CN115061501A - Method and system for identifying coded data of air formation - Google Patents
Method and system for identifying coded data of air formation Download PDFInfo
- Publication number
- CN115061501A CN115061501A CN202210894545.6A CN202210894545A CN115061501A CN 115061501 A CN115061501 A CN 115061501A CN 202210894545 A CN202210894545 A CN 202210894545A CN 115061501 A CN115061501 A CN 115061501A
- Authority
- CN
- China
- Prior art keywords
- code
- formation
- coding
- index
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000015572 biosynthetic process Effects 0.000 title claims abstract description 212
- 238000000034 method Methods 0.000 title claims abstract description 43
- 239000013598 vector Substances 0.000 claims abstract description 47
- 230000000694 effects Effects 0.000 claims abstract description 38
- 239000011159 matrix material Substances 0.000 claims abstract description 30
- 238000012545 processing Methods 0.000 claims abstract description 12
- 238000012935 Averaging Methods 0.000 claims abstract description 11
- 210000002569 neuron Anatomy 0.000 claims description 46
- 238000013528 artificial neural network Methods 0.000 claims description 24
- 238000004422 calculation algorithm Methods 0.000 claims description 13
- 238000010276 construction Methods 0.000 claims description 11
- 238000007637 random forest analysis Methods 0.000 claims description 10
- 238000012706 support-vector machine Methods 0.000 claims description 10
- 230000001133 acceleration Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000005755 formation reaction Methods 0.000 description 164
- 238000012549 training Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 10
- 238000010801 machine learning Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 210000004027 cell Anatomy 0.000 description 2
- 230000009194 climbing Effects 0.000 description 2
- 230000009918 complex formation Effects 0.000 description 2
- 230000007123 defense Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
- G05D1/10—Simultaneous control of position or course in three dimensions
- G05D1/101—Simultaneous control of position or course in three dimensions specially adapted for aircraft
- G05D1/104—Simultaneous control of position or course in three dimensions specially adapted for aircraft involving a plurality of aircrafts, e.g. formation flying
Landscapes
- Engineering & Computer Science (AREA)
- Aviation & Aerospace Engineering (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Image Analysis (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides a method and a system for identifying coded data of an aerial formation form, belonging to the field of aerial formation form identification, wherein the method comprises the following steps: dividing the formation activity area into grids for binary coding according to the formation motion direction and the formation activity area, and constructing a local area code; establishing a feature matrix or a feature vector with the same dimension as the local code, filling corresponding feature values in the feature matrix or the feature vector at the same position of a target of the local code, and constructing a multilayer feature subcode; simplifying the local area code by adopting a position coding or index coding method; carrying out weighted averaging or coding combination processing on the multilayer characteristic subcodes; the local area code and the multilayer characteristic subcodes are connected in series to form an aerial formation situation coding vector; and inputting the air formation situation coding vector into a formation recognition model to obtain an air formation form. Compared with the traditional template-based method, the method has higher intelligent degree.
Description
Technical Field
The invention belongs to the field of identification of aerial formation formations, and particularly relates to a method and a system for identifying coded data of aerial formation formations.
Background
The air formation is the main battle style of modern air force, and can exert the power of the cluster and form the advantages of the body system. The identification of the formation in the air is the discrimination of the formation type formed by the objects in the air. The identification of the formation shape of the air formation is an important mode for judging the air behavior purpose, the task type and the like of the other party, and is important content for predicting the air defense early warning situation. However, due to the wide variety and high real-time performance of formation, it is difficult for people to identify the formation timely, accurately and continuously through experience.
Traditional computer-aided formation identification methods are template-based methods. In the template-based method, the method of constructing the template based on Hough transformation is common. The method is used for constructing a template for identifying the formation forms of various common linear formation forms by utilizing the point-line duality of Hough transformation. Under the noise condition, the formation characteristics are obtained based on line intersection clustering, then a template adaptive to the line intersection clustering is established, and the formation identification is completed based on template matching. The identification method based on Hough transformation and a clustering method can better realize the abstraction of formation in a noise environment and solve the uncertainty of formation, but the clustering algorithm has difficulties in the aspects of determining the number of clusters, initializing the centers of the clusters and the like, and solves the problem that adverse effects can influence the extraction of the formation characteristics. Meanwhile, as the template construction method has more human subjective factors, the template construction process is complex when complex formation identification is carried out, and more factors need to be considered.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a method and a system for identifying coded data of an aerial formation form, and aims to solve the problem that the existing template-based formation identification method has a plurality of human factors and the process of establishing a template is complex when complex formation identification is faced.
In order to achieve the above object, in one aspect, the present invention provides a method for identifying coded data of an air formation, including the following steps:
dividing the formation activity area into grids for binary coding with uniform size according to the formation motion direction and the formation activity area, and constructing a local area code; wherein, the local area code uses 1 to represent the existing target, and 0 to represent the non-existing target;
for each aerial target motion characteristic, establishing a characteristic matrix or a characteristic vector with the same dimension as the local code, filling corresponding characteristic values in the characteristic matrix or the characteristic vector at the same positions of the local code where the targets exist, and constructing a multilayer characteristic subcode;
simplifying the local area code by adopting a position coding or index coding method; carrying out weighting averaging or coding combination processing on the multilayer characteristic subcodes;
the simplified local codes and multilayer characteristic subcodes which are weighted, averaged or combined are connected in series to form aerial formation situation coding vectors;
inputting the air formation situation coding vector into a formation recognition model to obtain an air formation form; and the formation recognition model is a trained classifier.
