CN111694913B - Ship AIS track clustering method and device based on convolution self-encoder - Google Patents
Ship AIS track clustering method and device based on convolution self-encoder Download PDFInfo
- Publication number
- CN111694913B CN111694913B CN202010507856.3A CN202010507856A CN111694913B CN 111694913 B CN111694913 B CN 111694913B CN 202010507856 A CN202010507856 A CN 202010507856A CN 111694913 B CN111694913 B CN 111694913B
- Authority
- CN
- China
- Prior art keywords
- track
- ship
- sub
- feature
- encoder
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/29—Geographical information databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/284—Relational databases
- G06F16/285—Clustering or classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Remote Sensing (AREA)
- Navigation (AREA)
Abstract
The invention relates to a ship AIS track clustering method and device based on a convolution self-encoder. The ship AIS track clustering method based on the convolution self-encoder comprises the following steps: acquiring a continuous track of a ship, and dividing the continuous track into a plurality of sub-tracks; extracting characteristic engineering of a plurality of sub-tracks to obtain a sub-track characteristic matrix; inputting the sub-track feature matrix into a multi-feature fusion self-encoder to obtain a position feature vector, a speed feature vector and a course feature vector; performing splicing operation on the position feature vector, the speed feature vector and the heading feature vector to obtain potential feature vectors of the ship track; and performing track clustering operation on the extracted ship track feature vectors to obtain a ship track clustering result. The method of the invention does not need to select a space-time track measurement method according to the related data quantity and track type, the calculation complexity, the noise and other influencing factors and does not need a similarity distance formula, thereby saving calculation time and resources.
Description
Technical Field
The invention relates to the technical field of software engineering, in particular to a ship AIS track clustering method and device based on a convolution self-encoder.
Background
Satellite AIS (Automatic Identification System, automatic ship identification system) is a ship positioning technology, which receives AIS message information sent by a ship through a low-orbit satellite, and the satellite forwards the received and decoded AIS message information to a corresponding earth station, so that a land management mechanism can master the relevant dynamic information of the ship, and the monitoring of the navigation ship in the open sea area is realized.
In order to improve the shipping capacity and efficiency of the ship, the prior art clusters ship navigation track information data acquired from the AIS, so that a navigation scheme can be predicted according to a clustering result.
The trajectory clustering algorithm is a key and fundamental to solve the practical problem, and is widely used in numerous real-world applications, such as: anomaly detection, moving object behavior prediction, activity understanding, 3D structure reconstruction, traffic monitoring, and the like. Many researchers have now proposed a number of trajectory clustering methods that use some measure of trajectory similarity to quantify the trajectory similarity, and then apply some classical clustering algorithms, such as K-means, gaussian mixture models, and density-based application spatial clustering, to cluster trajectories.
The conventional track clustering method generally needs to select a space-time track measurement method according to the related data quantity and track type, computational complexity, noise and other influencing factors, and the main problem is that the selection of the optimal similarity measurement formula needs a great deal of priori knowledge and extensive experiments, so that a great deal of computational resources and time are wasted.
Disclosure of Invention
Based on the above, the invention aims to provide a ship AIS track clustering method and device based on a convolution self-encoder, so that a space-time track similarity measurement formula does not need to be manually selected, and therefore similarity calculation deviation based on track distance caused by a traditional similarity measurement method can be avoided.
In a first aspect, an embodiment of the present application provides a ship AIS track clustering method based on a convolution self-encoder, including the following steps:
acquiring a continuous track of a ship, and dividing the continuous track into a plurality of sub-tracks;
carrying out feature engineering extraction on a plurality of sub-tracks to obtain a sub-track feature matrix, wherein the ith row in the matrix represents the feature value of the ith sub-track;
inputting the sub-track feature matrix into a multi-feature fusion self-encoder, and carrying out feature extraction on the ship feature vectors of the position, the speed and the course through the multi-feature fusion self-encoder to obtain a position feature vector, a speed feature vector and a course feature vector;
performing splicing operation on the position feature vector, the speed feature vector and the heading feature vector to obtain a potential feature vector of a ship track;
and performing track clustering operation on the extracted ship track feature vectors to obtain a ship track clustering result.
Optionally, dividing the continuous track into a plurality of sub-tracks includes:
when the longitude difference or the change of the latitude difference of a certain track point relative to a previous track point in the continuous track is smaller than a set threshold value and the speed of the track point is close to 0, determining the track point as a starting point of a sub track;
and determining the track point with the speed close to 0 as the end point of the sub-track, wherein the longitude difference or the change of the latitude difference of the next track point relative to the previous track point in the continuous track is smaller than a set threshold value.
