CN113537347A - Unmanned aerial vehicle and flying bird target classification method based on track motion characteristics - Google Patents

Unmanned aerial vehicle and flying bird target classification method based on track motion characteristics Download PDF

Info

Publication number
CN113537347A
CN113537347A CN202110802116.7A CN202110802116A CN113537347A CN 113537347 A CN113537347 A CN 113537347A CN 202110802116 A CN202110802116 A CN 202110802116A CN 113537347 A CN113537347 A CN 113537347A
Authority
CN
China
Prior art keywords
target
track
course
joint probability
matrix
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
CN202110802116.7A
Other languages
Chinese (zh)
Other versions
CN113537347B (en
Inventor
刘佳
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.)
Beihang University
Original Assignee
Beihang University
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 Beihang University filed Critical Beihang University
Priority to CN202110802116.7A priority Critical patent/CN113537347B/en
Publication of CN113537347A publication Critical patent/CN113537347A/en
Application granted granted Critical
Publication of CN113537347B publication Critical patent/CN113537347B/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
    • 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
    • G06F18/24155—Bayesian classification
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention discloses an unmanned aerial vehicle and bird target classification method based on track motion characteristics, and belongs to the technical field of radar target tracking and identification. The method comprises the following steps: extracting a motion characteristic vector for each target track, wherein the motion characteristic vector comprises an average speed, a speed standard deviation, a course standard deviation, a maneuvering factor and a track oscillation factor; combining motion characteristic vectors extracted from different target tracks into a training sample set, establishing a hierarchical table for each characteristic element, and randomly selecting two characteristic elements to construct a joint probability matrix of each type of target; and extracting motion characteristic vectors of unknown target tracks, and identifying the target categories according to the hierarchical table and the joint probability matrix. The method and the device realize effective identification of the unmanned aerial vehicle and the flying bird target, avoid the problem of high identification difficulty of the flying bird target and the unmanned aerial vehicle target caused by high overlapping degree of radar scattering section values, and have high identification efficiency and strong universality.

