CN114764879A - Aerial target identification method based on radar infrared fusion characteristics - Google Patents

Aerial target identification method based on radar infrared fusion characteristics Download PDF

Info

Publication number
CN114764879A
CN114764879A CN202210259801.4A CN202210259801A CN114764879A CN 114764879 A CN114764879 A CN 114764879A CN 202210259801 A CN202210259801 A CN 202210259801A CN 114764879 A CN114764879 A CN 114764879A
Authority
CN
China
Prior art keywords
feature
feature set
radar
infrared
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.)
Pending
Application number
CN202210259801.4A
Other languages
Chinese (zh)
Inventor
谢荣
黄超
刘峥
刘莎莎
冉磊
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.)
Xidian University
Original Assignee
Xidian 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 Xidian University filed Critical Xidian University
Priority to CN202210259801.4A priority Critical patent/CN114764879A/en
Publication of CN114764879A publication Critical patent/CN114764879A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/211—Selection of the most significant subset of features
    • G06F18/2113—Selection of the most significant subset of features by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Aiming, Guidance, Guns With A Light Source, Armor, Camouflage, And Targets (AREA)

Abstract

The invention discloses an aerial target identification method based on radar infrared fusion characteristics, which comprises the following steps: acquiring a first feature set corresponding to a target to be trained, wherein the first feature set comprises: a first radar feature set and a first infrared feature set; determining a preset discriminator model based on the first feature set; acquiring a second feature set corresponding to a target to be identified, wherein the second feature set comprises: a second radar feature set and a second infrared feature set; analyzing the second feature set through the preset discriminator model to determine the category of the target to be recognized so as to realize target recognition; wherein the categories include in-library targets and out-of-library targets. The invention can improve the accuracy and efficiency of aerial target identification.

