CN113360655A - Track point classification and text generation method based on sequence annotation - Google Patents

Track point classification and text generation method based on sequence annotation Download PDF

Info

Publication number
CN113360655A
CN113360655A CN202110711121.7A CN202110711121A CN113360655A CN 113360655 A CN113360655 A CN 113360655A CN 202110711121 A CN202110711121 A CN 202110711121A CN 113360655 A CN113360655 A CN 113360655A
Authority
CN
China
Prior art keywords
track
point
sequence
representing
window
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
Application number
CN202110711121.7A
Other languages
Chinese (zh)
Other versions
CN113360655B (en
Inventor
徐建
阮国庆
王鑫鹏
吴蔚
李晓冬
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
CETC 28 Research Institute
Original Assignee
CETC 28 Research Institute
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by CETC 28 Research Institute filed Critical CETC 28 Research Institute
Priority to CN202110711121.7A priority Critical patent/CN113360655B/en
Publication of CN113360655A publication Critical patent/CN113360655A/en
Application granted granted Critical
Publication of CN113360655B publication Critical patent/CN113360655B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/10—Text processing
    • G06F40/166—Editing, e.g. inserting or deleting
    • G06F40/186—Templates
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Navigation (AREA)

Abstract

The invention provides a track point classification and text generation method based on sequence labeling, which comprises the steps of firstly defining concerned event types and machine related parameters, such as turning events, wherein the turning events need to be concerned about turning angles; for hover events, the number of hover turns needs to be of concern. And judging the event type corresponding to each time point by analyzing the characteristics of the track points and the angles, height changes and the like of the adjacent points. And generating corresponding texts by adopting a template filling mode for each event and the parameters thereof. Through the steps, the flight path data can be automatically analyzed, the key events can be extracted, and the text description which is high in readability and reflects the activity condition of the target is generated.

