) A kind of joint objective method for tracing based on local rarefaction representation and system thereof
CN105426929B (en) Object shapes alignment device, object handles devices and methods therefor
WO2017024691A1 (en) Analogue circuit fault mode classification method
CN105787448A (en) Facial shape tracking method based on space-time cascade shape regression
CN106022368B (en) A method of the increment track abnormality detection based on increment core principle component analysis
CN106204638A (en) A kind of based on dimension self-adaption with the method for tracking target of taking photo by plane blocking process
CN105608318A (en) Method for integrating crowdsourced annotations
CN107103326A (en) The collaboration conspicuousness detection method clustered based on super-pixel
CN104573621A (en) Dynamic gesture learning and identifying method based on Chebyshev neural network
CN110232308A (en) Robot gesture track recognizing method is followed based on what hand speed and track were distributed
CN109598220A (en) A kind of demographic method based on the polynary multiple dimensioned convolution of input
CN109934095A (en) A kind of remote sensing images Clean water withdraw method and system based on deep learning
CN104537686A (en) Tracing method and device based on target space and time consistency and local sparse representation
CN108898623A (en) Method for tracking target and equipment
CN102708294A (en) Self-adaptive parameter soft measuring method on basis of semi-supervised local linear regression
CN109325510A (en) A kind of image characteristic point matching method based on lattice statistical
CN104517123B (en) A kind of Subspace clustering method guided using local motion feature similitude
CN110008847B (en) Swimming stroke identification method based on convolutional neural network
CN104036528A (en) Real-time distribution field target tracking method based on global search
CN109657693B (en) Classification method based on correlation entropy and transfer learning
CN109886409B (en) Quantitative causal relationship judging method of multidimensional time sequence

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