CN110363115A - The extremely semi-supervised real-time detection method of shipping work based on AIS track data - Google Patents
The extremely semi-supervised real-time detection method of shipping work based on AIS track data Download PDFInfo
- Publication number
- CN110363115A CN110363115A CN201910574738.1A CN201910574738A CN110363115A CN 110363115 A CN110363115 A CN 110363115A CN 201910574738 A CN201910574738 A CN 201910574738A CN 110363115 A CN110363115 A CN 110363115A
- Authority
- CN
- China
- Prior art keywords
- operating area
- stage
- real
- point
- supervised
- 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
- G06F18/21—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
-
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Probability & Statistics with Applications (AREA)
- Alarm Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of extremely semi-supervised real-time detection methods of shipping work based on AIS track data.During harbour or river dredging operation, mud is not transported to specify by some operation ships as expected topples over region, but is thrown near operating area, it means that mud may put back to port in a short time or navigation channel.The invention proposes a kind of extremely semi-supervised detection methods of dredging operation for being based on automatic recognition system (AIS) data.It is used to extract behavioural characteristic from AIS data firstly, the present invention establishes a tag system.In addition, also T- distribution random neighborhood insertion (T-SNE) is combined by neural network with gauss hybrid models (GMM), the training detection model in a manner of semi-supervised.By training pattern, abnormal behaviour that can effectively during real-time detection dredging operation.
Description
Technical field
The present invention relates to harbour or the unusual checkings of river dredging operation ship, and in particular, to one kind is based on AIS
The extremely semi-supervised real-time detection method of the shipping work of track data, wherein AIS refers to Automatic Identification
System, i.e. ship automatic identification system.
Background technique
Harbour and river require periodically to carry out maintenance dredging, to access to the ports and navigation channel.In dredging operation, ship
Oceangoing ship digs out mud from operating area, and is thrown in specified dumping ground.But in actual dredging operation, one
Mud may be thrown the place near operating area by a little dredging vessels, rather than specifically topple over region.Meanwhile ship
The ship navigated by water on navigation channel is likely encountered during dredging operation, it is therefore necessary to leave navigation channel and be avoided.Therefore, when thin
When dredging ship and sailing out of operating area, it is necessary to distinguish abnormal mud illegal dumping behavior and normal evacuation behavior, i.e. dredger ship
Oceangoing ship operation abnormality detection.Accurate dredging vessel abnormality detection has not only standardized behavior of the dredger in operation process, but also
It ensure that the quality of entire engineering.
For the abnormality detection of inland waterway navigation, patent document CN106816038A discloses a kind of inland waters abnormal behaviour
Ship automatic identification system and method, by the AIS message information, the hydrometeorology department that obtain inland waters ship navigation in real time
Environmental information, CCTV video image and depth image, analyze ship abnormal behaviour pattern class, establish ship abnormal behaviour sample
This library；The analysis ship behavior of deep learning network model is established, ship abnormal behaviour mode and GPS positioning information are obtained；?
Ship is detected in CCTV video image, the three-dimensional spatial information of ship is obtained in conjunction with depth image, obtains ship video location
Information: by GPS positioning information, video location information, ship abnormal behaviour mode, ship detecting Fusion Features, ship mesh is carried out
Mark association, automatically identifies abnormal behaviour ship on CCTV video.
It is then less for the patent document of dredging vessel operation abnormality detection, in general, whether can be used according to the label of sample,
Existing dredging vessel operation method for detecting abnormality can be divided into the method for three types, i.e. unsupervised learning, supervised learning and
Semi-supervised learning.
First method is unsupervised learning, uses unlabelled sample.The basic thought of these methods is in some degree
Exception is found in amount.Exception is by threshold value or according to the ratio-dependent of the measurement.The more existing generation for unsupervised anomaly detection
Table method.The major advantage that label this characteristic is unsupervised learning method is not needed.But abnormal data is needed divided by threshold
Value or ratio, cause this method not flexible.
Second method is supervised learning, and abnormality detection is considered as classification problem, and target is that sort data into be normal
Or it is abnormal.Therefore, train classification models can be carried out using traditional machine learning and deep learning classification method.It proposes
Classical supervised learning classification method.With the offer of label, compared with unsupervised learning method, supervised learning method is realized
Better performance.
Since the cost that label obtains in practical problem is usually very high, before the third method-semi-supervised learning is
Compromise between two methods.It needs a small amount of flag data, but compared with unsupervised learning, obtains in terms of learning accuracy
Sizable progress.In general, semi-supervised learning algorithm can be divided into: generation method, common training method, and half based on figure
Supervised learning method and semisupervised support vector machines (S3VM).Generation method is based on generating model.It assumes that label and it is unmarked
Data be all to be generated by identical potential model, these models are by the parameter of potential model by unlabelled data connection
To learning objective.Common training method assumes that each data can be from different view classifications, wherein different disaggregated models can be with
It is trained by flag data.Based on trained model, it can classify to unlabelled data, then by reliable classification data
It is added in flag data.Data point is considered as vertex by the semi-supervised learning method based on figure, and the similitude between point is regarded
Carry out structural map for side.With mark point there is the unmarked point of larger similitude will be marked as same tag.It is different from SVM,
S3VM attempts to find a subregion hyperplane, the density regions that two class flag datas are separated and pass through data by it.