Description

Track point classification and text generation method based on sequence annotation
Technical Field
The invention relates to a track point classification and text generation method based on sequence labeling.
Background
The flight path data is a time sequence and reflects the real activity rule of the target. However, because the flight path exists in a boring digital form, the flight path is not easy to read by human beings, and the behavior of the target cannot be intuitively understood, how to describe the flight path data in a text form is a problem to be solved. The existing method only classifies the whole track, does not analyze the fine granularity of each track point, and does not consider the problem of text conversion of the track.
Disclosure of Invention
The purpose of the invention is as follows: the invention discloses a flight path analysis and text generation method based on sequence labeling, which aims to describe flight path data in a text form, and comprises the following steps:
step 1, defining an event set of a target, and constructing a multi-dimensional characteristic for each track point to predict an event type of the track point;
step 2, extracting additional parameters aiming at specific events;
and 3, extracting parameters and filling texts.
The step 1 comprises the following steps:
step 1-1, defining an event set of a target:
the event set of the target comprises target occurrence, target departure, target steering and circle events, wherein the circle events need to extract specific circle turns, and the steering events need to extract steering angles;
step 1-2, vectorizing longitude and latitude coordinates:
the latitude and longitude in the track sequence is a continuous variable that needs to be discretized for subsequent model processing. Grouping the longitude and latitude by using Geohash to realize continuous face-changing discretization;
each track point in the track sequence is analogized to a word, the whole track sequence is analogized to a sentence, the vector representation of each word is obtained by utilizing a word vector training algorithm skip-gram, and the vector representation of each word is the vector representation of each track point;
step 1-3, extracting statistical characteristics of the flight path sequence: constructing a sliding window w with the size of n by taking the current track point as the center, and marking the sliding window w as Tw＝{P0,P1...Pi...Pn-1I, wherein the ith track point Pi＝(xi,yi,hi,ti) Is a quadruple, xi,yi,hi,tiRespectively representing the latitude, longitude, altitude and time of the ith track point; the curvature s of the sliding window is calculated as shown in the following equationwindow：
Figure BDA0003133803860000021
Wherein the distance is calculated by using the Euler distance:
Figure BDA0003133803860000022
distance(Pi,Pj) Indicating the ith track point PiAnd the jth track point PjThe distance of (d);
the curvature represents the ratio of the sum of the moving distances in the window to the linear distance of the starting position and the ending position of the window, and is used for describing the bending degree of the window;
step 1-4, calculating the rotation angle of the window relative to the whole track:
diffw＝θwindow-θtrack
θwindow＝arctan(yn-1-yw0,xn-1-xw0)
θtrack＝arctan(yT-1-yT0,xT-1-xT0)
wherein diffwRepresenting the angle of rotation, theta, of the window relative to the entire trackwindowAnd thetatrackRespectively representing the direction angle of the window and the direction angle of the track; (x)w0,yw0) And (x)n-1,yn-1) Respectively indicate the slideLatitude and longitude coordinates of the start point and latitude and longitude coordinates of the end point of the moving window, (x)T0,yT0) And (x)T-1,yT-1) Coordinates representing a starting point and an ending point of the track respectively;
step 1-5, the height vectorization of the track points is represented as follows:
processing the height in the flight path sequence by the same method of the step 1-2, wherein the height in the flight path sequence is a continuous variable, the height is coded into a category variable by using quantile statistic, and the height of each time point is similar to a word; after discretization, the track of a single target is changed into a sequence, and the sequence is analogized into a sentence; obtaining a vector representation of each word by using a word vector training algorithm cbow, wherein the vector representation of each word is a height-based vector representation of each time point;
1-6, modeling time sequence data by using a bidirectional long-short term memory network (BilSTM), and inputting various characteristics of a flight path sequence at a given time step t:
BiLSTM([et,cft,difft,eht])
wherein etRepresenting the vector representation obtained by the track point of the current time point;
cfta warping factor representing a current point in time;
difftrepresenting an angle difference characteristic of a current time point;
ehta height-based vector representation representing a current point in time;
1-7, modeling the dependency relationship between the tags by using crf:
for a given track sequence TiAnd (3) recording the characteristic output after the long-short term memory network coding of the steps 1-6 as P, wherein the dimensionality is n multiplied by k, wherein n represents the number of track points, k represents the number of all events, and Pi,jRepresenting the score of mapping the track point i to the event j; for one of the output sequences y ═ (y)1,y2...yn) The score s (X, y) is recorded as:
Figure BDA0003133803860000031
wherein the matrix A is a transition matrix, in particular Ai，jRepresenting the transition probability of label i to label j;
to track sequence TiAll outputs are normalized to obtain a posterior probability p (y | X):
Figure BDA0003133803860000032
wherein Y isXRepresents all possible event sets for input sequence X;
Figure BDA0003133803860000033
is YXOne element of (1);
wherein
Figure BDA0003133803860000034
All possible event sequences representing the input sequence X for a loop;
the training process is intended to maximize the maximum likelihood of a correct tag sequence:
Figure BDA0003133803860000035
prediction phase selection maximizes the conditional probability:
Figure BDA0003133803860000036