Description

Aerial target identification method based on radar infrared fusion characteristics
Technical Field
The invention belongs to the technical field of information fusion, and particularly relates to an aerial target identification method based on radar infrared fusion characteristics.
Background
The guidance technology of multimode compound homing has become a mainstream research direction for accurate guidance weapons.
In the prior art, a millimeter wave radar seeker is usually adopted for aerial target identification, and although the millimeter wave radar seeker has the advantages of long detection distance, all-weather working and the like, the millimeter wave radar seeker is easy to be subjected to electronic interference and electronic deception, so that the detection precision and the detection efficiency of aerial target identification are influenced; in addition, an infrared thermal imaging seeker is adopted for aerial target identification in the prior art, and although the infrared thermal imaging seeker has the advantages of high detection precision, strong anti-interference capability and the like, the detection distance is short, so that the target identification range is limited.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides an aerial target identification method based on radar infrared fusion characteristics. The technical problem to be solved by the invention is realized by the following technical scheme:
an aerial target identification method based on radar infrared fusion characteristics comprises the following steps: step 1: acquiring a first feature set corresponding to a target to be trained, wherein the first feature set comprises: a first radar feature set and a first infrared feature set; and 2, step: determining a preset discriminator model based on the first feature set; and step 3: acquiring a second feature set corresponding to a target to be identified, wherein the second feature set comprises: a second radar feature set and a second infrared feature set; and 4, step 4: analyzing the second feature set through the preset discriminator model to determine the category of the target to be recognized so as to realize target recognition; wherein the categories include targets to be attacked and interfering targets.
In one embodiment of the present invention, the step 1 comprises: step 1-1: acquiring a feature set to be trained corresponding to a target to be trained, wherein the feature set to be trained comprises a radar feature set { x ] to be trained1,x2,…,xi,…,xmAnd the infrared feature set to be trained { y }1,y2,…,yj,…,yn}; wherein x isiE is R, i is more than or equal to 1 and less than or equal to m, and m is the dimension of radar characteristics; y isjE is R, j is more than or equal to 1 and less than or equal to n, n is the dimensionality of the infrared characteristic, and R is a real number set; step 1-2: respectively standardizing the radar feature set to be trained and the infrared feature set to be trained based on a preset standardization processing algorithm to obtain a first radar feature set X and a first infrared feature set Y, which are expressed as follows:
Figure BDA0003550343250000021
Figure BDA0003550343250000022
wherein, muiAnd σiRespectively representing features x of radar to be trainediMean and variance of (d), mujAnd σjRespectively represent the infrared characteristics y to be trainedjMean and variance of; the first radar feature set X is represented as: x ═ X1,X2,…,Xi,…,Xm}; the first set of infrared features Y is expressed as: y ═ Y1,Y2,…,Yj,…,Yn}。
In one embodiment of the present invention, the step 2 comprises: step 2-1: obtaining a first radar discrimination feature X from the first radar feature set based on a feature selection algorithm(f)And acquiring a first infrared identification characteristic Y from the first infrared characteristic set(g)(ii) a Step 2-2: performing feature fusion processing on the first radar identification feature and the first infrared identification feature based on a preset feature fusion algorithm to obtain a first fusion feature vector; step 2-3: based on the first fusionAnd establishing a preset discriminator model by using the characteristic vector.
The invention has the beneficial effects that:
1) has real-time processing capability.
The existing decision-level fusion target recognition method is difficult to mine deep information of a radar sensor and an infrared sensor. In the invention, two different feature sets of radar and infrared are subjected to feature weight sorting by a feature selection method, and a feature set with the best separability is screened out after sorting, so that the feature dimension is reduced, data can be processed in real time, and a good target identification effect can be achieved.
2) The capacity of the radar infrared system for identifying the empty target is improved.
The existing decision fusion method faces the problem that Basic Probability Assignment (BPA) construction conforms to practical difficulty and conflict evidence fusion in practical application, usually, the recognition effect is greatly reduced, and the performance of a decision fusion target recognition algorithm is greatly limited. In the invention, the feature fusion is applied to the feature layer and then the target identification is carried out, so that the infrared deep information of the radar is greatly reserved, and the identification capability of the information fusion system on the targets inside and outside the database can be greatly improved.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
FIG. 1 is a schematic flow chart of an air target identification method based on radar infrared fusion characteristics according to the present invention;
FIG. 2 is a schematic flow chart of another radar infrared fusion feature-based air target identification method provided by the invention;
FIG. 3 is a schematic flow chart of a multi-evaluation criterion feature selection algorithm;
FIG. 4 is a schematic flow chart of a feature fusion algorithm based on correlation discriminant analysis;
FIG. 5 is a schematic diagram of a target authentication implementation flow;