Description

Unmanned aerial vehicle and flying bird target classification method based on track motion characteristics
Technical Field
The invention belongs to the technical field of radar target tracking and identification, relates to radar target feature extraction and identification classification, and particularly relates to an unmanned aerial vehicle and a flying bird target classification method based on track motion features.
Background
Flying birds and unmanned aerial vehicle target all belong to typical "low little" target slowly, and this type target generally possesses the flight height and hangs down, flight speed is slow, radar scattering cross section is little, detectability low grade characteristic. The flying bird target has greater security threat to civil aviation airliners in the stage of entering and exiting ports, and the rapid development of commercial-grade small unmanned planes enables the small unmanned planes to be more easily utilized by illegal molecules, thereby seriously threatening the low-altitude security of key areas such as airports and the like. Therefore, the accurate identification and tracking of the targets of the flying birds and the unmanned aerial vehicles have important application and research values.
The low-altitude monitoring radar is applied to airport, wind power plant, border area and other important areas at present. The existing low-altitude monitoring radar generally adopts a Doppler signal processing technology based on solid-state power amplification, has small target detection capability in a complex low-altitude strong ground clutter environment, and realizes all-weather high-resolution continuous monitoring of a specific airspace. However, a bird target in a low-altitude airspace generally has a high similarity to a low-altitude unmanned machine in terms of a radar scattering cross section, polarization characteristics, flight speed, altitude, and the like. The radar target micro Doppler feature theoretically has the potential of accurately distinguishing flying birds from rotor unmanned aerial vehicles, but the micro Doppler feature quality is highly related to the monitoring distance in an actual application scene, the difficulty in extracting the small target long-distance micro Doppler feature is high, and the application value in actual radar monitoring needs to be further improved. Therefore, the existing low-altitude surveillance radar is difficult to effectively distinguish the bird and the unmanned aerial vehicle target.
Disclosure of Invention
The invention aims to solve the problems, provides a rotor unmanned aerial vehicle and a bird target classification and identification method based on radar track motion characteristics, and is suitable for effectively identifying the bird and the unmanned aerial vehicle target in a low-altitude complex environment.
A classification and identification method for a rotor unmanned aerial vehicle and a flying bird target based on radar track motion characteristics extracts a target motion characteristic vector by utilizing radar target track information and distinguishes the flying bird target from the unmanned aerial vehicle target by adopting a naive Bayes method, and comprises the following steps:
step one, extracting a motion characteristic vector for a target track acquired by a radar; the motion feature vector includes feature elements: average speed, speed standard deviation, course standard deviation, maneuvering factor and track oscillation factor;
step two, extracting motion characteristic vectors to form a training sample set for tracks of different targets by adopting the step 1, and then constructing a joint probability matrix of motion characteristics of the different targets;
and thirdly, judging the type of the unknown target track by using the obtained joint probability matrix.
In the first step, motion characteristic vectors including average speed, speed standard deviation, course standard deviation, maneuvering factors and flight path oscillation factors are extracted from a target flight path collected by a radar; the maneuvering factor is calculated as the ratio of the average speed to the standard deviation of the course; the track oscillation factor is calculated as follows: firstly, calculating the k-th course difference symbol in the course difference symbol vector O of the target track
Figure BDA0003165054080000021
Wherein, Δ h (k) is the course angle difference between the k +1 th track sampling point and the k-th track sampling point, and δeIs a measurement error threshold; then, the course oscillation frequency existing in the target track is judged according to the vector O, if one of the following two oscillation modes exists, the course oscillation is judged to exist once,
oscillation mode 1: o (i-1) + O (i) ≠ O (i) 0 and O (i-1) ≠ O (i);
oscillation mode 2: o (i-1) + O (i +1) ≠ O (i +1), O (i) ≠ 0;
finally, calculating the oscillation factor
Figure BDA0003165054080000022
Wherein, | Δ h (i) | is the absolute value of the change of the course angle under the ith oscillation, and w (i) is a weight factor.