Existing method cannot be directly used to solve to dredge operation exception test problems online.Method proposed by the present invention by
The inspiration of generation method.Real problems are converted into abnormality detection problem, need to extract by Feature Engineering some substantially special
Sign.It may cause that performance is bad, and the present invention passes through t- before using generation method due to being directly applied to high dimensional data
SNE carries out dimensionality reduction.
Summary of the invention
For the defects in the prior art, the object of the present invention is to provide a kind of shipping works based on AIS track data
Abnormal semi-supervised real-time detection method.
The extremely semi-supervised real-time detection method of a kind of shipping work based on AIS track data provided according to the present invention,
Including pretreatment stage, off-line model training stage and on-line checking stage；
Pretreatment stage: operation track is filtered out using tag system, extracts behavioural characteristic；
The off-line model training stage: according to the behavioural characteristic for the operation track that pretreatment stage obtains, to model parameter into
Row off-line training；
The on-line checking stage: executing online abnormality detection task, issues for abnormal behaviour potential during dredging operation
Alarm.
Preferably, the feature in the pretreatment stage includes global characteristics and/or real-time characteristic.
Preferably, the global characteristics include any one of following eight features or appoint multiple combinations:
1) the total time s of an operation track T；
2) the total travel distance d of an operation track T；
3) ship turns to frequency n in operation areaturn；
4) the total degree n of operating area is driven out in one-stop operationout；
5) total time t is stopped in one-stop operation outside operating areaout；
6) the interior total travel distance d outside operating area of one-stop operationout；
7) the interior longest residence time max outside operating area of one-stop operationt；
8) the maximum distance max behind operating area with boundary is sailed out ofd。
Preferably, the real-time characteristic includes any one of following four feature or appoints multiple combinations:
1) the time t of operating area is leftoutc；
2) the distance d of operating area is leftoutc；
3) last time leaves the time t of operating arealast；
4) present speed v.
Preferably, the off-line model training stage include Nonlinear Dimension Reduction T-SNE model and/or dimensionality reduction after behavior it is special
Levy the probabilistic model of modeling.
Preferably, the T-SNE model is corresponding to calculate by the probability for calculating the similitude of the point in higher dimensional space
The probability of the similitude of point in lower dimensional space, data point xiWith data point xjSimilarity Condition Probability pj|iMeet:
Wherein S (xi,xj) it is xiWith xjBetween similarity metric function, i, j, k be data point label number；In height
In dimension space, S (xi,xj) be defined asIn lower dimensional space, S (xi,xj) it is defined as (1+xi-xj2)-
1。
Preferably, the probabilistic model modeled using gauss hybrid models GMM as behavioural characteristic after dimensionality reduction；
Gauss hybrid models GMM is to maximize log-likelihood for the target of semi-supervised learning, i.e.,
Wherein, log is log operations mathematically, and L (θ) indicates loss function, PθIndicate posterior probability；xrWithIt is band
The sample and its label of label, xuIt is unlabeled exemplars, θ is GMM model parameter, including wi、μiAnd Σi, respectively represent classification i
Ratio, sample average and sample covariance.
Preferably, the parameter θ for maximizing logL (θ) is found using maximum likelihood algorithm EM；Maximum likelihood algorithm EM includes E
Step and M step；
In E step, the posterior probability P of unmarked sample is calculatedθ(Ci|xu), it may be assumed that
Wherein, CiIndicate that classification i, c indicate classification sum, xuIt is unlabeled exemplars, θ is GMM model parameter, including wi、μi
And Σi, respectively represent the ratio, sample average and sample covariance of classification i；
In M step, model parameter θ is updated according to its definition, it may be assumed that
Wherein, N is total sample number, NiIt is classification CiSample number, xrIndicate a sample, xuIt is unlabeled exemplars.wi、μi
And Σi, respectively represent the ratio, sample average and sample covariance of classification i；
E step and M step are repeated, until θ restrains.
The extremely semi-supervised real-time detection method of a kind of shipping work based on AIS track data provided according to the present invention,
Including pretreatment stage, off-line model training stage and on-line checking stage；
Pretreatment stage: it in pretreatment stage, according to the position of operating area and mud dumping ground, will be received first from AIS
Multiple operation tracks are marked off in the initial trace of collection: for each operation track, being extracted by tag system and characterized operation
Point outside region: after the pre-treatment, the training sample for having the tracing point of 12 features as next stage is extracted:
The off-line model training stage: firstly, original sample is mapped to two-dimensional space by t-SNE: meanwhile, training one
The neural network being fully connected comprising three hidden layers, after t-SNE, the data of mapping as input and use maximum seemingly
So algorithm EM training gauss hybrid models GMM: gauss hybrid models GMM is by two Gaussian Profiles corresponding to normal and exception class
Mixing generate:
The online abnormality detection stage: the main target in online abnormality detection stage is that the ongoing dredging of real-time detection is made
The tracing point p of industryiIt is whether normal: the point outside for operating area, first extraction real-time characteristic, then by trained nerve net
Network maps that two-dimensional space: then, mapping point miGauss hybrid models GMM is admitted to obtain P (Canomaly|mi) and P
(Cnormal|mi), respectively represent miBelong to the probability of abnormal and normal class: if P (Canomaly|mi) it is greater than P (Cnormal|mi), pi
Can be identified as that behavior may be abnormal:
In order to assess the intensity of anomaly of one-stop operation, the abnormal score of an operation be may be defined as,
Wherein, W indicates operating area,Indicate the point outside operating area, dist (pi,pi-1) indicate point piAnd pi-1
The distance between: abnormal score by each operating area outside point exceptional value it is cumulative form, which passes through abnormal probability
It is calculated multiplied by the distance to previous point.