Further preferably, the aerial target motion characteristics include target height, speed, acceleration, and climbing rate;
further preferably, the manner of position coding the local code includes absolute position coding or relative position coding;
the absolute position coding takes the upper left corner of the formation activity area as a starting point, calculates the absolute distance between the code element position corresponding to each target and the starting point according to the zigzag sequence, and takes the absolute distance as a code element value;
the absolute position is encoded as:
R=[r k ]，k=1，2，…，p
r k =（i k -1）×n+j k
the relative position coding takes the upper left corner of the formation activity area as a starting point, calculates the relative distance between the code element position corresponding to each target and the code element position corresponding to the previous target according to the zigzag sequence, and takes the relative distance as a code element value;
the relative position is encoded as:
R=[r k ]，k=1，2，…，p
wherein,i k ，j k respectively expressed in local area codeskThe number of rows and columns of elements corresponding to each object.
Further preferably, the manner of performing index coding on the local code comprises row-column index coding or grid index coding;
the row-column index coding mode is as follows: in the queuing activity area, code elements corresponding to each target are represented by row and column values of a local code matrix, and the row and column values of the code elements corresponding to each target are connected in series to form row and column index codes;
the row column index is encoded as:
M=[t k ]=[（i k ，j k ）]，k=1，2，…，p
the encoding mode of the grid index is as follows: dividing the formation activity area into grids, expressing the code elements corresponding to the targets by using a two-level grid index, and connecting the grid index values of the code elements corresponding to the targets in series to form a grid index code;
the grid index codes are as follows:
M=[t k ]=[（i k ，j k ）]，k=1，2，…，p
wherein,qthe number of the first-level grids;t k 1 andt k 2 respectively representing the first-level lattice and the second-level lattice codes;
and
respectively representing upward rounding and remainder calculation;i k ，j k respectively expressed in local area codeskThe number of rows and columns of elements corresponding to each target.
Further preferably, the input of the classifier is air formation situation code, and an air formation form is output; the classifier is a support vector machine, a BP neural network or a random decision forest;
the BP neural network adopts a three-layer structure, when the air formation situation code is based on position code, the number of neurons in an input layer is the same as the length of an air formation situation code vector, the number of neurons in a hidden layer is set to be 2 times of the number of neurons in the input layer, and the number of neurons in an output layer is the number of aerial formation forms; when the aerial formation situation code is based on index code, the number of neurons in an input layer is 2 times of the length of an aerial formation situation code vector, the number of neurons in a hidden layer is 2 times of the number of neurons in the input layer, and the number of neurons in an output layer is the number of aerial formation forms;
and optimizing the BP neural network parameters by adopting a GA algorithm.
In another aspect, the present invention provides a system for identifying encoded data of an aerial formation, including: the local area code construction module is used for dividing the formation activity area into grids for binary coding with uniform size according to the formation motion direction and the formation activity area, and constructing a local area code; wherein, the local area code uses 1 to represent the existing target, and 0 to represent the non-existing target;
the multilayer characteristic subcode construction module is used for establishing a characteristic matrix or a characteristic vector with the same dimension as the local code for each aerial target motion characteristic, filling characteristic values of corresponding targets in the characteristic matrix or the characteristic vector at the same positions of the targets where the local code exists, and constructing multilayer characteristic subcodes;
the encoding module is used for simplifying the local code by adopting a position encoding or index encoding method;
the characteristic subcode processing module is used for carrying out weighted averaging or coding combination processing on the multilayer characteristic subcodes;
the concatenation module is used for concatenating the simplified local codes and multilayer characteristic subcodes subjected to weighted averaging or code combination to form aerial formation situation code vectors;
and the classifier is used for inputting the air formation situation and outputting the air formation form.
Further preferably, the airborne target motion characteristics include target height, velocity, acceleration and climb rate.
Further preferably, the encoding module is a position encoding module, and the position encoding module is an absolute position encoding module or a relative position encoding module;
the absolute position coding module is used for calculating the absolute distance between the code element position corresponding to each target and the starting point according to the zigzag sequence by taking the upper left corner of the formation activity area as the starting point, and coding by taking the absolute distance as a code element value to obtain an absolute position code;
the absolute position is encoded as:
R=[r k ]，k=1，2，…，p
r k =（i k -1）×n+j k
the relative position coding module is used for calculating the relative distance between the code element position corresponding to each target and the code element position corresponding to the previous target according to the zigzag sequence by taking the upper left corner of the formation activity area as a starting point, and coding by taking the relative distance as a code element value to obtain a relative position code;
the relative position is encoded as:
R=[r k ]，k=1，2，…，p
wherein,i k ，j k are respectively shown in bureauIn the domain codekThe number of rows and columns of elements corresponding to each target.