FIG. 6 is a diagram showing the results of feature weights of three algorithms;
FIG. 7 is a graph of the average discrimination of three algorithms;
FIG. 8 is a schematic diagram of feature space comparison of three algorithms; wherein, the graph (a) is a characteristic space graph of a Relief algorithm; (b) the figure is a characteristic space diagram of the MRMR algorithm; (c) the figure is a characteristic space diagram of a multi-evaluation criterion algorithm.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but the embodiments of the present invention are not limited thereto.
The millimeter wave radar/infrared thermal imaging dual-mode composite guidance system utilizes the advantage combination detection of each single mode, can make up for the deficiency by taking the advantages and making up for the defects and deficiencies under the condition of single guidance. The information fusion technology based on the characteristic level can fuse multi-source characteristic information provided by a radar and an infrared seeker under various complex and uncertain conditions, filter useless characteristics and redundant characteristics, reduce characteristic dimensionality, solve the problem of data heterogeneity to a certain extent, and therefore improve the precision and efficiency of an information fusion system in a target classification task.
The target identification technology based on the fusion characteristics is to extract radar infrared characteristics of a target on the basis of detecting and positioning the target by a radar so as to judge the attribute and the type of the target. Aiming at the air battlefield environment, because the background is clean, the interference of the natural environment is not needed to be considered, and the air target identification only needs to identify the enemy fighter plane and eliminate several typical radar infrared interferences. And according to different data types of processed objects, the target attribute fusion is divided into data-level fusion, feature-level fusion and decision-level fusion. Compared with data level fusion, the characteristic level fusion adopted by the invention realizes information compression and is beneficial to real-time processing; compared with decision-level fusion, the method retains more information of the target and reduces information loss.
The invention aims to provide an aerial target identification method based on radar infrared fusion characteristics, which solves the problem of completing aerial target identification by using characteristic level information fusion in an aerial scene, improves the target identification performance of an information fusion system, and can be used for identity confirmation of enemy targets in radar and infrared composite guidance; the method can reduce the characteristic dimension while ensuring the maximum scattering matrix among the classes and the minimum scattering matrix in the classes, and is favorable for further improving the identification performance of the information fusion system.
Examples
Referring to fig. 1, fig. 1 is a schematic flow chart of an air target identification method based on radar infrared fusion characteristics according to an embodiment of the present invention.
The technical idea for realizing the purpose of the invention is as follows: 1) and in the off-line training stage, the characteristics of the radar and infrared detector target training samples are respectively extracted, and the separability of each characteristic is analyzed by utilizing simulation data to effectively select the characteristics. On the basis, the radar and the infrared features are fused by using a feature level fusion algorithm to obtain a fused feature vector of the training sample. And then, a Support Vector Data Description (SVDD) model is used as a target discriminator model, the model can effectively distinguish the targets in the database from the targets out of the database, the model is used for training, and a multi-sensor fused target training discrimination database is generated. 2) And in the on-line identification stage, a target radar and infrared features received by a sensor are fused to generate a feature vector, an identification library provides a support vector, a feature area radius, a feature area statistical mean vector and the like for the identifier, and finally, an SVDD model is used for distinguishing the obtained fusion feature vector to determine whether the target is a target in the library.
The method comprises the following steps:
step 1: acquiring a first feature set corresponding to a target to be trained, wherein the first feature set comprises: a first radar feature set and a first infrared feature set.
Optionally, the step 1 includes:
step 1-1: acquiring a feature set to be trained corresponding to a target to be trained, wherein the feature set to be trained comprises a radar feature set { x ] to be trained1,x2,…,xi,…,xm} and infrared feature set to be trained { y1,y2,…,yj,…,yn}。
Wherein x isiBelongs to R, i is more than or equal to 1 and less than or equal to m, and m is the dimension of radar characteristics; y isjE.r, j is more than or equal to 1 and less than or equal to n, n is infrared characteristicDimension, R is a set of real numbers.
Step 1-2: respectively standardizing the radar feature set to be trained and the infrared feature set to be trained based on a preset standardization processing algorithm to obtain a first radar feature set X and a first infrared feature set Y, which are expressed as follows:
Figure BDA0003550343250000061
Figure BDA0003550343250000062
wherein, muiAnd σiRespectively representing features x of radar to be trainediMean and variance of (d), mujAnd σjRespectively represent the infrared characteristics y to be trainedjMean and variance of.
The first radar feature set X is represented as: x ═ X1,X2,…,Xi,…,Xm}。
The first set of infrared features Y is expressed as: y ═ Y1,Y2,…,Yj,…,Yn}。
Presetting a normalization processing algorithm such as a Z-Score method, the normalization processing refers to scaling the feature data to [0, 1%]Within the interval, i.e. Xi∈[0,1]，Yj∈[0,1]。