A kind of computer readable storage medium for being stored with computer program provided according to the present invention, the computer journey
The step of the extremely semi-supervised real-time detection method of the above-mentioned shipping work based on AIS track data is realized when sequence is executed by processor
Suddenly.
Compared with prior art, the present invention have it is following the utility model has the advantages that
The present invention proposes a kind of operation ship abnormal behavior real-time detection method, and real-time detection is non-in construction operation management
Chang Guanjian, because inspector can go to on-site collection evidence immediately, this is offline inspection when real-time detection is to abnormal alarm
What method can not accomplish.
Detailed description of the invention
Upon reading the detailed description of non-limiting embodiments with reference to the following drawings, other feature of the invention,
Objects and advantages will become more apparent upon:
Fig. 1 is online dredging operation abnormality detection block schematic illustration of the invention；
Fig. 2 is t-SNE and GMM model cooperation relation matters figure of the invention.
Specific embodiment
The present invention is described in detail combined with specific embodiments below.Following embodiment will be helpful to the technology of this field
Personnel further understand the present invention, but the invention is not limited in any way.It should be pointed out that the ordinary skill of this field
For personnel, without departing from the inventive concept of the premise, several changes and improvements can also be made.These belong to the present invention
Protection scope.
The extremely semi-supervised real-time detection method of a kind of shipping work based on AIS track data provided according to the present invention,
Including pretreatment stage, off-line model training stage and on-line checking stage；
Pretreatment stage: operation track is filtered out using tag system, extracts behavioural characteristic；
The off-line model training stage: according to the behavioural characteristic for the operation track that pretreatment stage obtains, to model parameter into
Row off-line training；
The on-line checking stage: executing online abnormality detection task, issues for abnormal behaviour potential during dredging operation
Alarm.
Specifically, the feature in the pretreatment stage includes global characteristics and/or real-time characteristic.The global characteristics packet
It includes any one of following eight features or appoints multiple combinations:
1) the total time s of an operation track T；
2) the total travel distance d of an operation track T；
3) ship turns to frequency n in operation areaturn；
4) the total degree n of operating area is driven out in one-stop operationout；
5) total time t is stopped in one-stop operation outside operating areaout；
6) the interior total travel distance d outside operating area of one-stop operationout；
7) the interior longest residence time max outside operating area of one-stop operationt；
8) the maximum distance max behind operating area with boundary is sailed out ofd。
The real-time characteristic includes any one of following four feature or appoints multiple combinations:
1) the time t of operating area is leftoutc；
2) the distance d of operating area is leftoutc；
3) last time leaves the time t of operating arealast；
4) present speed v.
The off-line model training stage include Nonlinear Dimension Reduction T-SNE model and/or dimensionality reduction after behavioural characteristic model
Probabilistic model.The T-SNE model calculates corresponding low-dimensional by the probability of the similitude of the point in calculating higher dimensional space
The probability of the similitude of point in space, data point xiWith data point xjSimilarity Condition Probability pj|iMeet:
Wherein S (xi,xj) it is xiWith xjBetween similarity metric function, i, j, k be data point label number；In height
In dimension space, S (xi,xj) be defined asIn lower dimensional space, S (xi,xj) it is defined as (1+xi-xj2)-
1。
The probabilistic model modeled using gauss hybrid models GMM as behavioural characteristic after dimensionality reduction；Gauss hybrid models GMM is used
It is to maximize log-likelihood in the target of semi-supervised learning, i.e.,
Wherein, log is log operations mathematically, and L (θ) indicates loss function, PθIndicate posterior probability；xrWithIt is band
The sample and its label of label, xuIt is unlabeled exemplars, θ is GMM model parameter, including wi、μiAnd Σi, respectively represent classification i
Ratio, sample average and sample covariance.
More specifically, finding the parameter θ for maximizing logL (θ) using maximum likelihood algorithm EM；Maximum likelihood algorithm EM packet
Step containing E and M step；
In E step, the posterior probability P of unmarked sample is calculatedθ(Ci|xu), it may be assumed that
Wherein, CiIndicate that classification i, c indicate classification sum, xuIt is unlabeled exemplars, θ is GMM model parameter, including wi、μi
And Σi, respectively represent the ratio, sample average and sample covariance of classification i；
In M step, model parameter θ is updated according to its definition, it may be assumed that
Wherein, N is total sample number, NiIt is classification CiSample number, xrIndicate a sample, xuIt is unlabeled exemplars.wi、μi
And Σi, respectively represent the ratio, sample average and sample covariance of classification i；
E step and M step are repeated, until θ restrains.