Optionally, performing feature engineering extraction on the plurality of sub-tracks to obtain a sub-track feature matrix, and further including:
and carrying out normalization operation on the features in the sub-track feature matrix according to the following formula:
wherein, delta represents the feature to be normalized, min and max represent the maximum value and the minimum value of the feature, respectively, and delta' represents the feature obtained after normalization processing.
Optionally, after normalizing the features in the sub-track feature matrix, the method further includes:
and taking the number of the longest sub-track points as a standard, and filling 0 into other track data so that the lengths of the ship track feature vectors are the same.
Optionally, the multi-feature fusion self-encoder includes:
the first convolution automatic encoder is used for learning the position type characteristics of the ship track;
a second convolution automatic encoder for learning a speed type characteristic of the ship track;
and the third convolution automatic encoder is used for learning the heading type characteristics of the ship track.
Optionally, the location type feature includes longitude and latitude;
the speed type features include speed and acceleration;
the heading-type features include heading and turning rate.
Optionally, for each convolution automatic encoder, the structure includes:
the input layer is used for acquiring a characteristic matrix of each ship track characteristic data, expanding sub-track data into sub-track vectors and transmitting the sub-track vectors to the encoder part;
the encoder consists of a convolution layer and a pooling layer, an activation function is set as a relu function and is used for carrying out convolution operation on the sub-track vector and outputting extracted ship track characteristics;
a decoder for implementing a deconvolution function using a combination of a convolution layer and an up-sampling layer, adjusting a low-dimensional feature vector into a high-dimensional feature vector, and transmitting the high-dimensional feature vector to an output layer;
and the output layer is used for converting the high-dimensional feature vector into the original track feature vector so as to be the same as the latitude of the input data of the input layer.
Optionally, the mean square error is used to measure the similarity between the output layer data and the input layer data, as follows:
wherein MSE represents mean square error, observed M Representing real track data, predicted M Representing predicted trajectory data, N representing the total number of sub-trajectories for all vessels.
Optionally, the method further comprises:
and (3) evaluating the clustering effect by adopting an f1 value method, wherein the formula is as follows:
where TP represents a track correctly predicted as a positive example, FP represents a track incorrectly predicted as a positive example, and FN represents a track incorrectly predicted as a negative example.
In a second aspect, an embodiment of the present application provides a ship AIS track clustering device based on a convolution self-encoder, the device includes:
the track acquisition module is used for acquiring a continuous track of the ship and dividing the continuous track into a plurality of sub-tracks;
the characteristic engineering module is used for carrying out characteristic engineering extraction on a plurality of sub-tracks to obtain a sub-track characteristic matrix, wherein the ith row in the matrix represents the characteristic value of the ith sub-track;
the characteristic extraction module is used for inputting the sub-track characteristic matrix into a multi-characteristic fusion self-encoder, and carrying out characteristic extraction on the ship characteristic vectors of the position, the speed and the course through the multi-characteristic fusion self-encoder to obtain a position characteristic vector, a speed characteristic vector and a course characteristic vector;
the splicing module is used for carrying out splicing operation on the position feature vector, the speed feature vector and the heading feature vector to obtain potential feature vectors of the ship track;
and the clustering module is used for carrying out track clustering operation on the extracted ship track feature vectors to obtain a ship track clustering result.
In the embodiment of the application, performing track segmentation operation on an original ship AIS track to obtain segmented ship AIS sub-tracks, and performing feature engineering operation on each ship sub-track to obtain an AIS ship sub-track feature matrix, wherein an ith row in the matrix represents a feature value of the ith sub-track; the multi-feature fusion self-encoder is used for extracting the features of the ship feature vectors of the position, the speed and the course to obtain three types of feature vectors, then the three types of feature vectors are spliced to obtain potential feature vectors of the ship track, and the main stream clustering algorithm is used for carrying out track clustering operation on the extracted ship track feature vectors, so that a space-time track similarity measurement formula can be realized without manually selecting, similarity calculation deviation based on track distance caused by a traditional similarity measurement method can be avoided, the distribution of data of different ship feature types can be reserved, and the calculation performance is improved.
For a better understanding and implementation, the present invention is described in detail below with reference to the drawings.
Drawings
FIG. 1 is a flowchart of a ship AIS trajectory clustering method based on a convolutional self-encoder in one embodiment of the present application;
FIG. 2 is a flowchart illustrating steps for dividing a continuous track into sub-tracks according to one embodiment of the present application;
FIG. 3 is a schematic flow chart of a ship AIS track clustering method based on a convolution self-encoder in an embodiment of the present application;
FIG. 4 is a schematic diagram of a model structure applied by a ship AIS track clustering method based on a convolution self-encoder in an embodiment of the present application;
fig. 5 is a schematic structural diagram of a ship AIS track clustering device based on a convolution self-encoder according to an embodiment of the present application.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the present specification. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present description as detailed in the accompanying claims.