wherein y is*Representing the sequence of events that the flight path ultimately predicts.
In step 1-1, the number of turns of the coil is calculated by the following method:
note PmAnd PnRepresenting the start and end points of a hover event, respectively, and the set of trajectory points recording this time period is T ═ Pm,Pm+1...PnCalculating the direction (theta) and the rotation angle (diff) of two adjacent track points,summing all rotation angles as characteristic f of the spiral turn number decisioniAnd then judging the number of spiral turns by utilizing linear regression:
θi+1＝arctan(yi+2-yi+1,xi+2-xi+1)
θi＝arctan(yi+1-yi,xi+1-xi)
diffi＝θi+1-θi
Figure BDA0003133803860000041
wherein theta isiRepresenting two adjacent track points PiAnd Pi+1Direction of (d), diffiDenotes thetaiAnd thetai+1The corner of (d); point PiRespectively have a longitude and latitude of xi、yi；
The step 3 comprises the following steps:
step 3-1, generating a text for the track points predicted to be turned according to the following mode: finding more than two continuous track points with labels as steering, and storing longitude and latitude coordinate points of the track points as a coordinate set; for the set of coordinates, calculating an angle of the start and end positions:
angle＝arctan(y1-y0,x1-x0)
wherein (x)0,y0) Latitude and longitude (x) representing the starting position1,y1) Representing a longitude and latitude of the ending location; the angle value range is [ -180, 180 [ -180 [ ]]And according to 45 degrees, the angle is divided into 8 directions at intervals, namely east, northeast, north, northwest, west, southwest, south and southeast, and then the angle is mapped into the corresponding direction:
argminθ|angle-θ|
selecting the smallest theta as the predicted direction;
and 3-2, for track points predicted to be in a spiral, firstly finding a spiral starting point and an end point, predicting the number of spiral turns by using the point of the set of the spiral starting point and the end point (obtained by using a sequence expression model, namely the model in the step 2), and filling texts.
The method comprises the steps of firstly defining concerned event types and machine related parameters, such as steering angle of a steering event needing to be concerned; for hover events, the number of hover turns needs to be of concern. And judging the event type corresponding to each time point by analyzing the characteristics of the track points and the angles, height changes and the like of the adjacent points. And generating corresponding texts by adopting a template filling mode for each event and the parameters thereof. The method automatically converts the aircraft track into the text description which has strong readability and reflects the activity condition of the track target. The track data mentioned herein is a kind of structured time series data, including the longitude, latitude, altitude, speed and other raw information of the target at each time point; the corresponding text description is described for the event in the track, wherein the event comprises 4 event types of the occurrence, departure, turning, circling and the like of the target, wherein the turning event requires the extraction of information of turning angle, the circling event requires the extraction of circling number and the like. In the process, modeling the flight path data into a sequence labeling problem, extracting multi-dimensional characteristics of each time point, and predicting the event type of a given time point by utilizing the incidence relation between the bidirectional lstm and the crf network modeling event labels; extracting additional parameters for specific events, for example, judging the number of spiral turns by linear regression for the problem that the number of spiral turns needs to be judged in the spiral process; and for track generation, establishing a template for each event, and generating a text based on template filling.
The invention has the following remarkable advantages: modeling the flight path data into a sequence labeling problem, extracting multi-dimensional characteristics of each time point, and predicting the event type of a given time point by using the incidence relation between the bidirectional lstm and the crf network modeling event labels; for a specific event, additional parameters need to be extracted, for example, for the problem that the number of spiral turns needs to be judged in the spiral process, the number of turns is judged by using linear regression; and for track generation, establishing a template for each event, and generating a text based on template filling. Through the 3 steps, the event type corresponding to each time point track can be predicted, corresponding parameters of the event are extracted, corresponding texts are generated according to the template, and textual description of the flight path data is achieved.
Drawings
The foregoing and/or other advantages of the invention will become further apparent from the following detailed description of the invention when taken in conjunction with the accompanying drawings.
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
As shown in fig. 1, the invention provides a track point classification and text generation method based on sequence annotation, which specifically includes the following steps:
the step 1 comprises the following steps:
step 1-1, defining an event set of a target:
firstly, defining an event set of a target, including the appearance, departure, target steering and hovering of the target, wherein a specific circle number needs to be extracted for a hovering event, and a steering angle needs to be extracted for a steering event;
step 1-2, vectorizing longitude and latitude coordinates:
the latitude and longitude of the track sequence is a continuous variable, which needs to be discretized for subsequent model processing. Grouping the longitude and latitude by using Geohash to realize continuous face-changing discretization;
each track point in the track sequence is analogized to a word, the whole track sequence is analogized to a sentence, the vector representation of each word is obtained by utilizing a word vector training algorithm skip-gram, and the vector representation of each word is the vector representation of each track point; shown in table 1 are course point data:
TABLE 1
Figure BDA0003133803860000061
Step 1-3, extracting statistical characteristics of the flight path sequence: constructing a sliding window w with the size of n by taking the current track point as the center, and marking the sliding window w as Tw＝{P0,P1...Pi...Pn-1Therein, track point Pi＝(xi,yi,hi,ti) Is a quadruple representing latitude, longitude, altitude and time. The curvature of the sliding window is calculated as follows:
Figure BDA0003133803860000062
wherein the distance is calculated by using Euler distance
Figure BDA0003133803860000063
The curvature represents the ratio of the sum of the moving distances in the window to the linear distance of the starting position and the ending position of the window, and is used for describing the bending degree of the window.
Step 1-4, calculating the rotation angle of the window relative to the flight path:
diffw＝θwindow-θtrack
θwindow＝arctan(yn-1-yw0,xn-1-xw0)
θtrack＝arctan(yT-1-yT0,xT-1-xT0)
where diff represents the angle of rotation of the window relative to the entire track, θwindowAnd thetatrackThe direction angle of the window and the direction angle of the track are respectively represented, (x)w0,yw0) And (x)n-1,yn-1) Respectively representing the longitude and latitude coordinates of the start point and the end point of the sliding window,
(xT0,yT0) And (x)T-1,yT-1) Coordinates representing the starting point and the ending point of the track, respectively.
Step 1-5, the height vectorization of the track points is represented as follows:
processing the height in the flight path sequence by the same method of the step 1-2, wherein the height in the flight path sequence is a continuous variable, the height is coded into a category variable by using quantile statistic, and the height of each time point is similar to a word; after discretization, the track of a single target is changed into a sequence, and the sequence is analogized into a sentence; obtaining a vector representation of each word by using a word vector training algorithm cbow, wherein the vector representation of each word is a height-based vector representation of each time point;
1-6, modeling time sequence data by using a bidirectional long-short term memory network (BilSTM), and inputting various characteristics of a flight path sequence at a given time step t:
BiLSTM([et,cft,difft,eht])
wherein etRepresenting the vector representation obtained by the track point of the current time point;
cfta warping factor representing a current point in time;
difftrepresenting an angle difference characteristic of a current time point;
ehta height-based vector representation representing a current point in time;
1-7, modeling the dependency relationship between the tags by using crf:
for a given track sequence TiAnd (3) recording the characteristic output after the long-short term memory network coding of the steps 1-6 as P, wherein the dimensionality is n multiplied by k, wherein n represents the number of track points, k represents the number of all events, and the specific P isi,jRepresenting the score of a track point i mapped to an event j, for one of the output sequences y ═ y (y)1,y2...yn) The score s (X, y) is recorded as:
Figure BDA0003133803860000071
wherein the matrix A is a transition matrix, in particular Ai,jIndicating the transition probability of label i transitioning to label j.
To track sequence TiAll outputs are normalized to obtain a posterior probability:
Figure BDA0003133803860000081
the training process is intended to maximize the maximum likelihood of a correct tag sequence:
Figure BDA0003133803860000082
prediction phase selection maximizes the conditional probability:
Figure BDA0003133803860000083
wherein y is*Representing the sequence of events that the flight path ultimately predicts.
And (3) aiming at the spiral event, additionally extracting the number of spiral turns, and calculating the number of the spiral turns by adopting the following method:
note PmAnd PnRepresenting the start and end points of a hover event, respectively, and the set of trajectory points recording this time period is T ═ Pm,Pm+1...PnCalculating the directions and the rotation angles of two adjacent track points, and summing all the rotation angles to be used as the characteristic f for judging the number of turns of the circleiAnd then judging the number of spiral turns by utilizing linear regression.
θi+1＝arctan(yi+2-yi+1,xi+2-xi+1)
θi＝arctan(yi+1-yi,xi+1-xi)
diffi＝θi+1-θi
Figure BDA0003133803860000084
The step 3 comprises the following steps:
step 3-1, generating a text for the track points predicted to be turned according to the following mode: finding more than two continuous track points with labels as steering, and storing longitude and latitude coordinate points of the track points as a coordinate set; for the set of coordinates, calculating an angle of the start and end positions:
angle＝arctan(y1-y0,x1-x0)
wherein (x)0,y0) Latitude and longitude (x) representing the starting position1,y1) Representing a longitude and latitude of the ending location; the angle value range is [ -180, 180 [ -180 [ ]]And according to 45 degrees, the angle is divided into 8 directions at intervals, namely east, northeast, north, northwest, west, southwest, south and southeast, and then the angle is mapped into the corresponding direction:
argminθ|angle-θ|
selecting the smallest theta as the predicted direction;
and 3-2, for track points predicted to be in a spiral, firstly finding a spiral starting point and an end point, predicting the number of spiral turns by using the set of the spiral starting point and the end point, and filling texts. Shown below are track correspondence texts:
015108H, type 1 XX appearing near Lon85.8Lat126.2, flying toward west;
between 021728H and 030631H, the airplane space Lon84.1Lat126.2-Lon84.2Lat126.2-Lon84.2Lat126.1-Lon84.1Lat126.1 flies in the west and south directions in a reciprocating way for 3 weeks;
032927H, the position Lon82.5Lat126.3 disappears.
The invention provides a track point classification and text generation method based on sequence annotation, and a plurality of methods and ways for implementing the technical scheme are provided. All the components not specified in the present embodiment can be realized by the prior art.