And 2, step: based on the first set of features, a preset discriminator model is determined.
Optionally, the step 2 includes:
step 2-1: obtaining a first radar discrimination feature X from the first radar feature set based on a feature selection algorithm(f)And acquiring a first infrared identification characteristic Y from the first infrared characteristic set(g)；
Optionally, before the step 2-1, the method further includes:
step S11: calculating a first weight value corresponding to each feature in the first radar feature set and the first infrared feature set based on a Relief evaluation criterion, and sorting the first weight values, wherein the first weight values are represented as:
Sort(Relief)＝[SR(1),SR(2),…,SR(i),…,SR(N)]，
wherein N represents the number of features, SR() represents a first weight value;
it should be noted that the larger the calculated weight value is, the stronger the classification performance of the features is.
Step S12: calculating a second weight value corresponding to each feature in the first radar feature set and the first infrared feature set based on the MRMR evaluation criterion, and sorting the second weight values, wherein the second weight values are expressed as:
Sort(MRMR)＝[SM(1),SM(2),…,SM(i),…,SM(N)]，
wherein S isM() represents a second weight value;
step S13: based on an order relation analysis algorithm, sorting the first weight value and the second weight value to obtain a comprehensive sorting result S (i), which is expressed as:
S(i)＝w1SR(i)+w2SM(i)，
wherein, w1And w2Respectively representing the weight coefficients of different evaluation criteria.
And S (i) is a weight sorting sequence number of the ith feature in the N-dimensional feature set after fusion processing.
Optionally, step 2-1 includes:
according to a feature selection algorithm, obtaining a first radar identification feature X from a sequencing result of a first weight value and a second weight value(f)And a first infrared identifying characteristic Y(g)。
The identification features obtained by the feature selection algorithm have the characteristics of good identification effect and minimum dimension.
Step 2-2: performing feature fusion processing on the first radar identification feature and the first infrared identification feature based on a preset feature fusion algorithm to obtain a first fusion feature vector;
the preset feature fusion algorithm is a related discriminant analysis feature fusion algorithm.
Optionally, step 2-2 includes:
step 2-21: the fusion criteria function is represented as:
Figure BDA0003550343250000071
wherein α and β represent a T-dimensional non-zero Correlation Discrimination Vector (CDV), SbxInter-class scatter matrix of X, SwxAn intra-class scatter matrix of X, ScxAn autocovariance matrix of X, ScIs a cross covariance matrix;
step 2-22: solving the fusion criterion function based on a Lagrange multiplier method to obtain an optimized objective function, which is expressed as:
Figure BDA0003550343250000081
for solving, the three independent parts can be formed by splitting the three parts, and the solutions are respectively carried out by utilizing a Lagrange multiplier method. The three parts split into the Fisher discriminant model parts are used for realizing analysis and optimization of separability among different types of targets; a typical Correlation Analysis (CCA) model part is used for maximizing the intra-class Correlation between two characteristic components and realizing fusion; an improved CCA model part is used for removing the correlation between classes.
Step 2-23: obtaining a solution corresponding to the first radar feature set, expressed as:
Figure BDA0003550343250000082
in the same way, a solution corresponding to the first infrared feature set is obtained, which is expressed as:
Figure BDA0003550343250000083
step 2-24: obtaining a conversion weight W of the first radar feature set and the first infrared feature setCDVxAnd WCDVyExpressed as:
Figure BDA0003550343250000084
step 2-25: based on the conversion weight, performing feature fusion processing on the first radar identification feature and the first infrared identification feature to obtain a first fusion feature vector Z, which is expressed as:
Figure BDA0003550343250000091
step 2-3: and establishing a preset discriminator model based on the first fusion feature vector.
In step 2-3, an SVDD single classifier is preselected, the boundary of the sample characteristic region is determined by searching for a support vector, and the class judgment is carried out by utilizing the distance between the boundary and the sample to be detected. An aerial target SVDD discriminator model is designed. In the training stage, a K-Means clustering method based on correlation coefficient preprocessing is used for carrying out region division on a target sample feature space in a library, a support vector field description method is used for determining the boundary and the support vector of the sample feature space, the boundary of the sample feature space is used for carrying out first judgment, the weighted K nearest neighbor principle is used for calculating the distance between a test sample and the support vector of each feature distribution region, and therefore the final judgment result is obtained.
After the target feature space is subjected to clustering processing, each subarea presents the characteristic of central polymeric distribution. And aiming at each characteristic subregion, searching a hypersphere containing all the sample radii of the characteristic subregion according to the sample space distribution condition of the characteristic subregion, and simultaneously enabling the radius R to be minimum so as to expect to obtain a compact hypersphere boundary capable of accurately describing all the sample distributions of the characteristic subregion.
Figure BDA0003550343250000092