Further preferably, the encoding module is an index encoding module, and the index encoding module is a row-column index encoding module or a grid index encoding module;
the row-column index coding module is used for representing code elements corresponding to all targets by row-column values of a local code matrix in a queuing activity area and connecting the code element row-column values corresponding to all targets in series to form row-column index codes;
the row column index is coded as:
M=[t k ]=[（i k ，j k ）]，k=1，2，…，p
the grid index coding module is used for dividing the formation activity area into grids, expressing the code elements corresponding to the targets by using a two-level grid index, and connecting grid index values of the code elements corresponding to the targets in series to form a grid index code;
the grid index codes are:
M=[t k ]=[（i k ，j k ）]，k=1，2，…，p
wherein,qthe number of the first-level grids;t k 1 andt k 2 respectively representing the first-level grid and the second-level grid codes;
and
respectively representing upward rounding and remainder calculation;i k ，j k respectively expressed in local area codeskThe number of rows and columns of elements corresponding to each target.
Further preferably, the input of the classifier is air formation situation code, and an air formation form is output; the classifier is a support vector machine, a BP neural network or a random decision forest;
the BP neural network adopts a three-layer structure, when the air formation situation code is based on position code, the number of neurons in an input layer is the same as the length of an air formation situation code vector, the number of neurons in a hidden layer is set to be 2 times of the number of neurons in the input layer, and the number of neurons in an output layer is an air formation form; when the aerial formation situation code is based on index code, the number of neurons in an input layer is 2 times of the length of an aerial formation situation code vector, the number of neurons in a hidden layer is 2 times of the number of neurons in the input layer, and the number of neurons in an output layer is an aerial formation form;
and optimizing the BP neural network parameters by adopting a GA algorithm.
Generally, compared with the prior art, the above technical solution conceived by the present invention has the following beneficial effects:
the invention provides a coded data identification method and a coded data identification system for an aerial formation form. Firstly, providing a coding method for acquiring the situation of air formation, dividing a formation activity area into grids according to a formation motion direction and a formation activity area, carrying out binary coding with uniform size, and constructing a local area code; for each aerial target motion characteristic, establishing a characteristic matrix or a characteristic vector with the same dimension as the local code, filling corresponding characteristic values in the characteristic matrix or the characteristic vector at the same position of the local code where the target exists, and constructing a multilayer characteristic subcode; simplifying the local area code by adopting a position coding or index coding method; carrying out weighted averaging or coding combination processing on the multilayer characteristic subcodes; the simplified local codes and multilayer characteristic subcodes which are weighted, averaged or combined are connected in series to form aerial formation situation coding vectors; successfully converting the position, the motion characteristic and other attribute characteristics of the air formation into coding information, and training by applying a machine learning technology to obtain a classifier based on the air formation situation code, so that the classifier can intelligently identify the air formation situation code and output the air formation form. The algorithm provided by the invention realizes the intelligent identification of the formation from end to end, and the formation type can be directly obtained on the basis of unified situation coding. Compared with the traditional template-based method, the intelligent degree is higher.
In the process of acquiring the air formation situation codes, in order to avoid the complexity of matrix operation processing, position codes or index codes are adopted to simplify the local codes, wherein the method for carrying out the position codes on the local codes is absolute position codes or relative position codes, and the method for carrying out the index codes on the local codes is row-column index codes or grid index codes; the efficiency of formation identification is improved, and different coding modes can be selected for coding according to actual coded data.
The classifier can select various support vector machines, BP neural networks or random decision forests and the like, different classifiers can be selected for training according to the number of the formation flying frames when real-time aerial formation is identified, and different formation identification models can be obtained by combining different coding modes.
Drawings
FIG. 1 is a diagram of a coding structure provided by an embodiment of the present invention;
fig. 2(a) is an air formation diagram provided by an embodiment of the present invention;
fig. 2(b) is a corresponding enqueue local area code of fig. 2(a) according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a high-level feature subcode corresponding to FIG. 2(b) according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a neural network-based recognition model provided by an embodiment of the present invention;
FIG. 5(a) is a schematic diagram of a wedge formation provided by an embodiment of the present invention;
FIG. 5(b) is a schematic diagram of a column formation provided by an embodiment of the present invention;
FIG. 5(c) is a schematic diagram of a echelon formation provided by an embodiment of the present invention;
fig. 5(d) is a schematic diagram of the formation of the cross-platoon according to the embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The overall technical scheme of the invention is as follows: bringing the air formation into a unified situation coding structure; based on the situation code of the air formation, applying a machine learning technology and designing an intelligent recognition algorithm frame based on machine learning; generating air formation data through simulation, establishing a multi-type formation data set, and training a model of the multi-type formation data set;
in one aspect, the invention provides a method for identifying coded data of an aerial formation form, which comprises the following steps:
the method comprises the following steps: and (3) encoding the air formation situation:
the coding structure shown in fig. 1:
the air formation situation code comprises a local area code and a characteristic subcode; in order to unify the input of subsequent situation prediction tasks, realize standardization and modular model construction, and design a code into a two-layer structure;
the local area code is the description of formation and the activity area thereof and can be represented in a binary matrix or vector form; the characteristic subcodes are codes of attribute characteristics of targets in the formation, and can also be expressed in a matrix or vector form, have the same dimension as the local area codes, and have corresponding elements; the situation code can uniquely describe and calibrate a target or formation and the movement thereof;