Claims (10)

1. A track point classification and text generation method based on sequence labeling is characterized by comprising the following steps:
step 1, defining an event set of a target, and constructing a multi-dimensional characteristic for each track point to predict an event type of the track point;
step 2, extracting additional parameters aiming at specific events;
and 3, extracting parameters and filling texts.
2. The method of claim 1, wherein step 1 comprises:
step 1-1, defining an event set of a target:
the event set of the target comprises target occurrence, target departure, target steering and circle events, wherein the circle events need to extract specific circle turns, and the steering events need to extract steering angles;
step 1-2, vectorizing longitude and latitude coordinates;
step 1-3, extracting statistical characteristics of a track sequence;
step 1-4, calculating the turning angle of the window relative to the whole track;
step 1-5, the height vectorization of the track points is represented as follows:
1-6, modeling time sequence data by using a bidirectional long-short term memory network (BilSTM);
and 1-7, modeling the dependency relationship between the labels by using crf.
3. A method according to claim 2, characterized in that in step 1-1 the number of turns is calculated as follows:
note PmAnd PnRepresenting the start and end points of a hover event, respectively, and the set of trajectory points recording this time period is T ═ Pm,Pm+1...PnCalculating the directions and the rotation angles of two adjacent track points, and summing all the rotation angles to be used as the characteristic f for judging the number of turns of the circleiAnd then judging the number of spiral turns by utilizing linear regression:
θi+1＝arctan(yi+2-yi+1,xi+2-xi+1)
θi＝arctan(yi+1-yi,xi+1-xi)
diffi＝θi+1-θi
Figure FDA0003133803850000011
wherein theta isiRepresenting two adjacent track points PiAnd Pi+1Direction of (d), diffiDenotes thetaiAnd thetai+1The corner of (d); point PiRespectively have a longitude and latitude of xi、yi。
4. The method of claim 3, wherein steps 1-2 comprise: the longitude and latitude in the flight path sequence are continuous variables, and the longitude and latitude are grouped by using Geohash to realize the discretization of continuous face changing;
each track point in the track sequence is analogized to a word, the whole track sequence is analogized to a sentence, the vector representation of each word is obtained by utilizing a word vector training algorithm skip-gram, and the vector representation of each word is the vector representation of each track point.
5. The method of claim 4, wherein steps 1-3 comprise: constructing a sliding window w with the size of n by taking the current track point as the center, and marking the sliding window w as Tw＝{P0,P1...Pi...Pn-1I, wherein the ith track point Pi＝(xi,yi,hi,ti) Is a quadruple, xi,yi,hi,tiRespectively representing the latitude, longitude, altitude and time of the ith track point; the curvature s of the sliding window is calculated as shown in the following equationwindow：
Figure FDA0003133803850000021
Wherein the distance is calculated by using the Euler distance:
Figure FDA0003133803850000022
distance(Pi,Pj) Indicating the ith track point PiAnd the jth track point PjThe distance of (d);
the curvature represents the ratio of the sum of the moving distances in the window to the linear distance of the starting position and the ending position of the window, and is used for describing the bending degree of the window.
6. The method of claim 5, wherein steps 1-4 comprise: and calculating the rotation angle of the window relative to the whole track by adopting the following formula:
diffw＝θwindow-θtrack
θwindow＝arctan(yn-1-yw0,xn-1-xw0)
θtrack＝arctan(yT-1-yT0,xT-1-xT0)
wherein diffwRepresenting the angle of rotation, theta, of the window relative to the entire trackwindowAnd thetatrackRespectively representing the direction angle of the window and the direction angle of the track; (x)w0,yw0) And (x)n-1,yn-1) Respectively representing longitude and latitude coordinates of a start point and a longitude and latitude coordinate of an end point of the sliding window, (x)T0,yT0) And (x)T-1,yT-1) Respectively representing the coordinates of the starting point and the ending point of the track.
