CN112307935B - Multi-attribute information weighted fusion target identification method based on DS rule - Google Patents
Multi-attribute information weighted fusion target identification method based on DS rule Download PDFInfo
- Publication number
- CN112307935B CN112307935B CN202011168979.5A CN202011168979A CN112307935B CN 112307935 B CN112307935 B CN 112307935B CN 202011168979 A CN202011168979 A CN 202011168979A CN 112307935 B CN112307935 B CN 112307935B
- Authority
- CN
- China
- Prior art keywords
- data signal
- radar
- signal
- attribute
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 58
- 230000004927 fusion Effects 0.000 title claims abstract description 51
- 239000008186 active pharmaceutical agent Substances 0.000 title claims abstract description 29
- 230000005855 radiation Effects 0.000 claims abstract description 40
- 230000006870 function Effects 0.000 claims description 58
- 238000012549 training Methods 0.000 claims description 39
- 238000005457 optimization Methods 0.000 claims description 25
- 238000004590 computer program Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 description 4
- 238000012795 verification Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000007499 fusion processing Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Signal Processing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a DS rule-based multi-attribute information weighted fusion target identification method, which comprises the steps of obtaining radar radiation source signals and extracting corresponding data signals; using a data signal as input information, and determining a classification result of the data signal based on EKNN classifier; determining the type of the radar radiation source signal according to the classification result of the data signal; when the data signals corresponding to the radar radiation source signals are classified by the EKNN classifier, the weight value is given to the dimension attribute in the class corresponding to each data signal, different radar signal attributes of each class are utilized in a distinguishing way, and the higher/lower weight value is given to the radar signal attribute with larger/smaller influence degree on the data signal classification result, so that each radar signal attribute has different influence on different data signal classes, the accuracy of data signal classification can be effectively improved, and the classification accuracy of the radar radiation source signals is further improved.
Description
Technical Field
The invention belongs to the technical field of computers, and particularly relates to a DS rule-based multi-attribute information weighted fusion target identification method.
Background
Radar is a device that uses radio waves to determine the orientation of an object. The radar can be used for detecting military targets such as airplanes, ships and the like, can be used for navigating the airplanes, can be used for researching stars, detecting abrupt weather such as typhoons, thunderstorms and the like, and is very wide in application range.
Radar plays an important role in many fields, such as ground radar, airborne radar, pulsed doppler radar, synthetic aperture radar, space-borne information radar, missile guidance radar, weather radar, etc., and the range of radar applications is beginning to go from ground to sea and also in air, even to space. Therefore, radar has been applied in various aspects, playing an important role. And intercepting radar signals in a battlefield, judging the radar type according to the radar signals, and further judging the type of the enemy equipment, so that the enemy equipment can be dealt with in advance according to the type of the enemy equipment.
Radar countermeasures are a technology for performing reconnaissance and interference on an enemy radar by using specific electronic equipment, the enemy radar parameters and performances are reconnaissance to acquire the enemy army information, and appropriate measures are adopted to take corresponding measures, so that the normal combat capability of the enemy radar is greatly reduced, and the strength of the army is saved. Then it is of great importance how to determine which radar type the intercepted radar signal belongs to. The conventional judging method is to conduct classification judgment according to multiple attributes of radar signals, but the method is high in error rate and low in accuracy.
Disclosure of Invention
The invention aims to provide a DS rule-based multi-attribute information weighted fusion target identification method, which improves the classification accuracy of radar radiation sources by giving different weight values to different dimension attributes in radar data.
The invention adopts the following technical scheme: a DS rule-based multi-attribute information weighted fusion target identification method comprises the following steps:
Acquiring radar radiation source signals and extracting corresponding data signals; wherein the data signal comprises radar signal attributes of at least two dimensions;
using the data signal as input information, and determining a classification result of the data signal based on EKNN classifier; when the data signals are classified by the EKNN classifier, a weight value is given to the radar signal attribute in the class corresponding to each data signal in the EKNN classifier;
and determining the type of the radar radiation source signal according to the classification result of the data signal.
Further, the calculation method of the weight value comprises the following steps:
constructing a training data set, wherein the training data set comprises a plurality of data signals corresponding to radar radiation source signals of different types and categories corresponding to each data signal;
respectively calculating basic trust function fusion values of each data signal and all neighbors in the training data set; wherein the basic trust function fusion value comprises a weight value;
constructing an optimization objective function according to the basic trust function fusion value, and solving the optimization objective function to obtain the Euclidean distance between the data signal and the neighbor thereof;
and determining the weight value of the radar signal attribute in the category corresponding to the data signal according to the Euclidean distance.