The terminology used in the description presented herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the description. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any or all possible combinations of one or more of the associated listed items.
Next, embodiments of the present specification will be described in detail.
Aiming at the technical problem that a space-time track measurement method is generally required to be selected according to the related data quantity and track type, the calculation complexity, the noise and other influencing factors in the prior art, and a great deal of calculation resources and time are wasted, the embodiment of the application provides a ship AIS track clustering method based on a convolution self-encoder, as shown in fig. 1, in one embodiment, the method comprises the following steps:
s101: and obtaining a continuous track of the ship, and dividing the continuous track into a plurality of sub-tracks.
In the embodiment of the present application, it is first necessary to divide one continuous track TR of a ship into several sub-tracks: i.e., tr= { TR1, TR2,.. trj }, where trj is the ship sub-track and j represents the total number of sub-tracks of one ship.
The embodiment of the application divides the whole ship track TR into a series of ship sub-tracks, so that each ship sub-track can accurately represent a feasible single sub-track from one port to another port, and the longitude difference or the latitude difference between two continuous points of the track needs to be calculated to identify the starting point and the end point of the ship sub-track. Theoretically and empirically, the speed of a ship is close to 0 when it approaches the destination, and the latitude and longitude change is small.
In a specific example, as shown in fig. 2, the continuous track is divided into a plurality of sub-tracks, including the following steps:
s201: when the longitude difference or the change of the latitude difference of a certain track point relative to a previous track point in the continuous track is smaller than a set threshold value and the speed of the track point is close to 0, determining the track point as a starting point of a sub track;
s202: and determining the track point with the speed close to 0 as the end point of the sub-track, wherein the longitude difference or the change of the latitude difference of the next track point relative to the previous track point in the continuous track is smaller than a set threshold value.
S102: and carrying out feature engineering extraction on a plurality of sub-tracks to obtain a sub-track feature matrix, wherein the ith row in the matrix represents the feature value of the ith sub-track.
In the embodiments of the present application, several concepts are first defined: given a data set a= { a 1, a 2, a 3, …, a n }, where n is the number of vessels and ai is the time series data points of sample points of one vessel trajectory. That is, ai= (p 1, p 2, p 3,., pm), where m is the number of sub-track points, pm is the point defined as the tuple (t, lat, lon, speed, cog, rot, ac), where t is the timestamp, lon is the longitude of the ship track point, lat is the latitude of the ship track point, cog is the heading of the ship track point, speed is the instantaneous speed of the ship track point, rot is the ship turning rate, ac is the acceleration of the ship at time t.
Assuming that in some cases only the time, longitude and latitude of the locus point can be obtained, the characteristic extraction result can still be calculated from the raw data using the ship locus characteristic engineering formula as follows:
△lat ap ＝lat ap -lat ap-1
△lon ap ＝lon ap -lon ap-1
△t ap ＝time ap -time ap-1
h＝hav(△lat ap )+cos(lat ap-1 )*cos(lat ap )*hav(△lon ap )
wherein Deltalat ap ，△lon ap ，△t ap The difference in elevation, the difference in longitude and the difference in time of the p-th point of the track a are represented, respectively. rot is the turn rate of the p-th point of the track a, and Courseap is a piecewise function used for calculating the heading value of the p-th point of the track a. h and haverin (θ) use haverin's formula to calculate the distance between two points between any two tracks on the earth, R is the earth radius, and the average value is 6371km. The speed is the instantaneous speed of the track point calculated by dividing the distance between two points of the track by the time difference. acap denotes the acceleration ac calculated from the speed difference and the time difference of the adjacent two points of the trajectory.
In one embodiment, in order to scale the features of different tracks to values between (0, 1), so that the data extraction process has a more rapid and accurate effect, the feature engineering extraction is performed on a plurality of sub-tracks, and after obtaining a sub-track feature matrix, the method further comprises the following operations:
and carrying out normalization operation on the features in the sub-track feature matrix according to the following formula:
wherein, delta represents the feature to be normalized, min and max represent the maximum value and the minimum value of the feature, respectively, and delta' represents the feature obtained after normalization processing.
In one embodiment, to simplify the training process, the method converts the trajectories of the vessels with different lengths into fixed lengths, and after normalizing the features in the sub-trajectory feature matrix, the method further includes:
and taking the number of the longest sub-track points as a standard, and filling 0 into other track data so that the lengths of the ship track feature vectors are the same.