The second step comprises the following steps: (1) firstly, establishing a hierarchical table for each feature element; (2) secondly, two characteristic elements in the motion characteristic vector are selected randomly to construct a joint probability matrix of each type of target; when a joint probability matrix of a t-type target is constructed, initializing the joint probability matrix to be a 0 matrix, traversing motion characteristic vectors of each flight path of the t-type target in a training sample set, acquiring levels m and n corresponding to the numerical values of two selected characteristic elements according to a hierarchical table, increasing the value of the element (m, n) in the joint probability matrix by 1, and normalizing the element in the joint probability matrix after traversing is completed; the value of the element (m, n) in the obtained joint probability matrix represents that for the t-th class object, the probability that the first selected feature element belongs to the level m under the condition that the first selected feature element belongs to the level n is p.
And in the third step, extracting a motion characteristic vector from the unknown target track acquired by the radar, acquiring the grade of each characteristic value according to the hierarchical table, searching the probability value of the unknown target track belonging to the class of targets from 20 joint probability matrixes for each class of targets, and finally selecting the class with the highest probability as the class of the unknown target track.
Compared with the prior art, the invention has the advantages and positive effects that:
(1) the method provided by the invention extracts the target motion characteristics by using the target track information, realizes effective identification of the flying bird and the unmanned aerial vehicle target without depending on radar echo intensity information such as a radar scattering cross section, and avoids the problem of high identification difficulty of two types of targets caused by high numerical overlapping degree of the radar scattering cross sections.
(2) The method has the advantages of high speed of constructing the target characteristic vector, high identification efficiency and strong flexibility, and can be expanded and applied to identification of other low-altitude interference targets such as precipitation clutter and the like.
(3) According to the method, software and hardware equipment of the low-altitude airspace radar monitoring system do not need to be changed or upgraded, and effective identification of the unmanned aerial vehicle and the flying bird target can be realized only by performing feature extraction and feature vector reconstruction on the radar target track, so that the universality is high.
Drawings
FIG. 1 is a schematic diagram of a radar target identification process implemented by the method of the present invention;
fig. 2 is a diagram illustrating a distribution of trajectories of the small unmanned aerial vehicle and the bird target according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples.
The invention relates to an unmanned aerial vehicle based on track motion characteristics and a flying bird target classification method, which utilizes the difference of target motion modes of flying birds and rotor unmanned aerial vehicles to realize effective classification and identification of light unmanned aerial vehicles and flying bird targets in complex low-altitude environments, and comprises the following steps:
step one, extracting a radar target track motion characteristic vector.
Collecting the track of a moving target by a radar, and defining the target track Z ═ Z1,z2,...,zN]And extracting the speed and the course information vectors V and H according to the space-time information of each sampling point in the track as follows:
V＝[v1,v2,...,vN]；H＝[h1,h2,...,hN] (1)
wherein, N represents the number of sampling points in the flight path, i.e. the flight path length. v. ofi、hiRespectively, the speed and heading of the target at the ith sampling point, i ═ 1,2, … N. The length of the target track is closely related to the characteristics of the target, the motion mode and the radar tracking algorithm, and the lengths of the tracks are usually inconsistent. The invention constructs the motion characteristic vector of the target track by extracting the statistical descriptor of the track motion characteristic, realizes the consistency of the feature dimension of the target track, and is beneficial to realizing target identification by adopting a supervised machine learning algorithm. The invention extracts the following five types of track motion characteristics:
(1) average velocity vmeanThe calculation is as follows:
Figure BDA0003165054080000031
(2) standard deviation of velocity vstdThe calculation is as follows:
Figure BDA0003165054080000032
(3) heading standard deviation hstdThe calculation is as follows:
Figure BDA0003165054080000033
parameters in equation (4)Δh(i) Is defined as:
Figure BDA0003165054080000034
wherein, deltahThe present invention sets the threshold value of the heading angle difference to 50 °. h ismeanThe average value of the course of the target track is obtained by averaging the courses of the N sampling points in the track.
(4) The mobility factor σ, defined as shown in equation (6), describes the mobility performance of the target. The larger the value of σ, the simpler the motion pattern representing the target within track, and the lower the corresponding maneuverability.
Figure BDA0003165054080000035
(5) And a track oscillation factor zeta.