the mathematical expression of situation encoding is:R·F(ii) a Wherein,Ris a local area code;Fis a feature subcode; (. -) represents the operation of the local code and the corresponding element of the characteristic code matrix;
local area code:
dividing grids for the formation active area and carrying out binary coding to form a binary matrix with uniform size, namely a local area code; in the local area code, 1 represents a target, 0 represents a target does not exist, as shown in fig. 2(a), a queuing spatial situation diagram is shown, and fig. 2(b) is a corresponding queuing local area code; the size unification is the key of local area code coding, and aims to arrange the formation of different areas and densities in a matrix with unified dimension to realize the unique expression of the codes;
the local area code coding solves the problem that the dimensionality of an area matrix obtained by directly coding an original area is not uniform, and is convenient for the application of intelligent algorithms such as machine learning and the like;
the characteristic subcode:
the codes reflect the target attribute characteristics in the formation and correspond to the local area codes;
the aerial target motion characteristics comprise attributes such as height, speed, acceleration and climbing rate; in the study of formation type identification, height is the most important feature; assuming that the formation targets are at the same height level and all are h meters, fig. 3 is a height feature subcode corresponding to fig. 2 (b);
in practical application, the feature subcodes are not necessarily simple one-layer structures, but often consist of multiple layers of feature subcodes, such as feature subcodes of height, RCS, communication frequency, and the like; the characteristic subcodes can be processed in various ways, such as weighted averaging, coding combination and the like, and specific situations of different scenes need to be considered and targeted processing is performed;
local area code encoding:
the position-based coding and the index-based coding are four coding methods in two types;
(1) location-based coding:
the local area coding mode based on the position can be divided into absolute position coding and relative position coding, which are expressed in a vector form;
(1.1) absolute position coding:
absolute position coding takes the upper left corner of a coding airspace as a starting point, calculates the absolute distance (the unit is a unit cell) between the code element position corresponding to each target and the starting point according to the zigzag sequence, and takes the distance as a code element value; if the area matrixA m×n If p targets exist, the absolute position code can be expressed as:
R=[r k ]，k=1，2，…，p
r k =（i k -1）×n+j k
wherein,i k ，j k respectively representing area matricesA m×n To middlekThe number of element rows and columns corresponding to each target;
(1.2) relative position coding:
relative position coding takes the upper left corner of the airspace as a starting point, and according to a zigzag sequence, the relative distance (the unit is a unit cell) between the code element position corresponding to each target and the code element position corresponding to the previous target is obtained, and the distance is taken as a code element value;
for the same local area code matrix, the relative position code can be expressed as:
R=[r k ]，k=1，2，…，p
wherein,i k ，j k has the same meaning as absolute position coding;
(2) index-based encoding:
the local area code coding mode based on the index can be divided into row-column index coding and grid index coding which are expressed in a vector form;
(2.1) row-column index coding:
in the queuing activity area, the code elements corresponding to the targets are represented by row and column values of a local code matrix, and the row and column values of the code elements corresponding to the targets are connected in series to form a vector, wherein the vector is row and column index coding;
the row column index encoding can be expressed as:
M=[t k ]=[（i k ，j k ）]，k=1，2，…，p
wherein,i k ，j k meaning the same as absolute position coding;
(2.2) encoding the grid index:
dividing the formation airspace into grids, expressing code elements corresponding to all targets by using a two-level grid index, and connecting grid index values of the code elements corresponding to all targets in series to form a vector, wherein the vector is a grid index code;
if the first level lattice number isqThen the grid index code can be expressed as:
M=[t k ]=[（t k 1 ，t k 2 ）]，k=1，2，…，p
wherein,t k 1 andt k 2 respectively representing the first-level lattice and the second-level lattice codes;i k ，j k meaning the same as absolute position coding;
and
respectively representing upward rounding and remainder calculation;
after the air defense early warning system acquires the position, the motion characteristics and other attribute data of an air formation, firstly, determining the motion direction of the formation, representing the area where the formation is located as a binary matrix, and selecting any one of four codes, namely a relative position code, an absolute position code, a row and column index code and a grid index code for coding to form a local area code; then, selecting the formation characteristics and constructing characteristic subcodes; finally, the local code is connected with the characteristic subcode in series, and the formed vector is the aerial formation situation coding vector;
step two: constructing a formation identification model:
designing a formation recognition model based on machine learning on the basis of encoding the situation of the formation in the air, and realizing the recognition of the formation shape of the formation through the training of the formation recognition model;
the formation recognition model is constructed by a classifier algorithm of Machine learning, a typical classifier includes a Support Vector Machine (SVM), a Back Propagation neural network (BP), a Random Decision Forest (RDF), and the like, and one of the two types of construction recognition models can be selected optionally, and fig. 4 is a schematic diagram of the formation recognition model constructed by the BP neural network;
when different classifier algorithms are selected for the formation recognition model, the corresponding settings of model parameters are different; when the SVM is selected as the formation recognition model, the selection of a kernel function and the parameter determination of the SVM are key, and the kernel function and the parameter are selected in a cross validation mode;