S103: inputting the sub-track feature matrix into a multi-feature fusion self-encoder, and carrying out feature extraction on the ship feature vectors of the position, the speed and the course through the multi-feature fusion self-encoder to obtain the position feature vector, the speed feature vector and the course feature vector.
The multi-feature fusion self-encoder is used for carrying out data dimension reduction on the high-dimension ship track feature data to obtain low-latitude ship track feature data.
Specifically, the multi-feature fusion self-encoder in the embodiment of the present application includes three convolutional automatic encoders with the same structure, which are respectively:
the first convolution automatic encoder is used for learning the position type characteristics of the ship track;
a second convolution automatic encoder for learning a speed type characteristic of the ship track;
and the third convolution automatic encoder is used for learning the heading type characteristics of the ship track.
In one example, the location type characteristics include longitude and latitude; the speed type features include speed and acceleration; the heading-type features include heading and turning rate.
S104: and performing splicing operation on the position feature vector, the speed feature vector and the heading feature vector to obtain potential feature vectors of the ship track.
S105: and performing track clustering operation on the extracted ship track feature vectors to obtain a ship track clustering result.
In the embodiment of the application, the three types of the obtained trajectory feature vectors are spliced into a feature vector Z through the ship trajectory position feature vector Z1, the ship trajectory speed feature vector Z2 and the ship trajectory heading feature vector Z3 which are obtained after the feature extraction of the three convolution self-encoders. And taking the characteristic vector Z as the characteristic representation of all the ship tracks, and finally carrying out track clustering operation on the extracted ship track characteristic vector to obtain a ship track clustering result.
In one example, the tracks can be clustered based on the AIS ship track low-dimensional feature vector Z, and the clustering effect is evaluated by adopting an f1 value method, wherein the formula is as follows:
where TP represents a track correctly predicted as a positive example, FP represents a track incorrectly predicted as a positive example, and FN represents a track incorrectly predicted as a negative example.
In the embodiment of the application, performing track segmentation operation on an original ship AIS track to obtain segmented ship AIS sub-tracks, and performing feature engineering operation on each ship sub-track to obtain an AIS ship sub-track feature matrix, wherein an ith row in the matrix represents a feature value of the ith sub-track; the multi-feature fusion self-encoder is used for extracting the features of the ship feature vectors of the position, the speed and the course to obtain three types of feature vectors, then the three types of feature vectors are spliced to obtain potential feature vectors of the ship track, and the main stream clustering algorithm is used for carrying out track clustering operation on the extracted ship track feature vectors, so that a space-time track similarity measurement formula can be realized without manually selecting, similarity calculation deviation based on track distance caused by a traditional similarity measurement method can be avoided, the distribution of data of different ship feature types can be reserved, and the calculation performance is improved.
In one embodiment, for each convolution automatic encoder described above, the structure includes:
and the input layer is used for acquiring the characteristic matrix of each ship track characteristic data, expanding the sub-track data into sub-track vectors and transmitting the sub-track vectors to the encoder part.
In one example, a feature matrix of dimension (30, 2) is obtained for each piece of ship track feature data, where 30 represents the number of ship sub-track points and is also the sub-track length, and 2 represents each type of feature information contained in a single track point, including: location, speed, or heading type characteristics. Meanwhile, in order to ensure the timing relationship of the track sequence and make the convolution kernel filter cover all track points completely, the sub-track data is expanded into sub-track vectors with the length of 60×1 and transferred to the encoder section.
The encoder consists of a convolution layer and a pooling layer, and the activation function is set as a relu function and is used for carrying out convolution operation on the sub-track vectors and outputting the extracted ship track characteristics.
In one example, the encoder is comprised of a convolutional layer and a pooling layer. The convolutional layer 1 consists of 4 (4, 1) convolutional kernels, followed by a pooling layer for downsampling, which uses a maximum pooling model. The activation function is set to the relu function and the step size is set to 2 because two trace points are convolved at a time so that the network structure of the convolutional layer arrangement can ensure that the timing relationship between the two trace points is preserved. The ship track input of the model input is (60, 1), the ship track feature output dimension extracted after convolution operation is (1, 4), wherein 1 represents 1 feature, and 4 represents 4 filters. The one-dimensional convolution formula is as follows:
where f and e represent the trajectory feature vector and the convolution kernel. Alpha represents the length of the feature vector and beta represents the convolution kernel size. The feature extraction of two track points can be carried out by one-time convolution operation of the convolution kernel, so that the time sequence relation of the track points is ensured, local and contextual information about the ship track can be saved, and the analysis of abnormal conditions of the ship local track is facilitated.