Wherein, Xi(i＝1,2,…,np) Is the feature vector of the p-th feature sub-region, npThe number of feature vectors of the p-th feature region, wherein R is SVDD radius, C represents weight factor, xiiDenotes the relaxation vector, alpha denotes the center of the hyper sphere, p (X)i) Is the deviation of the feature space density center by a weight, δ (X)i) Is the density weight of the sample feature space area.
The solution is carried out by using a Lagrange operator,
Figure BDA0003550343250000093
optionally, after the step 2-3, the method further includes:
step S21: obtaining a third feature set corresponding to a target to be tested, wherein the third feature set comprises: a third radar feature set and a third infrared feature set;
step S22: and adjusting parameters of the preset discriminator model based on the third feature set to obtain a preset target discriminator model.
After the target feature space is subjected to the clustering algorithm and the SVDD algorithm, the feature space can be simplified by the support vector X'i,pCharacteristic region radius RpStatistical mean vector O with feature regionpAnd (4) showing. Extracting the characteristics of the target to be tested to obtain a characteristic vector Z, and calculating the weight corresponding to each support vector in the vector D:
Figure BDA0003550343250000101
calculating the mean vector of the K neighbor feature vectors of the target to be tested on the basis:
Figure BDA0003550343250000102
calculating XmeanL1 norm distance from the target to be tested:
Figure BDA0003550343250000103
setting a threshold value when d2(Z,Xmean) When the value is more than or equal to xi, the target to be tested is an out-of-library target, and when d is greater than or equal to xi2(Z,Xmean)<And xi, the target to be tested is a target in the library.
The invention can adjust the parameters of the preset discriminator model through the target to be tested, so that the discrimination effect is optimal.
And step 3: acquiring a second feature set corresponding to a target to be identified, wherein the second feature set comprises: a second set of radar features and a second set of infrared features.
And 4, step 4: analyzing the second feature set through the preset discriminator model to determine the category of the target to be recognized so as to realize target recognition; wherein the categories include in-library targets and out-of-library targets.
The in-library targets are also called targets to be attacked, and the out-of-library targets are also called interference targets.
Optionally, the analyzing the second feature set by the preset evaluator model includes:
and analyzing the second feature set through the preset target discriminator model.
It should be noted that, in the present invention, a second fusion feature vector is obtained based on a second feature set corresponding to a target to be recognized, an obtaining manner of the second fusion feature vector corresponding to the target to be recognized is the same as an obtaining manner of a first fusion feature vector corresponding to a target to be trained, that is, based on a second radar feature set and a second infrared feature set, a second radar identification feature and a second infrared identification feature are obtained, a second fusion feature vector is further obtained, and the second fusion feature vector is analyzed through the preset discriminator model, so as to determine whether the target to be recognized is an in-library target or an out-of-library target. In addition, the discriminator model is obtained based on the theoretical design of a single classifier of a support vector machine.
Fig. 2 is a schematic flow chart of another radar infrared fusion feature-based air target identification method provided by the invention;
in conclusion, the invention has the beneficial effects that:
1) has real-time processing capability.
The existing decision-level fusion target recognition method is difficult to mine deep information of a radar sensor and an infrared sensor. In the invention, two different feature sets of radar and infrared are subjected to feature weight sorting by a feature selection method, and a feature set with the best separability is screened out after sorting, so that the feature dimension is reduced, data can be processed in real time, and a good target identification effect can be achieved.
2) The capacity of the radar infrared system for identifying the empty target is improved.
The existing decision fusion method faces the problems that BPA construction is difficult to conform to reality and evidence fusion conflicts in practical application, and usually the recognition effect is greatly reduced, so that the performance of a decision fusion target recognition algorithm is greatly limited. In the invention, the feature fusion is applied to the feature layer and then the target identification is carried out, so that the infrared deep information of the radar is greatly reserved, and the identification capability of the information fusion system on the targets inside and outside the database can be greatly improved.
The beneficial effects of the invention are further verified based on experiments:
1. the experimental conditions are as follows:
after the collected radar infrared sensor information is subjected to feature extraction, the aforementioned feature selection algorithm is adopted for sorting, the sorting result is shown in fig. 6, and the features with good feature separability occupy higher weight. The sorting result is combined with the relationship graph of the feature subset dimension of fig. 7 and the average identification accuracy of the three algorithms, so that the optimal feature subset can be obtained, the identification effect is ensured, the feature subset dimension for identification is minimized, and the real-time performance of the algorithms is ensured. The final feature selection and identification results are as follows:
the feature sets of the extracted radar and infrared are respectively { signal-to-noise ratio feature, amplitude variance feature, frequency domain variance feature }, { compactness feature, mean contrast feature, standard deviation feature and Hull invariant moment feature }, and specific sample sets are shown in Table 1.
TABLE 1 sample set
Figure BDA0003550343250000121