when the formation recognition model adopts a BP neural network, the BP neural network adopts a three-layer structure, when the aerial formation situation code is based on the position code, the number of neurons in an input layer is the same as the vector length of the aerial formation situation code, the number of neurons in a hidden layer is set to be 2 times of the number of the neurons in the input layer, and the number of the neurons in an output layer is the number of aerial formation forms; when the air formation situation code is based on index coding, the number of neurons of an input layer is 2 times of the length of an air formation situation code vector, and the number of neurons of a hidden layer and an output layer is consistent with the air formation situation code based on position coding; in order to avoid overfitting of the BP neural network, a Genetic Algorithm (GA) is adopted to optimize parameters of the BP neural network; the GA algorithm population scale is set to be 20, individuals in the population are all weights and thresholds of a BP neural network, the evolution times are 50 times, the cross probability is 0.4, the variation probability is 0.1, and an individual fitness function is the sum of the absolute values of the prediction errors of the training data; when the model selects RDF, the number of the base estimators of the RDF method is selected to be 20;
the output of the formation identification model is the type of the formation shape in the air;
step three: training a formation recognition model:
(3.1) data set construction:
constructing a data set with a proper scale is the basis of model training; the formation of the air formation can be divided into wedge-shaped formation, longitudinal formation, echelon and transverse formation according to forms, as shown in fig. 5(a) - (d); therefore, the formation type is set to 4 types; the number of airplane frames forming the formation is generally not fixed, but 3 to 10 are common, so the number of airplane frames forming the formation is set to be 3 to 10; respectively simulating each frame formation and constructing corresponding data sets, wherein each data set comprises 2000 samples; dividing a data set into a training set and a testing set, and segmenting the training set and the testing set according to a proportion of 10: 3;
generating formation data by taking a long machine as a base point according to the position layout specified by the combat ordinance and the formation motion direction; considering the influence of observation noise and motion noise, representing the uncertainty caused by the influence of the noise by using probability distribution; according to statistical analysis, the motion noise follows normal distribution with a mean value of 0m and a variance of 4m at the transverse interval of formation, follows normal distribution with a mean value of 0m and a variance of 6m at the longitudinal distance of formation, and follows normal distribution with a mean value of 0m and a variance of 2 m; sampling by adopting an accepting and rejecting method according to the description of the formation form and the probability distribution to generate simulation formation form data; carrying out situation coding on the generated data, and using the situation coding for training and identification;
(3.2) model parameter setting:
setting parameters according to a classifier selected during model design, and turning to the step two in detail;
after training is finished, air formation identification models corresponding to different air formation airplane frame numbers, classifiers and local area code coding methods can be obtained;
step four: aerial formation identification:
in another aspect, the present invention provides a system for identifying encoded data of an aerial formation, including: the local area code construction module is used for dividing the formation activity area into grids for binary coding with uniform size according to the formation motion direction and the formation activity area, and constructing a local area code; wherein, the local area code uses 1 to represent the existing target, and 0 to represent the non-existing target;
the multilayer characteristic subcode construction module is used for establishing a characteristic matrix or a characteristic vector with the same dimension as the local code for each aerial target motion characteristic, filling characteristic values of corresponding targets in the characteristic matrix or the characteristic vector at the same positions of the targets where the local code exists, and constructing multilayer characteristic subcodes;
the encoding module is used for simplifying the local code by adopting a position encoding or index encoding method;
the characteristic subcode processing module is used for carrying out weighted averaging or coding combination processing on the multilayer characteristic subcodes;
the concatenation module is used for concatenating the simplified local codes and multilayer characteristic subcodes subjected to weighted averaging or code combination to form aerial formation situation code vectors;
and the classifier is used for inputting the air formation situation and outputting the air formation form.
Further preferably, the airborne target motion characteristics include target height, velocity, acceleration and climb rate.
Further preferably, the encoding module is a position encoding module, and the position encoding module is an absolute position encoding module or a relative position encoding module;
the absolute position coding module is used for calculating the absolute distance between the code element position corresponding to each target and the starting point according to the zigzag sequence by taking the upper left corner of the formation activity area as the starting point, and coding by taking the absolute distance as a code element value to obtain an absolute position code;
the absolute position is encoded as:
R=[r k ]，k=1，2，…，p
r k =（i k -1）×n+j k
the relative position coding module is used for calculating the relative distance between the code element position corresponding to each target and the code element position corresponding to the previous target according to the zigzag sequence by taking the upper left corner of the formation activity area as a starting point, and coding by taking the relative distance as a code element value to obtain a relative position code;
the relative position is encoded as:
R=[r k ]，k=1，2，…，p
wherein,i k ，j k respectively expressed in local area codeskThe number of rows and columns of elements corresponding to each target.
Further preferably, the encoding module is an index encoding module, and the index encoding module is a row-column index encoding module or a grid index encoding module;
the row-column index coding module is used for representing code elements corresponding to all targets by row-column values of a local code matrix in a queuing activity area and connecting the code element row-column values corresponding to all targets in series to form row-column index codes;
the row column index is coded as:
M=[t k ]=[（i k ，j k ）]，k=1，2，…，p
the grid index coding module is used for dividing the formation activity area into grids, expressing the code elements corresponding to the targets by using a two-level grid index, and connecting grid index values of the code elements corresponding to the targets in series to form a grid index code;
the grid index codes are:
M=[t k ]=[（i k ，j k ）]，k=1，2，…，p
wherein,qthe number of the first-level grids;t k 1 andt k 2 respectively representing the first-level lattice and the second-level lattice codes;
and
respectively representing upward rounding and remainder calculation;i k ，j k respectively expressed in local area codeskThe number of rows and columns of elements corresponding to each target.