The extremely semi-supervised real-time detection method of a kind of shipping work based on AIS track data provided according to the present invention,
Including pretreatment stage, off-line model training stage and on-line checking stage；
Pretreatment stage: it in pretreatment stage, according to the position of operating area and mud dumping ground, will be received first from AIS
Multiple operation tracks are marked off in the initial trace of collection: for each operation track, being extracted by tag system and characterized operation
Point outside region: after the pre-treatment, the training sample for having the tracing point of 12 features as next stage is extracted:
The off-line model training stage: firstly, original sample is mapped to two-dimensional space by t-SNE: meanwhile, training one
The neural network being fully connected comprising three hidden layers, after t-SNE, the data of mapping as input and use maximum seemingly
So algorithm EM training gauss hybrid models GMM: gauss hybrid models GMM is by two Gaussian Profiles corresponding to normal and exception class
Mixing generate:
The online abnormality detection stage: the main target in online abnormality detection stage is that the ongoing dredging of real-time detection is made
The tracing point p of industryiIt is whether normal: the point outside for operating area, first extraction real-time characteristic, then by trained nerve net
Network maps that two-dimensional space: then, mapping point miGauss hybrid models GMM is admitted to obtain P (Canomaly|mi) and P
(Cnormal|mi), respectively represent miBelong to the probability of abnormal and normal class: if P (Canomaly|mi) it is greater than P (Cnormal|mi), pi
Can be identified as that behavior may be abnormal:
In order to assess the intensity of anomaly of one-stop operation, the abnormal score of an operation be may be defined as,
Wherein, W indicates operating area,Indicate the point outside operating area, dist (pi,pi-1) indicate point piAnd pi-1
The distance between: abnormal score by each operating area outside point exceptional value it is cumulative form, which passes through abnormal probability
It is calculated multiplied by the distance to previous point.
A kind of computer readable storage medium for being stored with computer program provided according to the present invention, the computer journey
The step of the extremely semi-supervised real-time detection method of the above-mentioned shipping work based on AIS track data is realized when sequence is executed by processor
Suddenly.
Further, the preferred embodiment of the present invention is described as follows in conjunction with attached drawing:
1. overall architecture
It being inspired by generation method used in semi-supervised learning, the present invention detects abnormal dredging behavior in a probabilistic manner,
The present invention establishes the distribution that a probabilistic model carrys out simulated sediment dredging behavior, and will have the dredging behavior of high abnormal probability as
It is abnormal to return.
It is a kind of up-and-coming method extremely using probabilistic model checking, because the model can be counted as a kind of pressure
The summary of the expression of contracting and history dredging track.Therefore, it for a kind of new dredging behavior, need to only be obtained just by probabilistic model
Often with abnormal probability, track is dredged without searching history, helps to improve detection speed.
Fig. 1 provides shipping work method for detecting abnormality frame, is broadly divided into three phases: pretreatment stage, off-line model
Training stage and on-line checking stage.
Pretreatment stage filters out operation track using tag system, extracts behavioural characteristic.
Key component of the off-line model training stage as system uses two kinds of models: one is for non-linear
The T-SNE model of dimensionality reduction, another kind are the probabilistic model modeled with behavioural characteristic after dimensionality reduction, i.e. gauss hybrid models (GMM).Root
The behavioural characteristic for the operation track that the Data preprocess stage obtains carries out off-line training to the model parameter of two kinds of models.
After model parameter deduction, online abnormality detection task is executed, it is therefore an objective to for potential during dredging operation
Abnormal behaviour sounds an alarm.
2. tag system
Since original AIS information cannot be directly used to detection abnormal behaviour, it is therefore desirable to construct some advanced features and come area
Divide these behaviors.These new features can be divided into global characteristics and real-time characteristic.Intuitively, it extracts to indicate complete dredging operation
Global characteristics are abnormal for offline inspection.When detecting abnormal during operation, global characteristics are unavailable, it is therefore desirable to design
Real-time characteristic is to ensure that detection method still can identify abnormal behaviour in real time.In other words, global characteristics are a dredging rails
The expression of mark, and real-time characteristic depicts a point in dredging track.
(1) global characteristics
Global characteristics focus on the index that can be collected when an operation is completed.For clearer description, from AIS data
In be extracted following 8 global characteristics.
1) the total time s of an operation track T
2) the total travel distance d of an operation track T
3) ship turns to frequency n in operation areaturn
4) the total degree n of operating area is driven out in one-stop operationout
5) total time t is stopped in one-stop operation outside operating areaout
6) the interior total travel distance d outside operating area of one-stop operationout
7) the interior longest residence time max outside operating area of one-stop operationt
8) the maximum distance max behind operating area with boundary is sailed out ofd
(2) real-time characteristic
Real-time characteristic be mainly used for operation during online abnormality detection.Since operation does not complete, can not capture whole
Body information.Nevertheless, the track from starting point to current point can be considered as primary complete operation, for more real-time
Statistical nature.Other real-time characteristics further include:
1) the time t of operating area is leftoutc
2) the distance d of operating area is leftoutc
3) last time leaves the time t of operating arealast
4) present speed v
3.t-SNE Feature Dimension Reduction
It is a kind of visual machine learning algorithm that T distributed random neighborhood, which is embedded in (t-SNE),.It is a kind of non-linear drop
Dimension technology is very suitable to insertion high dimensional data, to be visualized in two dimension or three-dimensional lower dimensional space.T-SNE algorithm
The probability of the similitude of the point in corresponding lower dimensional space is calculated by the probability of the similitude of the point in calculating higher dimensional space.
Data point xiWith data point xjSimilarity Condition Probability pj|i
Wherein S (xi,xj) it is xiWith xjBetween similarity metric function, i, j, k be data point label number.In height
In dimension space, S (xi,xj) be defined asIn lower dimensional space, S (xi,xj) it is defined as (1+xi-xj2)-
1.Then, t-SNE attempts to minimize the difference in higher-dimension and lower dimensional space between these conditional probabilities, so as to compared with lower dimensional space
Middle perfect expression data point.For the minimum of design conditions probability difference summation, t-SNE minimizes institute using gradient descent method
There is the summation of the Kullback-Leibler divergence of data point.