7. The method of claim 6, wherein steps 1-5 comprise: the height in the flight path sequence is a continuous variable, the height is coded into a category variable by utilizing quantile statistics, and the height of each time point is similar to a word; after discretization, the track of a single target is changed into a sequence, and the sequence is analogized into a sentence; the word vector training algorithm cbow is used to obtain a vector representation of each word, which is a height-based vector representation at each time point.
8. The method according to claim 7, wherein in steps 1-6, the time series data is modeled by using a bidirectional long-short term memory network (BilSTM), and the input is a plurality of characteristics of the flight path sequence at a given time step t:
BiLSTM([et,cft,difft,eht])
wherein etRepresenting the vector representation obtained by the track point of the current time point;
cfta warping factor representing a current point in time;
difftrepresenting an angle difference characteristic of a current time point;
ehta height-based vector representation representing a current point in time.
9. The method of claim 8, wherein steps 1-7 comprise: for a given track sequence TiAnd (3) recording the characteristic output after the long-short term memory network coding of the steps 1-6 as P, wherein the dimensionality is n multiplied by k, wherein n represents the number of track points, k represents the number of all events, and Pi,jRepresenting the score of mapping the track point i to the event j; for one of the output sequences y ═ (y)1,y2...yn) The score s (X, y) is recorded as:
Figure FDA0003133803850000031
wherein the matrix A is a transition matrix, in particular Ai,jRepresenting the transition probability of label i to label j;
to track sequence TiAll outputs are normalized to obtain a posterior probability p (y | X):
Figure FDA0003133803850000032
wherein Y isXRepresents all possible event sets for input sequence X;
Figure FDA0003133803850000033
is YXOne element of (1);
the training process is intended to maximize the maximum likelihood of a correct tag sequence:
Figure FDA0003133803850000041
prediction phase selection maximizes the conditional probability:
Figure FDA0003133803850000042
wherein y is*Representing the sequence of events that the flight path ultimately predicts.
10. The method of claim 9, wherein step 3 comprises:
step 3-1, generating a text for the track points predicted to be turned according to the following mode: finding more than two continuous track points with labels as steering, and storing longitude and latitude coordinate points of the track points as a coordinate set; for the set of coordinates, calculating an angle of the start and end positions:
angle＝arctan(y1-y0,x1-x0)
wherein (x)0,y0) Latitude and longitude (x) representing the starting position1,y1) Representing a longitude and latitude of the ending location; the angle value range is [ -180, 180 [ -180 [ ]]And according to 45 degrees, the angle is divided into 8 directions at intervals, namely east, northeast, north, northwest, west, southwest, south and southeast, and then the angle is mapped into the corresponding direction:
argminθ|angle-θ|
selecting the smallest theta as the predicted direction;
and 3-2, for track points predicted to be in a spiral, firstly finding a spiral starting point and an end point, predicting the number of spiral turns by using the set of the spiral starting point and the end point, and filling texts.
CN202110711121.7A 2021-06-25 2021-06-25 Track point classification and text generation method based on sequence annotation Active CN113360655B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110711121.7A CN113360655B (en) 2021-06-25 2021-06-25 Track point classification and text generation method based on sequence annotation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110711121.7A CN113360655B (en) 2021-06-25 2021-06-25 Track point classification and text generation method based on sequence annotation