Further preferably, the input of the classifier is air formation situation code, and an air formation form is output; the classifier is a support vector machine, a BP neural network or a random decision forest;
the BP neural network adopts a three-layer structure, when the air formation situation code is based on position code, the number of neurons in an input layer is the same as the length of an air formation situation code vector, the number of neurons in a hidden layer is set to be 2 times of the number of neurons in the input layer, and the number of neurons in an output layer is an air formation form; when the air formation situation code is based on index code, the number of neurons in an input layer is 2 times of the length of the vector of the air formation situation code, the number of neurons in a hidden layer is set to be 2 times of the number of neurons in the input layer, and the number of neurons in an output layer is the air formation form;
and optimizing the BP neural network parameters by adopting a GA algorithm.
The model can realize the end-to-end recognition effect; when real-time aerial formation is identified, firstly, selecting a corresponding trained model according to the number of the formation flying frames, the type of a classifier and a coding mode; and then coding the formation data according to the selected coding mode, and inputting the obtained coding vector into the selected model. The output of the model is the formation.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. A coded data identification method of an air formation is characterized by comprising the following steps:
dividing the formation activity area into grids for binary coding with uniform size according to the formation motion direction and the formation activity area, and constructing a local area code; wherein, the local area code uses 1 to represent the existing target, and 0 to represent the non-existing target;
for each aerial target motion characteristic, establishing a characteristic matrix or a characteristic vector with the same dimension as the local code, filling characteristic values of corresponding targets in the characteristic matrix or the characteristic vector at the same positions of the targets existing in the local code, and constructing a multilayer characteristic subcode;
simplifying the local code by adopting a position coding or index coding method; carrying out weighted averaging or coding combination processing on the multilayer characteristic subcodes;
the simplified local codes and multilayer characteristic subcodes which are weighted, averaged or combined are connected in series to form aerial formation situation coding vectors;
inputting the air formation situation coding vector into a formation recognition model to obtain an air formation form; and the formation recognition model is a trained classifier.
2. The coded data recognition method of claim 1, wherein the airborne target motion characteristics include target altitude, velocity, acceleration and climb rate.
3. A coded data recognition method according to claim 1 or 2, wherein the local code is position-coded in absolute position coding or relative position coding;
the absolute position coding takes the upper left corner of the formation activity area as a starting point, calculates the absolute distance between the code element position corresponding to each target and the starting point according to the zigzag sequence, and codes by taking the absolute distance as a code element value;
the absolute position is encoded as:
R=[r k ]，k=1，2，…，p
r k =（i k -1）×n+j k
the relative position coding takes the upper left corner of the formation activity area as a starting point, calculates the relative distance between the code element position corresponding to each target and the code element position corresponding to the previous target according to the zigzag sequence, and codes by taking the relative distance as a code element value to obtain the relative position coding;
the relative position is encoded as:
R=[r k ]，k=1，2，…，p
wherein,i k ，j k respectively expressed in local area codeskThe number of rows and columns of elements corresponding to each target.
4. The coded data identification method according to claim 1 or 2, characterized in that the local code is index coded by row-column index coding or grid index coding;
the row-column index coding mode is as follows: in the queuing activity area, code elements corresponding to each target are represented by row and column values of a local code matrix, and the row and column values of the code elements corresponding to each target are connected in series to form row and column index codes;
the row-column index is encoded as:
M=[t k ]=[（i k ，j k ）]，k=1，2，…，p
the encoding mode of the grid index is as follows: dividing the formation activity area into grids, expressing the code elements corresponding to the targets by using a two-level grid index, and connecting the grid index values of the code elements corresponding to the targets in series to form a grid index code;
the grid index codes are as follows:
M=[t k ]=[（i k ，j k ）]，k=1，2，…，p
wherein,qthe number of the first-level grids;t k 1 andt k 2 respectively representing the first-level lattice and the second-level lattice codes;
and
respectively representing upward rounding and remainder calculation;i k ，j k respectively expressed in local area codeskThe number of rows and columns of elements corresponding to each object.
5. The coded data identification method according to claim 1, characterized in that the input of the classifier is air formation situation code, and air formation is output; the classifier is a support vector machine, a BP neural network or a random decision forest;
the BP neural network adopts a three-layer structure, when the air formation situation code is based on position code, the number of neurons in an input layer is the same as the length of a vector of the air formation situation code, the number of neurons in a hidden layer is set to be 2 times of the number of neurons in the input layer, and the number of neurons in an output layer is the number of aerial formation forms; when the aerial formation situation code is based on index code, the number of neurons in an input layer is 2 times of the length of an aerial formation situation code vector, the number of neurons in a hidden layer is 2 times of the number of neurons in the input layer, and the number of neurons in an output layer is the number of aerial formation forms;
and optimizing the BP neural network parameters by adopting a GA algorithm.