T-SNE study is nonparametric mapping, it means that it will not learn for data to be mapped to from higher dimensional compared with low-dimensional
The explicit function of degree.Therefore, it is impossible to be directly embedded into new point in existing mapping.A kind of potential method is by directly minimizing
T-SNE loss function trains multivariate regression amount.Or training neural network is to adapt to map from initial data to by t-SNE
Relatively low-dimensional data mapping.
The dimension of characteristic is reduced to two dimension from higher dimensional space using t-SNE by the present invention, and by being fully connected
Neural network reserved mapping.In general, using t-SNE, there are two major advantages.On the one hand, t-SNE can help Visual Inference
Characteristic and check whether data usually separable.On the other hand, mapping data are more had based on their similitude
It may be clustered in certain classes, this helps to train GMM.
4.GMM
Gauss hybrid models (GMM) are a probabilistic models, assume that all data points are by having with unknown parameter
Limit the mixing generation of the Gaussian Profile of quantity.For semi-supervised learning, GMM is a kind of generation method.In supervised learning
GMM is different, and GMM is to maximize log-likelihood for the target of semi-supervised learning, i.e.,
Log is log operations mathematically, and L (θ) indicates loss function, PθIndicate posterior probability.
Wherein, xrWithIt is the sample and its label of tape label, xuIt is unlabeled exemplars, θ is GMM model parameter, including
wi、μiAnd Σi, respectively represent the ratio, sample average and sample covariance of classification i.
Maximum likelihood (EM) algorithm can be used for finding the parameter θ for maximizing logL (θ).With it is initial used in supervised learning
Change difference, EM algorithm initializes the parameter of each Gauss model using the sample of label.EM algorithm includes E step and M step.
In E step, the posterior probability P of unmarked sample is calculatedθ(Ci|xu), that is,
Wherein, CiIndicate that classification i, c indicate classification sum, xuIt is unlabeled exemplars, θ is GMM model parameter, including wi、μi
And Σi, respectively represent the ratio, sample average and sample covariance of classification i.
In M step, model parameter θ is updated according to its definition, that is,
Wherein, N is total sample number, NiIt is classification CiSample number, xrIndicate a sample, xuIt is unlabeled exemplars.wi、μi
And Σi, respectively represent the ratio, sample average and sample covariance of classification i.
E step and M step repeat, until θ restrains.
When θ is fixed, GMM model is determining.Therefore, for a new samples x, it is easy to calculate x and belong to the general of i class
Rate Pθ(Ci|x).In other words, when classification is defined as normal or abnormal, it is normal that we can be readily available a sample
Or abnormal probability.
GMM is very suitable to the abnormality detection in semi-supervised situation.Compared with unsupervised anomaly detection method, GMM being capable of benefit
With hard-won label data, to obtain better performance.Compared with most of semi-supervised methods, GMM is able to carry out soft
Distribution, rather than when being classified directly by sample labeling be some class.
5. dredging operation abnormality detection process
(1) it pre-processes
In pretreatment stage, according to the position of operating area and mud dumping ground, initial trace that will be collected first from AIS
In mark off multiple operation tracks.For each operation track, is extracted by tag system and characterize the point outside operating area.?
After pretreatment, the training sample for having the tracing point of 12 features as next stage is extracted.
(2) off-line model training
Fig. 2 shows the cooperation relation between t-SNE and GMM model.Firstly, original sample is mapped to two by t-SNE
Dimension space.Meanwhile training one include three hidden layers the neural network being fully connected, every layer of element number is 5 respectively,
5 and 10, to keep being used for the mapping relations of new embedding data.After t-SNE, the data of mapping are calculated as input and using EM
Method trains GMM.GMM is generated by the mixing for corresponding to two Gaussian Profiles of normal and exception class.
(3) online abnormality detection
The main target in online abnormality detection stage is the tracing point p of the ongoing dredging operation of real-time detectioniWhether
Normally.Point outside for operating area, first extraction real-time characteristic, then map that two dimension by trained neural network
Space.Then, mapping point miGMM is admitted to obtain P (Canomaly|mi) and P (Cnormal|mi), respectively represent miBelong to it is abnormal and
The probability of normal class.If P (Canomaly|mi) it is greater than P (Cnormal|mi), piCan be identified as that behavior may be abnormal.
In order to assess the intensity of anomaly of one-stop operation, the abnormal score of an operation be may be defined as,
Wherein, W indicates operating area,Indicate the point outside operating area, dist (pi,pi-1) indicate point piAnd pi-1
The distance between.Abnormal score by each operating area outside point exceptional value it is cumulative form, which passes through abnormal probability
It is calculated multiplied by the distance to previous point.
Further, technology of the invention solve the problems, such as be:
To the ship for implementing dredging operation in specified port area, when it sails out of operating area, real-time detection its be normal
Evacuation behavior or abnormal mud topple over behavior, to improve the quality of engineering management and the dynamics of environmental protection.
Technical solution of the invention are as follows:
The present invention establishes a tag system and is used to extract behavioural characteristic from AIS data.In addition, also by T- be distributed with
Machine neighborhood insertion (T-SNE) is combined by neural network with gauss hybrid models (GMM), the training detection mould in a manner of semi-supervised
Type.
In the description of the present application, it is to be understood that term " on ", "front", "rear", "left", "right", " is erected at "lower"
Directly ", the orientation or positional relationship of the instructions such as "horizontal", "top", "bottom", "inner", "outside" is orientation based on the figure or position
Relationship is set, description the application is merely for convenience of and simplifies description, rather than the device or element of indication or suggestion meaning are necessary
It with specific orientation, is constructed and operated in a specific orientation, therefore should not be understood as the limitation to the application.