The difference between the heading angles of the track sampling points is defined as delta h (k) hk+1-hkK is 1,2, …, N-1. Defining a measurement error threshold δ taking into account the actual radar measurement errore0.5 degrees, define the k-th course difference symbol J of the target trackh(k) Comprises the following steps:
Figure BDA0003165054080000041
Jh(k) is the kth element O (k) in the vector O of the heading difference symbols of the target track, the vector O can be described as O ═ 1, -1,0,1, …]In the form of (1). According to the heading difference symbol vector O, if there is one of the following two oscillation modes:
oscillation mode 1: o (k-1) + O (k) ≠ O (k)
Oscillation mode 2: o (k-1) + O (k +1) ≠ O (k +1), O (k) ≠ 0
Judging that a course oscillation exists in the flight path. And judging the oscillation frequency of the O according to the above criterion. If there is one or more than one oscillation, define the oscillation factor:
Figure BDA0003165054080000042
wherein, | Δ h (i) | is the absolute value of the change of the heading angle under the ith oscillation. The weight factors w (i) are defined as shown in Table 1:
TABLE 1 weight factor definition
Number of oscillations 1 2 3 4 Greater than or equal to 5
Weight factor 1 1.5 2 3 5
And (3) constructing a track motion characteristic vector by adopting the five types of track motion characteristic descriptors:
S＝[vmean,vstd,hstd,σ,ζ] (9)
as shown in fig. 2, examples of flying birds and drone tracking tracks observed in low altitude airspace near the seattle international airport in usa based on the bird detection casade radar system by accident corporation, canada are provided. And constructing a target track motion characteristic vector according to the above mode according to the corresponding information such as speed, direction and the like on each track point in the track.
And step two, constructing a target motion characteristic joint probability matrix.
The invention provides a rotor unmanned aerial vehicle and bird target classification and identification method based on radar track motion characteristics, which adopts a supervised learning algorithm. And (3) assuming that the elements of the track motion characteristic vectors are mutually independent, acquiring radar tracking tracks of different types of targets, and constructing the motion characteristic vectors of the target tracks by adopting the method in the step one. And then constructing a training database and a joint distribution probability matrix based on a large amount of motion feature vector sample sets.
A hierarchical table is built for each motion feature descriptor in equation (9), denoted vmeanThe descriptor is an example for explaining the descriptor hierarchical boundary numerical definition method. For v within a defined range of valuesmeanV of each class of object subset in the training sample setmeanThe corresponding probability value in the database is
Figure BDA0003165054080000043
Wherein
Figure BDA0003165054080000044
And the probability that the target class is the T-th class under the condition that the average speed value is v is shown, wherein T is the number of classes of the targets in the training sample set. In the embodiment of the present invention, the target categories are divided into two categories, i.e., a bird and a drone, i.e., T ═ 2. V is to bemeanDividing the value range into Q levels, and defining the lower bound and the upper bound of the value range corresponding to each level as
Figure BDA0003165054080000045
1,2, Q-1, T1, 2, T, numerically satisfying the following condition:
Figure BDA0003165054080000051
at the lower bound of the determined value range
Figure BDA0003165054080000052
Then, when the integral value satisfies the definition of the formula (10), the upper bound of the interval is found at this time
Figure BDA0003165054080000053
By counting the values of the feature elements in the training data set, the maximum value and the minimum value of the feature elements can be obtained, and the minimum value is taken as the lower bound of level 1, namely
Figure BDA0003165054080000054
The upper bound is then obtained according to equation (10)
Figure BDA0003165054080000055
Namely the upper bound of level 1, and so on, and then sequentially obtaining the 3 rd to the Q th boundary values according to the formula (10).
For different classes of targets T (T1, 2.. T.), a boundary value array composed of Q levels
Figure BDA0003165054080000056
There is a certain difference in the value, and the final boundary value is obtained by using a method of performing statistical average on T-type targets, as shown in the following formula:
Figure BDA0003165054080000057
rv(i) is the ith boundary value of the average velocity.
Other classes of feature descriptors can be calculated using similar methods as described in equations (10) and (11) above.
After a hierarchical table of all track motion feature descriptors is obtained, two feature elements { S (i), S (j) } in a motion feature vector S can be selected randomly to construct a joint probability matrix
Figure BDA0003165054080000058
The dimension of the joint probability distribution matrix is 5 x 5, according to the definition of S. The two feature elements selected are different, i ≠ j, where i, j denote the ith and jth feature elements, respectively, in the motion feature vector S. For the training sample set with the target class of t and the corresponding joint probability distribution matrix