And a decoder for implementing a deconvolution function using a combination of the convolution layer and the upsampling layer, adjusting the low-dimensional feature vector into a high-dimensional feature vector, and transmitting the high-dimensional feature vector to the output layer.
In one example, in the decoder section, the deconvolution function is implemented using a combination of a convolution layer and an upsampling layer to ensure that the dimension of the resulting output after the operation is not reduced, where convolution layer 2 consists of 4 (4, 1) size convolution kernel filters. The activation function is the relu function and the convolution step size is set to 1. The effect of the final deconvolution is to adjust the low-dimensional feature vector z to a (60, 4) -dimensional feature vector and transmit it to the output layer.
And the output layer is used for converting the high-dimensional feature vector into the original track feature vector so as to be the same as the latitude of the input data of the input layer.
In one example, the (60, 4) feature vector is converted into the original track feature vector size (60, 1), a convolution layer is used for the output layer, the convolution kernel with the dimension of 1 (4, 1) is used for the output layer, the dimension adjustment of the output data is realized through convolution, and finally the purpose identical to the latitude of the input data of the input layer is realized.
In one example, to achieve quantization and evaluation of errors in output layer data and input layer data, the mean square error is used to measure similarity between output layer data and input layer data, as follows:
wherein MSE represents mean square error, observed M Representing real track data, predicted M Representing predicted trajectory data, N representing the total number of sub-trajectories for all vessels.
As shown in fig. 3 and fig. 4, fig. 3 and fig. 4 are schematic diagrams of a flow chart and an applied model structure of a ship AIS track clustering method based on a convolution self-encoder according to an embodiment of the present application.
In one example, the ship AIS track clustering method based on the convolution self-encoder of the present application first divides one ship continuous AIS track TR into a plurality of sub-tracks: i.e., tr= { TR1, TR2,.,. Trj }, where trj is the ship sub-track, j represents the total number of sub-tracks of one ship; carrying out characteristic engineering operation and normalization on each ship sub-track, and finally obtaining a normalized AIS ship sub-track characteristic matrix, wherein the ith row in the matrix represents the characteristic value of the ith sub-track; track filling is carried out on the matrix, so that the lengths of all sub track vectors are equal, and the subsequent feature extraction operation is facilitated; extracting the characteristics of the ship characteristic vectors of the position, the speed and the course by using a multi-characteristic fusion self-encoder to obtain three types of characteristic vectors; and then performing splicing operation on the three types of feature vectors to obtain potential feature vectors of the ship track, and performing track clustering operation on the extracted ship track feature vectors by using a main stream clustering algorithm.
Fig. 5 is a schematic structural diagram of a ship AIS track clustering device based on a convolution self-encoder according to an embodiment of the present application, as shown in fig. 5, the ship AIS track clustering device 50 based on the convolution self-encoder includes:
the track acquisition module 51 is configured to acquire a continuous track of a ship, and divide the continuous track into a plurality of sub-tracks;
the feature engineering module 52 is configured to perform feature engineering extraction on a plurality of sub-tracks to obtain a sub-track feature matrix, where an ith row in the matrix represents a feature value of an ith sub-track;
the feature extraction module 53 is configured to input the sub-track feature matrix into a multi-feature fusion self-encoder, and perform feature extraction on the ship feature vectors of the position, the speed and the heading through the multi-feature fusion self-encoder to obtain a position feature vector, a speed feature vector and a heading feature vector;
the splicing module 54 is configured to perform a splicing operation on the position feature vector, the speed feature vector, and the heading feature vector, so as to obtain a potential feature vector of the ship track;
and the clustering module 55 is used for carrying out track clustering operation on the extracted ship track feature vectors to obtain a ship track clustering result.
In an exemplary embodiment, the track acquisition module 51 includes:
a start point determining unit, configured to determine a track point as a start point of a sub-track when a longitude difference or a change in altitude difference of the track point relative to a previous track point is smaller than a set threshold value and a speed of the track point is close to 0 in the continuous track;
and the end point determining unit is used for determining the track point with the speed close to 0 as the end point of the sub-track, wherein the longitude difference or the change of the altitude difference of the next track relative to the previous track point in the continuous track is smaller than a set threshold value.
In an exemplary embodiment, the ship AIS track clustering device 50 based on a convolution self-encoder further comprises:
the normalization module is used for carrying out feature engineering extraction on a plurality of sub-tracks to obtain a sub-track feature matrix, and then carrying out normalization operation on features in the sub-track feature matrix according to the following formula:
wherein, delta represents the feature to be normalized, min and max represent the maximum value and the minimum value of the feature, respectively, and delta' represents the feature obtained after normalization processing.