Specific embodiments of the present invention are described above.It is to be appreciated that the invention is not limited to above-mentioned
Particular implementation, those skilled in the art can make a variety of changes or modify within the scope of the claims, this not shadow
Ring substantive content of the invention.In the absence of conflict, the feature in embodiments herein and embodiment can any phase
Mutually combination.
Claims (10)
1. a kind of extremely semi-supervised real-time detection method of the shipping work based on AIS track data, which is characterized in that including locating in advance
Reason stage, off-line model training stage and on-line checking stage；
Pretreatment stage: operation track is filtered out using tag system, extracts behavioural characteristic；
The off-line model training stage: according to the behavioural characteristic for the operation track that pretreatment stage obtains, to model parameter carry out from
Line training；
The on-line checking stage: executing online abnormality detection task, sounds an alarm for abnormal behaviour potential during dredging operation.
2. the extremely semi-supervised real-time detection method of the shipping work based on AIS track data according to claim 1, special
Sign is that the feature in the pretreatment stage includes global characteristics and/or real-time characteristic.
3. the extremely semi-supervised real-time detection method of the shipping work based on AIS track data according to claim 2, special
Sign is that the global characteristics include any one of following eight features or appoint multiple combinations:
1) the total time s of an operation track T；
2) the total travel distance d of an operation track T；
3) ship turns to frequency n in operation areaturn；
4) the total degree n of operating area is driven out in one-stop operationout；
5) total time t is stopped in one-stop operation outside operating areaout；
6) the interior total travel distance d outside operating area of one-stop operationout；
7) the interior longest residence time max outside operating area of one-stop operationt；
8) the maximum distance max behind operating area with boundary is sailed out ofd。
4. the extremely semi-supervised real-time detection method of the shipping work based on AIS track data according to claim 2, special
Sign is that the real-time characteristic includes any one of following four feature or appoints multiple combinations:
1) the time t of operating area is leftoutc；
2) the distance d of operating area is leftoutc；
3) last time leaves the time t of operating arealast；
4) present speed v.
5. the extremely semi-supervised real-time detection method of the shipping work based on AIS track data according to claim 1, special
Sign is, the off-line model training stage include Nonlinear Dimension Reduction T-SNE model and/or dimensionality reduction after behavioural characteristic model
Probabilistic model.
6. the extremely semi-supervised real-time detection method of the shipping work based on AIS track data according to claim 5, special
Sign is that the T-SNE model calculates corresponding lower dimensional space by the probability of the similitude of the point in calculating higher dimensional space
In point similitude probability, data point xiWith data point xjSimilarity Condition Probability pj|iMeet:
Wherein S (xi,xj) it is xiWith xjBetween similarity metric function, i, j, k be data point label number；In higher dimensional space
In, S (xi,xj) be defined asIn lower dimensional space, S (xi,xj) it is defined as (1+xi-xj2) -1.
7. the extremely semi-supervised real-time detection method of the shipping work based on AIS track data according to claim 6, special
Sign is, the probabilistic model modeled using gauss hybrid models GMM as behavioural characteristic after dimensionality reduction；
Gauss hybrid models GMM is to maximize log-likelihood for the target of semi-supervised learning, i.e.,
Wherein, log is log operations mathematically, and L (θ) indicates loss function, PθIndicate posterior probability；xrWithIt is tape label
Sample and its label, xuIt is unlabeled exemplars, θ is GMM model parameter, including wi、μiAnd Σi, respectively represent the ratio of classification i
Example, sample average and sample covariance.
8. the extremely semi-supervised real-time detection method of the shipping work based on AIS track data according to claim 7, special
Sign is, the parameter θ for maximizing logL (θ) is found using maximum likelihood algorithm EM；Maximum likelihood algorithm EM includes E step and M
Step；
In E step, the posterior probability P of unmarked sample is calculatedθ(Ci|xu), it may be assumed that
Wherein, CiIndicate that classification i, c indicate classification sum, xuIt is unlabeled exemplars, θ is GMM model parameter, including wi、μiWith
Σi, respectively represent the ratio, sample average and sample covariance of classification i；
In M step, model parameter θ is updated according to its definition, it may be assumed that
Wherein, N is total sample number, NiIt is classification CiSample number, xrIndicate a sample, xuIt is unlabeled exemplars.wi、μiWith
Σi, respectively represent the ratio, sample average and sample covariance of classification i；
E step and M step are repeated, until θ restrains.