6. An identification system for coded data of an air formation, comprising:
the local area code construction module is used for dividing the formation activity area into grids for binary coding with uniform size according to the formation motion direction and the formation activity area, and constructing a local area code; wherein, the local area code uses 1 to represent the existing target, and 0 to represent the non-existing target;
the multilayer characteristic subcode construction module is used for establishing a characteristic matrix or a characteristic vector with the same dimension as the local code for each aerial target motion characteristic, filling characteristic values of corresponding targets in the characteristic matrix or the characteristic vector at the same positions of the targets where the local code exists, and constructing multilayer characteristic subcodes;
the encoding module is used for simplifying the local code by adopting a position encoding or index encoding method;
the characteristic subcode processing module is used for carrying out weighted averaging or coding combination processing on the multilayer characteristic subcodes;
the concatenation module is used for concatenating the simplified local codes and multilayer characteristic subcodes subjected to weighted averaging or code combination to form aerial formation situation code vectors;
and the classifier is used for inputting the air formation situation and outputting the air formation form.
7. The coded data recognition system of claim 6, wherein the airborne target motion characteristics include target altitude, velocity, acceleration, and rate of climb.
8. A coded data recognition system according to claim 6 or 7, wherein the coding module is a position coding module, the position coding module being either an absolute position coding module or a relative position coding module;
the absolute position coding module is used for calculating the absolute distance between the code element position corresponding to each target and the starting point according to the zigzag sequence by taking the upper left corner of the formation activity area as the starting point, and coding by taking the absolute distance as a code element value to obtain an absolute position code;
the absolute position code is:
R=[r k ]，k=1，2，…，p
r k =（i k -1）×n+j k
the relative position coding module is used for calculating the relative distance between the code element position corresponding to each target and the code element position corresponding to the previous target according to the zigzag sequence by taking the upper left corner of the formation activity area as a starting point, and coding by taking the relative distance as a code element value to obtain a relative position code;
the relative position codes are:
R=[r k ]，k=1，2，…，p
wherein,i k ，j k respectively expressed in local area codeskThe number of rows and columns of elements corresponding to each target.
9. The coded data identification system of claim 6 or 7, wherein the coding module is an index coding module, and the index coding module is a row-column index coding module or a grid index coding module;
the row-column index coding module is used for representing code elements corresponding to all targets by row-column values of a local code matrix in a queuing activity area and connecting the code element row-column values corresponding to all targets in series to form row-column index codes;
the row-column index is encoded as:
M=[t k ]=[（i k ，j k ）]，k=1，2，…，p
the grid index coding module is used for dividing the formation activity area into grids, expressing the code elements corresponding to the targets by using a two-level grid index, and connecting grid index values of the code elements corresponding to the targets in series to form a grid index code;
the grid index codes are as follows:
M=[t k ]=[（i k ，j k ）]，k=1，2，…，p
wherein,qthe number of the first-level grids;t k 1 andt k 2 respectively representing the first-level lattice and the second-level lattice codes;
and
respectively representing upward rounding and remainder calculation;i k ，j k respectively expressed in local area codeskThe number of rows and columns of elements corresponding to each target.
10. The coded data recognition system of claim 6, wherein the input to the classifier is an air formation situation code, and the output is an air formation; the classifier is a support vector machine, a BP neural network or a random decision forest;
the BP neural network adopts a three-layer structure, when the air formation situation code is based on position code, the number of neurons in an input layer is the same as the length of an air formation situation code vector, the number of neurons in a hidden layer is set to be 2 times of the number of neurons in the input layer, and the number of neurons in an output layer is the number of aerial formation forms; when the aerial formation situation code is based on index code, the number of neurons in an input layer is 2 times of the length of an aerial formation situation code vector, the number of neurons in a hidden layer is 2 times of the number of neurons in the input layer, and the number of neurons in an output layer is the number of aerial formation forms;
and optimizing the BP neural network parameters by adopting a GA algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210894545.6A CN115061501B (en) | 2022-07-28 | 2022-07-28 | Method and system for identifying coded data of air formation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210894545.6A CN115061501B (en) | 2022-07-28 | 2022-07-28 | Method and system for identifying coded data of air formation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115061501A true CN115061501A (en) | 2022-09-16 |
| CN115061501B CN115061501B (en) | 2022-11-22 |
Family
ID=83207183
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210894545.6A Active CN115061501B (en) | 2022-07-28 | 2022-07-28 | Method and system for identifying coded data of air formation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115061501B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160272340A1 (en) * | 2014-12-24 | 2016-09-22 | Environmental Tectonics Corporation | Aircraft-vision systems and methods for maintaining situational awareness and spatial orientation |