In an exemplary embodiment, the ship AIS track clustering device 50 based on a convolution self-encoder further comprises:
and the track data filling unit is used for carrying out 0 filling on other track data by taking the number of the longest sub-track points as a standard after carrying out normalization operation on the features in the sub-track feature matrix, so that the lengths of the ship track feature vectors are the same.
In one exemplary embodiment, the multi-feature fusion self-encoder includes:
the first convolution automatic encoder is used for learning the position type characteristics of the ship track;
a second convolution automatic encoder for learning a speed type characteristic of the ship track;
and the third convolution automatic encoder is used for learning the heading type characteristics of the ship track.
In an exemplary embodiment, the location type characteristics include longitude and latitude;
the speed type features include speed and acceleration;
the heading-type features include heading and turning rate.
In one exemplary embodiment, for each convolutional automatic encoder, the structure comprises:
the input layer is used for acquiring a characteristic matrix of each ship track characteristic data, expanding sub-track data into sub-track vectors and transmitting the sub-track vectors to the encoder part;
the encoder consists of a convolution layer and a pooling layer, an activation function is set as a relu function and is used for carrying out convolution operation on the sub-track vector and outputting extracted ship track characteristics;
a decoder for implementing a deconvolution function using a combination of a convolution layer and an up-sampling layer, adjusting a low-dimensional feature vector into a high-dimensional feature vector, and transmitting the high-dimensional feature vector to an output layer;
and the output layer is used for converting the high-dimensional feature vector into the original track feature vector so as to be the same as the latitude of the input data of the input layer.
In an exemplary embodiment, the ship AIS track clustering device 50 based on a convolution self-encoder further comprises:
the similarity evaluation unit is used for measuring the similarity between the output layer data and the input layer data by using the mean square error, and the formula is as follows:
wherein MSE represents mean square error, observed M Representing real track data, predicted M Representing predicted trajectory data, N representing the total number of sub-trajectories for all vessels.
In an exemplary embodiment, the ship AIS track clustering device 50 based on a convolution self-encoder further comprises:
the clustering effect evaluation unit is used for evaluating the clustering effect by adopting an f1 value method, and the formula is as follows:
where TP represents a track correctly predicted as a positive example, FP represents a track incorrectly predicted as a positive example, and FN represents a track incorrectly predicted as a negative example.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.
The above examples illustrate only a few embodiments of the invention, which are described in detail and are not to be construed as limiting the scope of the invention. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the invention, which are all within the scope of the invention.
Claims (2)
1. The ship AIS track clustering method based on the convolution self-encoder is characterized by comprising the following steps of:
acquiring a continuous track of a ship, and dividing the continuous track into a plurality of sub-tracks, wherein the method comprises the following steps of:
when the longitude difference or the change of the latitude difference of a certain track point relative to a previous track point in the continuous track is smaller than a set threshold value and the speed of the track point is close to 0, determining the track point as a starting point of a sub track;
determining a track point with the speed close to 0 as the end point of the sub track, wherein the longitude difference or the change of the latitude difference of the next track point relative to the previous track point in the continuous track is smaller than a set threshold value;
carrying out feature engineering extraction on a plurality of sub-tracks to obtain a sub-track feature matrix, wherein the ith row in the matrix represents the feature value of the ith sub-track;
further comprises:
and carrying out normalization operation on the features in the sub-track feature matrix according to the following formula:
wherein, delta represents the feature to be normalized, min and max represent the maximum value and the minimum value of the feature respectively, and delta' represents the feature obtained after normalization processing;
further comprises:
taking the number of the longest sub-track points as a standard, and filling other track data with 0 so that the lengths of the ship track feature vectors are the same;
inputting the sub-track feature matrix into a multi-feature fusion self-encoder, and carrying out feature extraction on the ship feature vectors of the position, the speed and the course through the multi-feature fusion self-encoder to obtain a position feature vector, a speed feature vector and a course feature vector;
the multi-feature fusion self-encoder includes:
the first convolution automatic encoder is used for learning the position type characteristics of the ship track; the location type characteristics include longitude and latitude;
a second convolution automatic encoder for learning a speed type characteristic of the ship track; the speed type features include speed and acceleration;
the third convolution automatic encoder is used for learning the course type characteristics of the ship track; the heading type features include heading and turning rate;
for each convolutional automatic encoder, its structure comprises:
the input layer is used for acquiring a characteristic matrix of each ship track characteristic data, expanding sub-track data into sub-track vectors and transmitting the sub-track vectors to the encoder part;
the encoder consists of a convolution layer and a pooling layer, an activation function is set as a relu function and is used for carrying out convolution operation on the sub-track vector and outputting extracted ship track characteristics;
a decoder for implementing a deconvolution function using a combination of a convolution layer and an up-sampling layer, adjusting a low-dimensional feature vector into a high-dimensional feature vector, and transmitting the high-dimensional feature vector to an output layer;
the output layer is used for converting the high-dimensional feature vector into the original track feature vector so as to enable the latitude of the high-dimensional feature vector to be the same as that of the input data of the input layer;
the mean square error is used to measure the similarity between the output layer data and the input layer data, and the formula is as follows:
wherein MSE represents mean square error, observed M Representing real track data, predicticted M Representing predicted trajectory data, N representing a total number of sub-trajectories for all vessels;
performing splicing operation on the position feature vector, the speed feature vector and the heading feature vector to obtain a potential feature vector of a ship track;
performing track clustering operation on the extracted ship track feature vectors to obtain a ship track clustering result;
and (3) evaluating the clustering effect by adopting an f1 value method, wherein the formula is as follows:
where TP represents a track correctly predicted as a positive example, FP represents a track incorrectly predicted as a positive example, and FN represents a track incorrectly predicted as a negative example.