9. a kind of extremely semi-supervised real-time detection method of the shipping work based on AIS track data, which is characterized in that including locating in advance
Reason stage, off-line model training stage and on-line checking stage；
Pretreatment stage: it in pretreatment stage, according to the position of operating area and mud dumping ground, will be collected first from AIS
Multiple operation tracks are marked off in initial trace: for each operation track, being extracted by tag system and characterized operating area
Outer point: after the pre-treatment, the training sample for having the tracing point of 12 features as next stage is extracted:
The off-line model training stage: firstly, original sample is mapped to two-dimensional space by t-SNE: meanwhile, training one includes
The neural network of three hidden layers being fully connected, after t-SNE, the data of mapping are calculated as input and using maximum likelihood
Method EM training gauss hybrid models GMM: gauss hybrid models GMM is by the mixed of two Gaussian Profiles corresponding to normal and exception class
Symphysis at:
The online abnormality detection stage: the main target in online abnormality detection stage is the ongoing dredging operation of real-time detection
Tracing point piIt is whether normal: the point outside for operating area, first extraction real-time characteristic, it then will by trained neural network
It is mapped to two-dimensional space: then, mapping point miGauss hybrid models GMM is admitted to obtain P (Canomaly|mi) and P (Cnormal
|mi), respectively represent miBelong to the probability of abnormal and normal class: if P (Canomaly|mi) it is greater than P (Cnormal|mi), piIt can be by
It is identified as that behavior may be abnormal:
In order to assess the intensity of anomaly of one-stop operation, the abnormal score of an operation be may be defined as,
Wherein, W indicates operating area,Indicate the point outside operating area, dist (pi,pi-1) indicate point piAnd pi-1Between
Distance: abnormal score by each operating area outside point exceptional value it is cumulative form, the exceptional value pass through abnormal probability multiplied by
It is calculated to the distance of previous point.
10. a kind of computer readable storage medium for being stored with computer program, which is characterized in that the computer program is located
Reason device realizes the shipping work extremely semi-supervised reality based on AIS track data described in any one of claims 1 to 9 when executing
When detection method the step of.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910574738.1A CN110363115B (en) | 2019-06-28 | 2019-06-28 | AIS (automatic identification system) track data based ship operation abnormity semi-supervised real-time detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910574738.1A CN110363115B (en) | 2019-06-28 | 2019-06-28 | AIS (automatic identification system) track data based ship operation abnormity semi-supervised real-time detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110363115A true CN110363115A (en) | 2019-10-22 |
| CN110363115B CN110363115B (en) | 2021-10-15 |
Family
ID=68217566
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910574738.1A Active CN110363115B (en) | 2019-06-28 | 2019-06-28 | AIS (automatic identification system) track data based ship operation abnormity semi-supervised real-time detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110363115B (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110824142A (en) * | 2019-11-13 | 2020-02-21 | 杭州鲁尔物联科技有限公司 | Geological disaster prediction method, device and equipment |
| CN110929790A (en) * | 2019-11-26 | 2020-03-27 | 山东中创软件工程股份有限公司 | Ship data screening method and related device |
| CN111339864A (en) * | 2020-02-17 | 2020-06-26 | 北京百度网讯科技有限公司 | Abnormal behavior alarm method and device |
| CN111682972A (en) * | 2020-08-14 | 2020-09-18 | 支付宝(杭州)信息技术有限公司 | Method and device for updating service prediction model |
| CN111785090A (en) * | 2020-06-24 | 2020-10-16 | 国家海洋环境监测中心 | Method for automatically identifying illegal dumping based on ship AIS track data |
| CN111931555A (en) * | 2020-06-14 | 2020-11-13 | 交通运输部科学研究院 | Method for identifying whether ship AIS is started or not by utilizing video image |
| CN112395382A (en) * | 2020-11-23 | 2021-02-23 | 武汉理工大学 | Ship abnormal track data detection method and device based on variational self-encoder |
| CN112699315A (en) * | 2020-12-30 | 2021-04-23 | 中南大学 | AIS data-based ship abnormal behavior detection method |
| CN113326472A (en) * | 2021-05-28 | 2021-08-31 | 东北师范大学 | Pattern extraction and evolution visual analysis method based on time sequence multivariable data |
| CN113542697A (en) * | 2021-09-15 | 2021-10-22 | 江西省水利科学院 | Adaptive monitoring method of sand production ship adaptive monitoring system based on deep learning |
| CN113935666A (en) * | 2021-12-17 | 2022-01-14 | 武汉精装房装饰材料有限公司 | Building decoration wall tile abnormity evaluation method based on image processing |
| CN114118243A (en) * | 2021-11-18 | 2022-03-01 | 中交疏浚技术装备国家工程研究中心有限公司 | Track data-based construction track identification method for trailing suction hopper dredger |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106218831A (en) * | 2016-07-22 | 2016-12-14 | 大连海事大学 | A kind of method and system obtaining Ship Controling behavior based on watercraft AIS track data |
| CN108228732A (en) * | 2016-12-14 | 2018-06-29 | 公立大学法人首都大学东京 | language storage method and language dialogue system |
-
2019
- 2019-06-28 CN CN201910574738.1A patent/CN110363115B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106218831A (en) * | 2016-07-22 | 2016-12-14 | 大连海事大学 | A kind of method and system obtaining Ship Controling behavior based on watercraft AIS track data |
| CN108228732A (en) * | 2016-12-14 | 2018-06-29 | 公立大学法人首都大学东京 | language storage method and language dialogue system |