Further, obtaining the euclidean distance between the data signal and its neighbors comprises:
determining a basic trust function value of the data signal and each neighbor thereof according to the solution of the optimized objective function;
And determining the Euclidean distance between the data signal and each neighbor according to the basic trust function value.
Further, determining the classification result of the data signal based on the EKNN classifier includes:
Assuming the data signal as any category in the category set, and respectively using EKNN classifiers to determine sub-classification results of the data signal until traversing the category in the category set to obtain a plurality of sub-classification results; in the process of determining the sub-classification result of the data signal by using EKNN classifier, a weight value is given to each radar signal attribute of the class;
And fusing the sub-classification results to determine the classification result of the data signal.
Further, the optimization objective function is:
Wherein, For a vector consisting of weight values for each radar signal attribute in the L t class, lambda t is the euclidean distance coefficient, q is the total number of data signals in the training dataset,T t is the class truth value of data signal x t for the basic trust function value of data signal x t and its neighbor data signal x k.
Further, the calculation method of the Euclidean distance between the data signal and the adjacent neighbor thereof comprises the following steps:
Where d (x, x t) is the Euclidean distance between the data signal x and its neighbor data signal x t, For the weight value of the nth radar signal attribute of category L t, x n is the nth radar signal attribute value of data signal x,Is the nth radar signal attribute value of data signal x t.
Another technical scheme of the invention is as follows: a DS rule-based multi-attribute information weighted fusion target recognition device comprises:
the extraction and conversion module is used for acquiring radar radiation source signals and extracting corresponding data signals; wherein the data signal comprises radar signal attributes of at least two dimensions;
a classification module for taking the data signal as input information and determining a classification result of the data signal based on the EKNN classifier; when the radar signals are classified by the EKNN classifier, a weight value is given to the radar signal attribute in the class corresponding to each data signal in the EKNN classifier;
And the first determining module is used for determining the type of the radar radiation source signal according to the classification result of the data signal.
Further, the classification module includes a weight value determination module, the weight value determination module including:
The first construction module is used for constructing a training data set, wherein the training data set comprises a plurality of data signals corresponding to radar radiation source signals of different types and categories corresponding to each data signal;
the computing module is used for respectively computing the basic trust function fusion value of each data signal and all the neighbors in the training data set; wherein the basic trust function fusion value comprises a weight value;
the second construction module is used for constructing an optimization objective function according to the basic trust function fusion value, solving the optimization objective function and obtaining the Euclidean distance between the data signal and the neighbor of the data signal;
And the second determining module is used for determining the weight value of each radar signal attribute in the category corresponding to the data signal according to the Euclidean distance.
Another technical scheme of the invention is as follows: a DS rule-based multi-attribute information weighted fusion target recognition device comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor realizes a DS rule-based multi-attribute information weighted fusion target recognition method when executing the computer program.
Another technical scheme of the invention is as follows: a computer readable storage medium storing a computer program which when executed by a processor implements a multi-attribute information weighted fusion target recognition method based on a DS rule of any one of the above.
The beneficial effects of the invention are as follows: when the data signals corresponding to the radar radiation source signals are classified by the EKNN classifier, the weight value is given to the dimension attribute in the class corresponding to each data signal, different radar signal attributes of each class are utilized in a distinguishing way, and the higher/lower weight value is given to the radar signal attribute with larger/smaller influence degree on the data signal classification result, so that each radar signal attribute has different influence on different data signal classes, the accuracy of data signal classification can be effectively improved, and the classification accuracy of the radar radiation source signals is further improved.
Drawings
FIG. 1 is a schematic flow chart of a DS rule-based multi-attribute information weighted fusion target recognition method according to an embodiment of the present application;
FIG. 2 is a flowchart of a method for calculating a weight value according to an embodiment of the present application;
FIG. 3 is a schematic structural diagram of a DS rule-based multi-attribute information weighted fusion target recognition device according to an embodiment of the present application;
fig. 4 is a schematic structural diagram of a multi-attribute information weighted fusion target recognition device based on DS rules according to another embodiment of the present application.
Detailed Description
The invention will be described in detail below with reference to the drawings and the detailed description.
In practical applications, for different radar signal attribute features (such as angle of arrival, carrier frequency, pulse width, repetition frequency, etc.) belonging to the same radar data signal class (such as ground radar, airborne radar, pulsed doppler radar, synthetic aperture radar, space-borne information radar, missile guidance radar, weather radar, etc.), the functions performed in the radar data signal classification process are often different, some radar signal attribute features have a higher degree of identification, and some radar signal attribute features are even irrelevant to the radar data signal classification decision.