| CN110390337A (en) * | 2019-06-27 | 2019-10-29 | 清华大学 | A kind of naval vessel individual discrimination method |
| CN111398903A (en) * | 2020-04-03 | 2020-07-10 | 嘉兴中科声学科技有限公司 | Shipborne short baseline positioning system and method |
| CN111914910A (en) * | 2020-07-15 | 2020-11-10 | 南京邮电大学 | Airplane formation identification method based on multi-observation-point shape context |
| CN113741513A (en) * | 2021-08-24 | 2021-12-03 | 北京理工大学 | Method for optimizing formation of ground search task formation of multiple unmanned aerial vehicles under implicit communication condition |
| CN113867342A (en) * | 2021-09-18 | 2021-12-31 | 中国人民解放军海军工程大学 | Anti-ship missile formation recognition target selection system based on Hough transformation and optimized K-means clustering |
| US20220027636A1 (en) * | 2020-07-20 | 2022-01-27 | National University Of Defense Technology | Target Task Intention Identifying Method and Device Based on Unit Distribution Thermal Grid |
| CN114117073A (en) * | 2021-11-29 | 2022-03-01 | 中国人民解放军国防科技大学 | Method and device for identifying group intention, computer equipment and readable medium |
| CN114359604A (en) * | 2021-12-07 | 2022-04-15 | 电子科技大学 | Formation recognition model construction method based on multi-aircraft trajectory data |
| CN114611911A (en) * | 2022-03-02 | 2022-06-10 | 博雷顿科技有限公司 | Multi-vehicle unmanned cooperative operation design method, algorithm system and vehicle |
-
2022
- 2022-07-28 CN CN202210894545.6A patent/CN115061501B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160272340A1 (en) * | 2014-12-24 | 2016-09-22 | Environmental Tectonics Corporation | Aircraft-vision systems and methods for maintaining situational awareness and spatial orientation |
| CN110390337A (en) * | 2019-06-27 | 2019-10-29 | 清华大学 | A kind of naval vessel individual discrimination method |
| CN111398903A (en) * | 2020-04-03 | 2020-07-10 | 嘉兴中科声学科技有限公司 | Shipborne short baseline positioning system and method |
| CN111914910A (en) * | 2020-07-15 | 2020-11-10 | 南京邮电大学 | Airplane formation identification method based on multi-observation-point shape context |
| US20220027636A1 (en) * | 2020-07-20 | 2022-01-27 | National University Of Defense Technology | Target Task Intention Identifying Method and Device Based on Unit Distribution Thermal Grid |
| CN113741513A (en) * | 2021-08-24 | 2021-12-03 | 北京理工大学 | Method for optimizing formation of ground search task formation of multiple unmanned aerial vehicles under implicit communication condition |
| CN113867342A (en) * | 2021-09-18 | 2021-12-31 | 中国人民解放军海军工程大学 | Anti-ship missile formation recognition target selection system based on Hough transformation and optimized K-means clustering |
| CN114117073A (en) * | 2021-11-29 | 2022-03-01 | 中国人民解放军国防科技大学 | Method and device for identifying group intention, computer equipment and readable medium |
| CN114359604A (en) * | 2021-12-07 | 2022-04-15 | 电子科技大学 | Formation recognition model construction method based on multi-aircraft trajectory data |
| CN114611911A (en) * | 2022-03-02 | 2022-06-10 | 博雷顿科技有限公司 | Multi-vehicle unmanned cooperative operation design method, algorithm system and vehicle |
Non-Patent Citations (3)
| Title |
|---|
| SURAJ BIJJAHALLI，ROBERTOS ABATINI；ALESSANDRO GARDI: "Advances in intelligent and autonomous navigation systems for small UAS", 《PROGRESS IN AEROSPACE SCIENCES》 * |
| 刘文一; 孙伟; 朱良明; 赵志博: "舰载飞行器打击水面舰艇编队队形识别和目标选择方法", 《兵器装备工程学报》 * |
| 周强: "基于计算机视觉的无人机目标检测与队形识别技术研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115061501B (en) | 2022-11-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111931902B (en) | Generating countermeasure network model and vehicle track prediction method using generating countermeasure network model | |
| CN105488528B (en) | Neural network image classification method based on improving expert inquiry method | |
| CN109492748B (en) | Method for establishing medium-and-long-term load prediction model of power system based on convolutional neural network | |
| CN105279555A (en) | Self-adaptive learning neural network implementation method based on evolutionary algorithm | |
| CN110164129B (en) | Single-intersection multi-lane traffic flow prediction method based on GERNN | |
| CN111753207B (en) | Collaborative filtering method for neural map based on comments | |
| CN113487855B (en) | Traffic flow prediction method based on EMD-GAN neural network structure | |
| CN109948716A (en) | A kind of airport delay prediction technique based on region residual sum LSTM network | |
| CN117746260B (en) | Remote sensing data intelligent analysis method and system | |
| CN110689184A (en) | Method for predicting rail traffic stream of people through deep learning | |
| CN112926645B (en) | Electricity stealing detection method based on edge calculation | |
| CN101740029B (en) | Three-particle cooperative optimization method applied to vector quantization-based speaker recognition | |
| CN115732034A (en) | Identification method and system of spatial transcriptome cell expression pattern | |
| CN113298817A (en) | High-accuracy semantic segmentation method for remote sensing image | |
| CN114693064A (en) | Building group scheme generation performance evaluation method | |
| CN115099328A (en) | Traffic flow prediction method, system, device and storage medium based on countermeasure network | |
| CN116844041A (en) | Cultivated land extraction method based on bidirectional convolution time self-attention mechanism | |
| CN115471016A (en) | Typhoon prediction method based on CISSO and DAED | |
| CN111861231A (en) | Flight training special situation intelligent generation method based on decision tree | |
| CN114004152B (en) | Multi-wind-field wind speed space-time prediction method based on graph convolution and recurrent neural network | |
| CN115061501B (en) | Method and system for identifying coded data of air formation | |
| CN117636183A (en) | Small sample remote sensing image classification method based on self-supervision pre-training | |
| CN116541593B (en) | Course recommendation method based on hypergraph neural network | |
| CN117290673A (en) | Ship energy consumption high-precision prediction system based on multi-model fusion | |
| CN109636194B (en) | Multi-source cooperative detection method and system for major change of power transmission and transformation project |
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 |