Non-Patent Citations (3)
| Title |
|---|
| A. P. DEMPSTER ET AL.: "Maximum likelihood from incomplete data via the em algorithm", 《JOURNAL OF THE ROYAL STATISTICAL SOCIETY》 * |
| 叶敏: "基于轨迹数据挖掘的异常检测方法研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 邓磊: "基于数据挖掘的船舶航行轨迹异常检测方法研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110824142B (en) * | 2019-11-13 | 2022-06-24 | 杭州鲁尔物联科技有限公司 | Geological disaster prediction method, device and equipment |
| CN110824142A (en) * | 2019-11-13 | 2020-02-21 | 杭州鲁尔物联科技有限公司 | Geological disaster prediction method, device and equipment |
| CN110929790A (en) * | 2019-11-26 | 2020-03-27 | 山东中创软件工程股份有限公司 | Ship data screening method and related device |
| CN111339864A (en) * | 2020-02-17 | 2020-06-26 | 北京百度网讯科技有限公司 | Abnormal behavior alarm method and device |
| CN111339864B (en) * | 2020-02-17 | 2023-11-10 | 北京百度网讯科技有限公司 | Abnormal behavior alarm method and device |
| CN111931555A (en) * | 2020-06-14 | 2020-11-13 | 交通运输部科学研究院 | Method for identifying whether ship AIS is started or not by utilizing video image |
| CN111931555B (en) * | 2020-06-14 | 2023-08-08 | 交通运输部科学研究院 | Method for identifying whether ship AIS is opened or not by utilizing video image |
| CN111785090A (en) * | 2020-06-24 | 2020-10-16 | 国家海洋环境监测中心 | Method for automatically identifying illegal dumping based on ship AIS track data |
| CN111682972B (en) * | 2020-08-14 | 2020-11-03 | 支付宝(杭州)信息技术有限公司 | Method and device for updating service prediction model |
| CN111682972A (en) * | 2020-08-14 | 2020-09-18 | 支付宝(杭州)信息技术有限公司 | Method and device for updating service prediction model |
| CN112395382A (en) * | 2020-11-23 | 2021-02-23 | 武汉理工大学 | Ship abnormal track data detection method and device based on variational self-encoder |
| CN112699315A (en) * | 2020-12-30 | 2021-04-23 | 中南大学 | AIS data-based ship abnormal behavior detection method |
| CN112699315B (en) * | 2020-12-30 | 2023-08-15 | 中南大学 | AIS data-based ship abnormal behavior detection method |
| CN113326472A (en) * | 2021-05-28 | 2021-08-31 | 东北师范大学 | Pattern extraction and evolution visual analysis method based on time sequence multivariable data |
| CN113326472B (en) * | 2021-05-28 | 2022-07-15 | 东北师范大学 | Pattern extraction and evolution visual analysis method based on time sequence multivariable data |
| CN113542697A (en) * | 2021-09-15 | 2021-10-22 | 江西省水利科学院 | Adaptive monitoring method of sand production ship adaptive monitoring system based on deep learning |
| CN113542697B (en) * | 2021-09-15 | 2021-12-10 | 江西省水利科学院 | Adaptive monitoring method of sand production ship adaptive monitoring system based on deep learning |
| CN114118243A (en) * | 2021-11-18 | 2022-03-01 | 中交疏浚技术装备国家工程研究中心有限公司 | Track data-based construction track identification method for trailing suction hopper dredger |
| CN113935666A (en) * | 2021-12-17 | 2022-01-14 | 武汉精装房装饰材料有限公司 | Building decoration wall tile abnormity evaluation method based on image processing |
| CN113935666B (en) * | 2021-12-17 | 2022-03-22 | 武汉精装房装饰材料有限公司 | Building decoration wall tile abnormity evaluation method based on image processing |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110363115B (en) | 2021-10-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110363115A (en) | The extremely semi-supervised real-time detection method of shipping work based on AIS track data | |
| Zhang et al. | Towards a Convolutional Neural Network model for classifying regional ship collision risk levels for waterway risk analysis | |
| Murray et al. | An AIS-based deep learning framework for regional ship behavior prediction | |
| Pan et al. | Visual recognition based on deep learning for navigation mark classification | |
| CN106816038B (en) | A kind of inland waters abnormal behaviour ship automatic identification system and method | |
| Yetilmezsoy et al. | Artificial intelligence-based prediction models for environmental engineering | |
| Laxhammar et al. | Inductive conformal anomaly detection for sequential detection of anomalous sub-trajectories | |
| CN111368690B (en) | Deep learning-based video image ship detection method and system under influence of sea waves | |
| CN108596030A (en) | Sonar target detection method based on Faster R-CNN | |
| Dowden et al. | Sea ice classification via deep neural network semantic segmentation | |
| CN109977897A (en) | A kind of ship's particulars based on deep learning recognition methods, application method and system again | |
| Anneken et al. | Evaluation and comparison of anomaly detection algorithms in annotated datasets from the maritime domain | |
| CN105574550A (en) | Vehicle identification method and device | |
| Yan et al. | Vessel movement analysis and pattern discovery using density-based clustering approach | |
| Capizzi et al. | A clustering based system for automated oil spill detection by satellite remote sensing | |
| CN110097044A (en) | Stage car plate detection recognition methods based on deep learning | |
| CN110598693A (en) | Ship plate identification method based on fast-RCNN | |
| Rong et al. | Evaluation of near-collisions in the Tagus River Estuary using a marine traffic simulation model | |
| Bækkegaard et al. | Target classification using kinematic data and a recurrent neural network | |
| Toloue et al. | Anomalous behavior detection of marine vessels based on Hidden Markov Model | |
| Yildirim et al. | Ship detection in optical remote sensing images using YOLOv4 and Tiny YOLOv4 | |
| Eljabu et al. | Spatial clustering method of historical ais data for maritime traffic routes extraction | |
| Chen et al. | The unsupervised method of vessel movement trajectory prediction | |
| Xia | Navigational risk analysis based on GIS spatiotemporal trajectory mining: a case study in Nanjing Yangtze River Bridge waters | |
| Wu et al. | Multiple detection model fusion framework for printed circuit board defect detection |
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 |