Figure BDA0003165054080000059
The following method can be used for calculation:
will be provided with
Figure BDA00031650540800000510
The array is initialized to be an array with dimension of 5 multiplied by 5 and all units of 0, and each flight path in the training sample set with the traversal type of t and the corresponding feature vector are traversed. For the values x corresponding to descriptors S (i) and S (j)1And x2Obtaining the corresponding levels m and n in the hierarchical table where the hierarchical table is located, and updating the matrix
Figure BDA00031650540800000511
Middle element
Figure BDA00031650540800000512
The following were used:
Figure BDA00031650540800000513
after traversing all the t-th class target training samples, obtaining a matrix
Figure BDA00031650540800000514
The matrix elements are normalized as follows:
Figure BDA00031650540800000515
for normalized matrix
Figure BDA00031650540800000516
Value of matrix element
Figure BDA00031650540800000517
It is shown that for the t-th class object, the probability that the track feature element belongs to the level m under the condition that s (i) belongs to the level n is P (j), and thus the conditional probability P can be further used as Pij(n | m) is described. The sequence of selecting the characteristic elements is different, and the obtained joint probability matrix is also different, namely the matrix
Figure BDA00031650540800000518
And
Figure BDA00031650540800000519
representing two different matrices. Thus, for the feature space where the feature vector is located, a feature vector can be constructed
Figure BDA00031650540800000520
A joint probability distribution matrix.
Tables 2 and 3 show five types of motion feature classification tables and joint probability distribution matrices of the average speed and heading standard deviation features based on experimental data of the embodiment, respectively.
TABLE 2 track motion characteristic grading Table
Figure BDA00031650540800000521
Figure BDA0003165054080000061
TABLE 3 example Joint probability matrix for mean speed-heading Standard deviation
Unmanned aerial vehicle/flying bird 1 2 3 4 5
1 0.25\0.18 0.27\0.35 0.43\0.23 0.02\0.14 0.03\0.1
2 0.19\0.11 0.24\0.31 0.41\0.26 0.12\0.2 0.02\0.12
3 0.21\0.18 0.33\0.41 0.32\0.28 0.07\0.08 0.07\0.05
4 0.15\0.21 0.37\0.35 0.38\0.31 0.06\0.08 0.04\0.05
5 0.12\0.23 0.35\0.42 0.29\0.27 0.15\0.06 0.09\0.02
As shown in tables 2 and 3, in the embodiment of the present invention, 5 levels are set for each feature element, table 2 shows a hierarchical table of 5 feature elements, two feature elements { average speed, standard deviation of course } are selected in table 3 to establish a joint probability matrix, and numerical values in the joint probability matrix are calculated for two types of targets, i.e., birds and unmanned aerial vehicles, respectively.
And step three, judging the target track type by utilizing the joint probability matrix.
Unknown target track for radar acquisition
Figure BDA00031650540800000612
Extracting motion characteristic vectors according to the method in the first step, obtaining the levels of all characteristic values according to the hierarchical table in the second step, searching corresponding probability values based on 20 joint probability matrixes constructed by various targets in the second step, and constructing probability vectors for the target types t
Figure BDA0003165054080000062
Wherein P isl tThe values are from the ith joint probability distribution matrix, l 1,2, …, 20; t1, 2.
According to Bayes' theorem, and presuming characteristic elements
Figure BDA0003165054080000063
The probability that the flight path belongs to the t-th class target is as follows:
Figure BDA0003165054080000064
suppose that
Figure BDA0003165054080000065
Independent of the category t, the main information of the category is determined to come from
Figure BDA0003165054080000066
Based on statistical independent assumptions, the joint probability model satisfies:
Figure BDA0003165054080000067
wherein the content of the first and second substances,
Figure BDA0003165054080000068
representing unknown target track obtained according to the l combined probability matrix of the t-type target
Figure BDA0003165054080000069
Probability of belonging to class t object, i.e. corresponding probability Pl t. Assuming that the objects of the classes occur with equal probability, i.e.
Figure BDA00031650540800000610
The category of the flight path is judged by the following method:
Figure BDA00031650540800000611
in this embodiment, the flight path observation data of 8, 9, and 10 months in 2016 and the flight path observation data of the experimental unmanned aerial vehicle are used for verification by the above identification method. The correct identification rates of the bird and drone targets are shown in table 4.
Table 4 identifying accuracy of flying bird/unmanned aerial vehicle
Flying bird/unmanned aerial vehicle Early morning 05:00-07:00 In the evening from 17:00 to 19:00
August 82/87 86/88
September 84/85 85/89
October 80/90 88/92
The embodiment shows that the method can effectively identify the flying birds and the unmanned aerial vehicle, and solves the problem of high identification difficulty of the two existing targets caused by high overlapping degree of radar scattering cross section values.