Figure BDA0003550343250000131
Here, two experiments were set up in total: experiment 1 adopts a characteristic space evaluation criterion to compare the advantages and disadvantages of three algorithms of Relief, MRMR and a multi-evaluation criterion in a multi-dimensional fusion characteristic space; experiment 2 fixes the dimension of the fusion features according to the result of experiment 1 to obtain the target identification result of the feature-level fusion features, compares the identification results of decision-level, and compares the identification accuracy of the two.
2. The experimental contents are as follows:
in the experiment 1, the radar feature set and the infrared feature set are respectively fused into 1-16 dimensional features through three algorithms of Relief, MRMR and multi-evaluation criterion for comparative analysis, and the experimental result is shown in FIG. 7.
Fig. 7 is a comparison of the three algorithm feature space scores. From fig. 7, the following conclusions can be drawn: (1) when three algorithms of Relief, MRMR and multi-evaluation criterion are set to be too small and too large in the dimension of the fused characteristic component, the scores of the algorithms are much smaller than the middle quantity of the dimension; through analysis, when the dimension is set to be too small, for example, 1 dimension, the target characteristic information contained in the obtained characteristic component is too little, the distribution of the internal and external characteristics of the formed characteristic space library is relatively disordered, and the internal and external targets cannot be effectively distinguished; when the dimensionality is set to be too large, for example, 16 dimensionalities, the feature digit is equal to the feature digit of the radar and infrared optimal feature subset, and the feature layer fusion is not performed, namely redundant feature information from a radar sensor and an infrared sensor is not abandoned, so that a result of low feature space evaluation is caused; (2) on the other hand, in general, the order of scoring the feature space is: the multiple evaluation criterion > MRMR > Release theoretically analyzes that the correlation of the features and the categories in the Release algorithm is based on the distinguishing capability of the features on close samples, the MRMR algorithm is to find a group of features which have the maximum correlation (Max-Rerelease) with the final output result but have the minimum correlation (Min-Rerelease) with each other in the original feature set, and the multiple evaluation criterion combines the advantages of the two so that the selected features are more separable.
In order to more intuitively see the feature space distribution of the three algorithms, the dimension Dim of the fused feature component is selected to be 3, and fig. 8 is a feature space comparison graph of the three algorithms when Dim is 3. The diagram is marked by a mark o to indicate an in-library target fighter, a mark to indicate an in-library target early warning machine, an □ to indicate an in-library target aerial tanker, and a mark to indicate out-of-library target radar towing interference.
As can be seen from fig. 8, the difference between the feature spaces obtained by the Relief algorithm and the MRMR algorithm is not large, and the features of the in-library and out-library targets are mutually adhered, which is very disadvantageous for eliminating the out-library targets in the authentication stage. The multi-evaluation criterion algorithm used in the method is relatively strong in intra-class cohesion and larger in inter-class distance, and is beneficial to better distinguishing targets inside and outside the library, so that the score of the feature space is higher than that of other two algorithms, and the reflected condition is consistent with that of the upper graph.
Experiment 2, the discrimination accuracy rates of the inside and outside targets under the four conditions of decision fusion, radar single mode, infrared single mode and radar infrared dual mode fusion are briefly compared, so that the influence of the algorithm on the discrimination performance of the composite seeker target is analyzed, and the result is shown in table 2.
TABLE 2 discrimination accuracy in four cases
Figure BDA0003550343250000141
In table 2, decision fusion is based on the discrimination results of the feature BPA, the discrimination rates of the radar single mode and the infrared dual mode respectively correspond to the discrimination results of the optimal feature subsets thereof, and dual mode fusion is the discrimination result of feature fusion performed after the radar and infrared optimal feature subsets are selected by multiple evaluation criteria.
It is clear from table 2 that the following conclusions can be drawn: in the aspect of identification accuracy, the algorithm > decision fusion algorithm > radar/infrared single mode. The main reason is that the algorithm provided by the invention is based on the fusion of radar and infrared features, and the identification accuracy is higher than that of other fusion algorithms by training the sample features in the off-line learning stage aiming at the fusion feature components. In addition, no matter what fusion algorithm is selected, compared with a single model, the authentication performance is improved.
In conclusion, the method carries out target identification on the characteristic layer, and can obtain better identification accuracy rate compared with the traditional method for carrying out fusion on the decision layer.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the present invention, "a plurality" means two or more unless specifically defined otherwise.
While the present application has been described in connection with various embodiments, other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed application, from a review of the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the word "a" or "an" does not exclude a plurality. A single processor or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.