Publications (2)

Publication Number Publication Date
CN113360655A true CN113360655A (en) 2021-09-07
CN113360655B CN113360655B (en) 2022-10-04

Family

ID=77536495

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110711121.7A Active CN113360655B (en) 2021-06-25 2021-06-25 Track point classification and text generation method based on sequence annotation

Country Status (1)

Country Link
CN (1) CN113360655B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114708227A (en) * 2022-03-31 2022-07-05 海门喜满庭纺织品有限公司 Dyeing diffusant adding control method and system based on image processing
CN115577118A (en) * 2022-09-30 2023-01-06 南京信息工程大学 Text generation method based on mixed grouping ordering and dynamic entity memory planning

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108829801A (en) * 2018-06-06 2018-11-16 大连理工大学 A kind of event trigger word abstracting method based on documentation level attention mechanism
CN108897989A (en) * 2018-06-06 2018-11-27 大连理工大学 A kind of biological event abstracting method based on candidate events element attention mechanism
WO2019179100A1 (en) * 2018-03-20 2019-09-26 苏州大学张家港工业技术研究院 Medical text generation method based on generative adversarial network technology
WO2019192397A1 (en) * 2018-04-04 2019-10-10 华中科技大学 End-to-end recognition method for scene text in any shape

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019179100A1 (en) * 2018-03-20 2019-09-26 苏州大学张家港工业技术研究院 Medical text generation method based on generative adversarial network technology
WO2019192397A1 (en) * 2018-04-04 2019-10-10 华中科技大学 End-to-end recognition method for scene text in any shape
CN108829801A (en) * 2018-06-06 2018-11-16 大连理工大学 A kind of event trigger word abstracting method based on documentation level attention mechanism
CN108897989A (en) * 2018-06-06 2018-11-27 大连理工大学 A kind of biological event abstracting method based on candidate events element attention mechanism

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114708227A (en) * 2022-03-31 2022-07-05 海门喜满庭纺织品有限公司 Dyeing diffusant adding control method and system based on image processing
CN114708227B (en) * 2022-03-31 2023-10-27 海门喜满庭纺织品有限公司 Method and system for controlling addition of dyeing diffusion agent based on image processing
CN115577118A (en) * 2022-09-30 2023-01-06 南京信息工程大学 Text generation method based on mixed grouping ordering and dynamic entity memory planning
CN115577118B (en) * 2022-09-30 2023-05-30 南京信息工程大学 Text generation method based on mixed grouping ordering and dynamic entity memory planning

Also Published As

Publication number Publication date
CN113360655B (en) 2022-10-04

Similar Documents

Publication Publication Date Title
US11775838B2 (en) Image captioning with weakly-supervised attention penalty
US20210382937A1 (en) Image processing method and apparatus, and storage medium
Ferstl et al. Streamline variability plots for characterizing the uncertainty in vector field ensembles
CN113360655B (en) Track point classification and text generation method based on sequence annotation
US11782928B2 (en) Computerized information extraction from tables
CN111695527B (en) Mongolian online handwriting recognition method
Jolly Machine learning with scikit-learn quick start guide: classification, regression, and clustering techniques in Python
Kosmala et al. On-line handwritten formula recognition using hidden Markov models and context dependent graph grammars
Qi et al. A generalized earley parser for human activity parsing and prediction
Singh et al. Online handwriting recognition systems for Indic and non-Indic scripts: a review
CN110334209A (en) File classification method, device, medium and electronic equipment
CN108256463B (en) Mobile robot scene recognition method based on ESN neural network
US11373043B2 (en) Technique for generating and utilizing virtual fingerprint representing text data
Deufemia et al. Petroglyph recognition using self-organizing maps and fuzzy visual language parsing
Singh et al. A self controlled RDP approach for feature extraction in online handwriting recognition using deep learning
Wang et al. Matching suitability analysis for geomagnetic aided navigation based on an intelligent classification method
Sen et al. Design of novel feature vector for recognition of online handwritten Bangla basic characters
CN113221628A (en) Video violence identification method, system and medium based on human skeleton point cloud interactive learning
Hu et al. On-line handwritten mathematical expression recognition method based on statistical and semantic analysis
Yang et al. IF-MCA: Importance factor-based multiple correspondence analysis for multimedia data analytics
Xiao et al. Motion retrieval based on Motion Semantic Dictionary and HMM inference
Pourian et al. Pixnet: A localized feature representation for classification and visual search
CN115130475A (en) Extensible universal end-to-end named entity identification method
Bezine et al. Handwriting perceptual classification and synthesis using discriminate HMMs and progressive iterative approximation
Lu et al. Self-reinforcing unsupervised matching

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