Claims (7)

1. An unmanned aerial vehicle and flying bird target classification method based on track motion characteristics is characterized by comprising the following steps:
step 1, extracting a motion characteristic vector for a target track acquired by a radar; the motion feature vector includes feature elements: average speed, speed standard deviation, course standard deviation, maneuvering factor and track oscillation factor;
step 2, extracting motion characteristic vectors to form a training sample set for tracks of different targets by adopting the step 1, and then constructing a joint probability matrix of motion characteristics of the different targets, wherein the joint probability matrix comprises the following steps: (1) firstly, the methodEstablishing a hierarchical table for each feature element; (2) secondly, two characteristic elements in the motion characteristic vector are selected randomly to construct a joint probability matrix of each type of target; when a joint probability matrix of a t-type target is constructed, initializing the joint probability matrix to be a 0 matrix, traversing motion characteristic vectors of each flight path of the t-type target in a training sample set, acquiring levels m and n corresponding to the numerical values of two selected characteristic elements according to a hierarchical table, increasing the value of the element (m, n) in the joint probability matrix by 1, and normalizing the element in the joint probability matrix after traversing is completed; the value of the element (m, n) in the obtained joint probability matrix represents that for the t-th class target, the probability that the selected first characteristic element belongs to the level m is p under the condition that the selected first characteristic element belongs to the level n; for each type of object to be constructed
Figure FDA0003165054070000011
A joint probability matrix;
and 3, judging the category of the unknown target track by using the obtained joint probability matrix, wherein the judgment comprises the following steps: extracting motion characteristic vectors from unknown target tracks acquired by the radar, acquiring the levels of all characteristic values according to a hierarchical table, searching the probability values of the unknown target tracks belonging to the targets from 20 joint probability matrixes for each type of targets, and finally selecting the type with the highest probability as the type of the unknown target tracks.
2. The method according to claim 1, wherein in step 1, said maneuver factor is represented by σ and is calculated as an average velocity vmeanStandard deviation h from coursestdThe ratio of (a) to (b).
3. The method according to claim 1, wherein in step 1, the track oscillation factor is represented as ζ and is calculated as follows: firstly, calculating course difference symbol vector O of target track, wherein the kth element O (k) in O is course difference symbol
Figure FDA0003165054070000012
Wherein, Δ h (k) is the course angle difference between the k +1 th track sampling point and the k-th track sampling point, and δeIs a measurement error threshold; then, judging the course oscillation frequency existing in the target track according to the vector O, and judging that one course oscillation exists if one of the following two oscillation modes exists;
oscillation mode 1: o (k-1) + O (k) ≠ O (k) 0 and O (k-1) ≠ O (k);
oscillation mode 2: o (k-1) + O (k +1) ≠ O (k +1), O (k) ≠ 0;
finally, calculating the oscillation factor
Figure FDA0003165054070000013
Wherein, | Δ h (i) | is the absolute value of the change of the course angle under the ith oscillation, and w (i) is a weight factor.
4. The method according to claim 3, wherein in step 1, the weight factors w (i) are set as:
when 1 course oscillation exists in the target track, w (i) is 1;
when 2 course oscillations exist in the target track, w (i) is 1.5;
when 3 course oscillations exist in the target track, w (i) is 2;
when 4 course oscillations exist in the target track, w (i) is 3;
when there are 5 or more course oscillations in the target track, w (i) ═ 5.
5. The method of claim 1, wherein in step 2, a hierarchical table is created for the feature elements, and the hierarchical boundary values are determined by:
at an average speed vmeanFor illustration, in the training sample set, let
Figure FDA0003165054070000021
Denotes the probability that the target class is class T under the average velocity value v, where T is 1,2The number of categories of the medium target; dividing the average speed into Q levels on a value range, and setting the lower bound and the upper bound of the value range corresponding to each level of the t-th class target as
Figure FDA0003165054070000022
Obtaining the upper and lower bound values by finding a definition that satisfies the following formula;
Figure FDA0003165054070000023
and finally, determining the grading boundary value of the average speed for the statistical average value of the boundary values of each grade of the T-class target to obtain a grading table of the average speed.
6. The method according to claim 1 or 5, wherein in step 2, two feature elements { S (i), S (j) } are selected to construct a joint probability matrix of the t-th class target
Figure FDA0003165054070000024
After traversing the training sample of the t-th class target, the obtained matrix is subjected to
Figure FDA0003165054070000025
The normalization method of each element is as follows:
Figure FDA0003165054070000026
wherein, the matrix
Figure FDA0003165054070000027
Dimension (d) is 5 × 5.
7. The method according to claim 1, wherein in step 3, the unknown target track is set as
Figure FDA00031650540700000218
Searching 20 joint probability matrixes of the t-th class target to obtain probability vectors
Figure FDA0003165054070000028
Let Pl tIs a probability value obtained from the ith joint probability distribution matrix, which value represents the flight path
Figure FDA0003165054070000029
Probability of belonging to class t object, l ═ 1,2, …, 20;
assuming that the motion characteristic elements are independent of each other, the track
Figure FDA00031650540700000210
Probability of belonging to class t object
Figure FDA00031650540700000211
Wherein, the probability is set
Figure FDA00031650540700000212
Independent of the category t, further we get:
Figure FDA00031650540700000213
given that the various types of objects occur with equal probability,
Figure FDA00031650540700000214
then track
Figure FDA00031650540700000215
The belonged category is judged by the following method:
Figure FDA00031650540700000216
t' denotes track
Figure FDA00031650540700000217
The category to which it belongs.
CN202110802116.7A 2021-07-15 2021-07-15 Unmanned aerial vehicle and flying bird target classification method based on track motion characteristics Active CN113537347B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110802116.7A CN113537347B (en) 2021-07-15 2021-07-15 Unmanned aerial vehicle and flying bird target classification method based on track motion characteristics