Claims (6)

1. An aerial target identification method based on radar infrared fusion characteristics is characterized by comprising the following steps:
step 1: acquiring a first feature set corresponding to a target to be trained, wherein the first feature set comprises: a first radar feature set and a first infrared feature set;
step 2: determining a preset discriminator model based on the first feature set;
and 3, step 3: acquiring a second feature set corresponding to a target to be identified, wherein the second feature set comprises: a second set of radar features and a second set of infrared features;
and 4, step 4: analyzing the second feature set through the preset discriminator model to determine the category of the target to be recognized so as to realize target recognition; wherein the categories include in-library targets and out-of-library targets.
2. The method of claim 1, wherein step 1 comprises:
step 1-1: acquiring a feature set to be trained corresponding to a target to be trained, wherein the feature set to be trained comprises a radar feature set { x ] to be trained1,x2,…,xi,…,xmAnd the infrared feature set to be trained { y }1,y2,…,yj,…,yn}；
Wherein x isiBelongs to R, i is more than or equal to 1 and less than or equal to m, and m is the dimension of radar characteristics; y isjBelongs to R, j is more than or equal to 1 and less than or equal to n, n is the dimensionality of the infrared characteristic, and R is a real number set;
step 1-2: based on a preset standardization processing algorithm, respectively standardizing the radar feature set to be trained and the infrared feature set to be trained to obtain a first radar feature set X and a first infrared feature set Y, which are expressed as follows:
Figure FDA0003550343240000011
Figure FDA0003550343240000012
wherein, muiAnd σiRespectively representing features x of radar to be trainediMean and variance of (d), mujAnd σjRespectively represent the infrared characteristics y to be trainedjMean and variance of;
the first radar feature set X is represented as: x ═ X1,X2,…,Xi,…,Xm}；
The first set of infrared features Y is expressed as: y ═ Y1,Y2,…,Yj,…,Yn}。
3. The method of claim 2, wherein step 2 comprises:
step 2-1: obtaining a first radar discrimination feature X from the first radar feature set based on a feature selection algorithm(f)And acquiring a first infrared identification characteristic Y from the first infrared characteristic set(g)；
Step 2-2: performing feature fusion processing on the first radar identification feature and the first infrared identification feature based on a preset feature fusion algorithm to obtain a first fusion feature vector;
step 2-3: and establishing a preset discriminator model based on the first fusion feature vector.
4. The method of claim 2, wherein prior to step 2-1, the method further comprises:
step S11: calculating a first weight value corresponding to each feature in the first radar feature set and the first infrared feature set based on a Relief evaluation criterion, and sorting the first weight values, wherein the first weight values are represented as:
Sort(Relief)＝[SR(1),SR(2),…,SR(i),…,SR(N)]，
wherein N represents the number of features, SR() represents a first weight value;
step S12: calculating a second weight value corresponding to each feature in the first radar feature set and the first infrared feature set based on the MRMR evaluation criterion, and sorting the second weight values, wherein the second weight values are expressed as:
Sort(MRMR)＝[SM(1),SM(2),…,SM(i),…,SM(N)]，
wherein S isM() represents a second weight value;
step S13: based on an order relation analysis algorithm, sorting the first weight value and the second weight value to obtain a comprehensive sorting result S (i), which is expressed as:
S(i)＝w1SR(i)+w2SM(i)，
wherein, w1And w2Respectively representing weight coefficients of different evaluation criteria;
the step 2-1 comprises the following steps:
according to a feature selection algorithm, obtaining a first radar identification feature X from a sequencing result of a first weight value and a second weight value(f)And a first infrared identifying characteristic Y(g)。
5. The method of claim 2, wherein the step 2-2 comprises:
step 2-21: the fusion criteria function is represented as:
Figure FDA0003550343240000031
wherein α and β represent a T-dimensional non-zero correlation discrimination vector, SbxInter-class scatter matrix of X, SwxAn intra-class scatter matrix of X, ScxAn autocovariance matrix of X, ScIs a cross covariance matrix;
step 2-22: solving the fusion criterion function based on a Lagrange multiplier method to obtain an optimized objective function, which is expressed as:
Figure FDA0003550343240000032
step 2-23: obtaining a solution corresponding to the first radar feature set, expressed as:
Figure FDA0003550343240000033
obtaining a solution corresponding to the first set of infrared features, expressed as:
Figure FDA0003550343240000041
step 2-24: obtaining a conversion weight W of the first radar feature set and the first infrared feature setCDVxAnd WCDVyExpressed as:
Figure FDA0003550343240000042
step 2-25: based on the conversion weight, performing feature fusion processing on the first radar identification feature and the first infrared identification feature to obtain a first fusion feature vector Z, which is expressed as:
Figure FDA0003550343240000043
6. the method of claim 3, wherein after the steps 2-3, the method further comprises:
step S21: acquiring a third feature set corresponding to a target to be tested, wherein the third feature set comprises: a third radar feature set and a third infrared feature set;
step S22: adjusting parameters of the preset discriminator model based on the third feature set to obtain a preset target discriminator model;
the analyzing the second feature set by the preset evaluator model includes:
and analyzing the second feature set through the preset target discriminator model.
CN202210259801.4A 2022-03-16 2022-03-16 Aerial target identification method based on radar infrared fusion characteristics Pending CN114764879A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210259801.4A CN114764879A (en) 2022-03-16 2022-03-16 Aerial target identification method based on radar infrared fusion characteristics