2. A ship AIS track clustering device based on a convolution self-encoder, which executes the ship AIS track clustering method based on the convolution self-encoder as claimed in claim 1, and is characterized by comprising the following steps:
the track acquisition module is used for acquiring a continuous track of the ship and dividing the continuous track into a plurality of sub-tracks;
the characteristic engineering module is used for carrying out characteristic engineering extraction on a plurality of sub-tracks to obtain a sub-track characteristic matrix, wherein the ith row in the matrix represents the characteristic value of the ith sub-track;
the characteristic extraction module is used for inputting the sub-track characteristic matrix into a multi-characteristic fusion self-encoder, and carrying out characteristic extraction on the ship characteristic vectors of the position, the speed and the course through the multi-characteristic fusion self-encoder to obtain a position characteristic vector, a speed characteristic vector and a course characteristic vector;
the splicing module is used for carrying out splicing operation on the position feature vector, the speed feature vector and the heading feature vector to obtain potential feature vectors of the ship track;
and the clustering module is used for carrying out track clustering operation on the extracted ship track feature vectors to obtain a ship track clustering result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010507856.3A CN111694913B (en) | 2020-06-05 | 2020-06-05 | Ship AIS track clustering method and device based on convolution self-encoder |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010507856.3A CN111694913B (en) | 2020-06-05 | 2020-06-05 | Ship AIS track clustering method and device based on convolution self-encoder |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111694913A CN111694913A (en) | 2020-09-22 |
| CN111694913B true CN111694913B (en) | 2023-07-11 |
Family
ID=72479622
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010507856.3A Active CN111694913B (en) | 2020-06-05 | 2020-06-05 | Ship AIS track clustering method and device based on convolution self-encoder |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111694913B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113033443B (en) * | 2021-03-31 | 2022-10-14 | 同济大学 | Unmanned aerial vehicle-based automatic pedestrian crossing facility whole road network checking method |
| CN113221450B (en) * | 2021-04-27 | 2024-03-12 | 中国科学院国家空间科学中心 | Space-time prediction method and system for sparse non-uniform time sequence data |
| CN113434617B (en) * | 2021-06-21 | 2022-05-13 | 武汉理工大学 | Behavior automatic division method and system based on ship track and electronic equipment |
| CN113535861B (en) * | 2021-07-16 | 2023-08-11 | 子亥科技(成都)有限公司 | Track prediction method for multi-scale feature fusion and self-adaptive clustering |
| CN113780395B (en) * | 2021-08-31 | 2023-02-03 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Mass high-dimensional AIS trajectory data clustering method |
| CN115512152B (en) * | 2022-09-08 | 2024-09-13 | 武汉大学 | Ship track classification method and system based on CNN and LSTM neural network combination |
| CN118378117B (en) * | 2024-06-24 | 2024-08-20 | 大连海大赢海科技有限公司 | Ship data real-time intelligent analysis method based on data acquisition |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103456192A (en) * | 2013-09-01 | 2013-12-18 | 中国民航大学 | Terminal area prevailing traffic flow recognizing method based on track spectral clusters |
| CN109708638A (en) * | 2018-12-03 | 2019-05-03 | 江苏科技大学 | A kind of ship track point extracting method |
| CN110210352A (en) * | 2019-05-23 | 2019-09-06 | 中国人民解放军海军工程大学 | Ship track method for detecting abnormality based on navigation channel model |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7567203B2 (en) * | 2005-04-11 | 2009-07-28 | Raytheon Canada Limited | Classification system for radar and sonar applications |
| CN105205145A (en) * | 2015-09-18 | 2015-12-30 | 中国科学院自动化研究所 | Track modeling and searching method |