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110802116.7A CN113537347B (en) 2021-07-15 2021-07-15 Unmanned aerial vehicle and flying bird target classification method based on track motion characteristics

Publications (2)

Publication Number Publication Date
CN113537347A true CN113537347A (en) 2021-10-22
CN113537347B CN113537347B (en) 2023-07-18

Family

ID=78128220

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110802116.7A Active CN113537347B (en) 2021-07-15 2021-07-15 Unmanned aerial vehicle and flying bird target classification method based on track motion characteristics

Country Status (1)

Country Link
CN (1) CN113537347B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2015121473A (en) * 2013-12-24 2015-07-02 三菱電機株式会社 Target tracking device and target tracking method
CN106546975A (en) * 2016-10-14 2017-03-29 中国民航科学技术研究院 A kind of small-sized unmanned plane based on radar data and flying bird classifying identification method
US20180308371A1 (en) * 2017-04-19 2018-10-25 Beihang University Joint search method for uav multiobjective path planning in urban low altitude environment

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2015121473A (en) * 2013-12-24 2015-07-02 三菱電機株式会社 Target tracking device and target tracking method
CN106546975A (en) * 2016-10-14 2017-03-29 中国民航科学技术研究院 A kind of small-sized unmanned plane based on radar data and flying bird classifying identification method
US20180308371A1 (en) * 2017-04-19 2018-10-25 Beihang University Joint search method for uav multiobjective path planning in urban low altitude environment

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
陈小龙 等: "飞鸟与无人机目标雷达探测与识别技术进展与展望", 《雷达学报》 *

Also Published As

Publication number Publication date
CN113537347B (en) 2023-07-18

Similar Documents

Publication Publication Date Title
CN112132042B (en) SAR image target detection method based on contrast domain adaptation
CN108510467B (en) SAR image target identification method based on depth deformable convolution neural network
CN110133599B (en) Intelligent radar radiation source signal classification method based on long-time and short-time memory model
CN110018453B (en) Intelligent model identification method based on aircraft track characteristics
Tao et al. Scene context-driven vehicle detection in high-resolution aerial images
CN111401168B (en) Multilayer radar feature extraction and selection method for unmanned aerial vehicle
CN109753874A (en) A kind of low slow small classification of radar targets method based on machine learning
CN111142085B (en) External radiation source radar target classification and identification method based on track feature extraction
Zhang et al. SEFEPNet: Scale expansion and feature enhancement pyramid network for SAR aircraft detection with small sample dataset
CN115943439A (en) Multi-target vehicle detection and re-identification method based on radar vision fusion
CN113808174A (en) Radar small target tracking method based on full convolution network and Kalman filtering
CN109117776A (en) Aircraft and meteorological clutter classifying identification method based on track information
Kahraman et al. Road extraction techniques from remote sensing images: A review
CN113191185A (en) Method for classifying targets of unmanned aerial vehicle by radar detection through Dense2Net
CN113960587B (en) Millimeter wave radar multi-target tracking method based on category information feedback
CN111368653B (en) Low-altitude small target detection method based on R-D graph and deep neural network
CN113537347B (en) Unmanned aerial vehicle and flying bird target classification method based on track motion characteristics
Sévigny et al. Unmanned aircraft (ua) telemetry data for track modelling and classification
CN112327266B (en) Clutter point trace eliminating method based on multidimensional information fusion
Yang et al. Shape feature aided target detection method for micro-drone surveillance radar
CN116304966A (en) Track association method based on multi-source data fusion
Xiao et al. Feature-based classification for image segmentation in automotive radar based on statistical distribution analysis
CN114677419A (en) Radar Doppler signal low-slow small target detection method based on three-dimensional convolution network
CN109375205B (en) Method and device for recognizing scenes of multiple types of unmanned aerial vehicles based on dictionary learning
Han et al. Aircraft target detection in polsar image based on region segmentation and multi-feature decision

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