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210259801.4A CN114764879A (en) 2022-03-16 2022-03-16 Aerial target identification method based on radar infrared fusion characteristics

Publications (1)

Publication Number Publication Date
CN114764879A true CN114764879A (en) 2022-07-19

Family

ID=82364971

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210259801.4A Pending CN114764879A (en) 2022-03-16 2022-03-16 Aerial target identification method based on radar infrared fusion characteristics

Country Status (1)

Country Link
CN (1) CN114764879A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116091883A (en) * 2023-04-10 2023-05-09 深圳市普雷德科技有限公司 Target detection and identification method, system and storage medium based on multi-source information fusion

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116091883A (en) * 2023-04-10 2023-05-09 深圳市普雷德科技有限公司 Target detection and identification method, system and storage medium based on multi-source information fusion

Similar Documents

Publication Publication Date Title
CN110309302B (en) Unbalanced text classification method and system combining SVM and semi-supervised clustering
Yoon et al. LFIQ: Latent fingerprint image quality
EP2273424A1 (en) Automated target detection and recognition system and method
US20070211964A1 (en) Image-based indexing and classification in image databases
CN109165639B (en) Finger vein identification method, device and equipment
Hambalık Fingerprint recognition system using artificial neural network as feature extractor: design and performance evaluation
Bhuyan et al. An effective method for fingerprint classification
CN114564982A (en) Automatic identification method for radar signal modulation type
CN108073940B (en) Method for detecting 3D target example object in unstructured environment
CN109613485A (en) A kind of Radar Signal Sorting Method based on support vector clustering and gray relation grades
CN111783837A (en) Feature fusion method based on multi-core learning
CN106529501B (en) Fingerprint and finger vein image fusion method based on weighted fusion and hierarchical serial structure
CN114764879A (en) Aerial target identification method based on radar infrared fusion characteristics
CN111896430A (en) Pollen monitoring method and device
Li et al. A score-level fusion fingerprint indexing approach based on minutiae vicinity and minutia cylinder-code
Li et al. A fingerprint indexing scheme with robustness against sample translation and rotation
CN106326927B (en) A kind of shoes print new category detection method
CN112861871A (en) Infrared target detection method based on target boundary positioning
Wahlberg Gaussian process classification as metric learning for forensic writer identification
Yue et al. Plant leaf recognition based on naive Bayesian classification and linear discriminant analysis model
Ma et al. An exceedingly fast model for low resolution handwritten digit string recognition
Li et al. A novel fingerprint indexing approach focusing on minutia location and direction
Le et al. Logo spotting for document categorization
CN114580482A (en) Radio signal characteristic acquisition method based on edge computing node and data center
Gao et al. Feature tracking for target identification in acoustic image sequences

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