| US10345106B1 (en) * | 2015-10-29 | 2019-07-09 | National Technology & Engineering Solutions Of Sandia, Llc | Trajectory analysis with geometric features |
| CN107392937B (en) * | 2017-07-14 | 2023-03-14 | 腾讯科技（深圳）有限公司 | Target tracking method and device and electronic equipment |
| CN108334905A (en) * | 2018-02-08 | 2018-07-27 | 中电科技（合肥）博微信息发展有限责任公司 | Ship action trail recognition methods |
| CN109447135A (en) * | 2018-10-12 | 2019-03-08 | 天津大学 | A kind of new ship method of trajectory clustering |
| CN110033051B (en) * | 2019-04-18 | 2021-08-20 | 杭州电子科技大学 | Fishing trawler behavior discrimination method based on multi-step clustering |
| CN110232319B (en) * | 2019-05-07 | 2021-04-06 | 杭州电子科技大学 | Ship behavior identification method based on deep learning |
| CN110443287B (en) * | 2019-07-19 | 2022-01-14 | 北京航空航天大学 | Crowd moving stream drawing method based on sparse trajectory data |
-
2020
- 2020-06-05 CN CN202010507856.3A patent/CN111694913B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103456192A (en) * | 2013-09-01 | 2013-12-18 | 中国民航大学 | Terminal area prevailing traffic flow recognizing method based on track spectral clusters |
| CN109708638A (en) * | 2018-12-03 | 2019-05-03 | 江苏科技大学 | A kind of ship track point extracting method |
| CN110210352A (en) * | 2019-05-23 | 2019-09-06 | 中国人民解放军海军工程大学 | Ship track method for detecting abnormality based on navigation channel model |
Non-Patent Citations (1)
| Title |
|---|
| 袁冠等. 移动对象轨迹数据挖掘技术.中国矿业大学出版社,2016,(第1版),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111694913A (en) | 2020-09-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111694913B (en) | Ship AIS track clustering method and device based on convolution self-encoder | |
| CN113780395B (en) | Mass high-dimensional AIS trajectory data clustering method | |
| CN114022847B (en) | Method, system, equipment and storage medium for predicting intelligent body track | |
| CN103235933B (en) | A kind of vehicle abnormality behavioral value method based on HMM | |
| CN103336863B (en) | The flight intent recognition methods of flight path observed data of flying based on radar | |
| CN112906858A (en) | Real-time prediction method for ship motion trail | |
| CN112465199B (en) | Airspace situation assessment system | |
| CN112465273B (en) | Unmanned vehicle track prediction method based on local attention mechanism | |
| CN113032378B (en) | Ship behavior pattern mining method based on clustering algorithm and pattern mining | |
| CN108460481B (en) | Unmanned aerial vehicle reconnaissance target evolution rule prediction method based on recurrent neural network | |
| CN110610165A (en) | Ship behavior analysis method based on YOLO model | |
| CN112489497A (en) | Airspace operation complexity evaluation method based on deep convolutional neural network | |
| CN111367901B (en) | Ship data denoising method | |
| CN114972918B (en) | Remote sensing image ship target identification method based on integrated learning and AIS data | |
| CN111178438A (en) | ResNet 101-based weather type identification method | |
| CN110933633A (en) | Onboard environment indoor positioning method based on CSI fingerprint feature migration | |
| CN116306790A (en) | Offshore ship track real-time prediction method, system, equipment and medium based on CNN-GRU and attention mechanism | |
| CN115311617A (en) | Method and system for acquiring passenger flow information of urban rail station area | |
| CN117135686B (en) | Bluetooth-based vehicle-mounted information interaction method and system | |
| CN111242028A (en) | Remote sensing image ground object segmentation method based on U-Net | |
| CN116502777B (en) | Transformer-based four-dimensional track long-time prediction method and device | |
| CN114091578A (en) | Ship track clustering method based on curve length distance | |
| CN116719060B (en) | Information fusion-based method for detecting tight combination navigation faults of deep learning network | |
| CN113158415A (en) | Vehicle track similarity evaluation method based on error analysis | |
| CN113434617B (en) | Behavior automatic division method and system based on ship track and electronic equipment |
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 |