Likewise, for the same radar signal attribute characteristics, the importance they play to the classification process is also different in radar radiation source signals of different radar data signal classes. If the classification is directly performed by adopting the traditional classifier, the classification result is greatly affected. In practical applications, how to obtain the weights of the attributes is a problem under the condition of lack of prior knowledge. Therefore, the invention realizes the multi-attribute information fine weighting fusion radar radiation source signal identification through the optimized training of the training data set on the basis of EKNN classifier.
Since the similarity between radar radiation source signals is a precondition for realizing classification, and the similarity is reflected according to the Euclidean distance between radar data signals, the difference of the calculated attribute weights is realized by processing the Euclidean distance function.
According to the invention, the radar data signals with known categories are used as a training data set, the weight value of each radar signal attribute characteristic of each radar data signal category is obtained by optimizing according to an improved Euclidean distance formula and the basic idea of a EKNN classifier, and the obtained weight value is used for carrying out decision classification on the radar data signals to be identified.
The method mainly comprises two parts:
The first part is the optimization calculation of the dimension attribute feature weight value based on the category. And (3) extracting features of each radar data signal sample in a leave-one-out mode in the training data set, converting the extracted features into data signals in a fixed format, processing the data signals, solving radar signal attribute features of each radar data signal category, carrying out radar data signal classification after weighting to obtain trust function values, and then obtaining attribute weight values through optimizing error criteria. The radar data signals are identified by extracting five-dimensional inter-pulse parameters of a pulse descriptor (Pulse Description Word, PDW) to form a feature vector.
The second part is a test of the true class of radar radiation source signals to be identified. For the data signals corresponding to the radar radiation source signals to be identified, respectively supposing that the data signals belong to the L t th class, classifying the data signals to be classified by using the weight value of the radar signal attribute of the L t class to obtain classification results, and the like to obtain L t group sub-classification results, and performing DS fusion on all the obtained sub-classification results to finally obtain a prediction result.
As shown in fig. 1, this embodiment specifically includes the following steps:
S110, acquiring radar radiation source signals and extracting corresponding data signals; wherein the data signal comprises radar signal attributes of at least two dimensions; s120, taking the data signals as input information, and determining classification results of the data signals based on EKNN classifiers; when the radar signals are classified by the EKNN classifier, a weight value is given to the radar signal attribute in the class corresponding to each data signal in the EKNN classifier; s130, determining the type of the radar radiation source signal according to the classification result of the data signal.
When the data signals corresponding to the radar radiation source signals are classified by the EKNN classifier, the weight value is given to the dimension attribute in the class corresponding to each data signal, different radar signal attributes of each class are utilized in a distinguishing way, and the higher/lower weight value is given to the radar signal attribute with larger/smaller influence degree on the data signal classification result, so that each radar signal attribute has different influence on different data signal classes, the accuracy of data signal classification can be effectively improved, and the classification accuracy of the radar radiation source signals is further improved.
In the embodiment of the present invention, the method for calculating the weight value is shown in fig. 2, and specifically includes:
S1210, constructing a training data set, wherein the training data set comprises a plurality of data signals corresponding to radar radiation source signals of different types and categories corresponding to each data signal; s1220, respectively calculating basic trust function fusion values of each data signal and all neighbors in the training data set; wherein the basic trust function fusion value comprises a weight value; s1230, constructing an optimization objective function according to the basic trust function fusion value, and solving the optimization objective function to obtain the Euclidean distance between the data signal and the neighbor thereof; s1240, determining the weight value of the radar signal attribute in the category corresponding to the data signal according to the Euclidean distance.
As a specific implementation, obtaining the euclidean distance between the data signal and its neighbors includes:
Determining a basic trust function value of the data signal and each neighbor thereof according to the solution of the optimized objective function; and determining the Euclidean distance between the data signal and each neighbor according to the basic trust function value.
Further, determining the classification result of the data signal based on the EKNN classifier includes:
Assuming the data signal as any category in the category set, and respectively using EKNN classifiers to determine sub-classification results of the data signal until traversing the category in the category set to obtain a plurality of sub-classification results; in the process of determining the sub-classification result of the data signal by using EKNN classifier, a weight value is given to each radar signal attribute of the class; and fusing the sub-classification results to determine the classification result of the data signal.
More specifically, in the present embodiment, it is assumed that the attribute space of the data signal is N-dimensional (i.e. S 1,S2,...,SN),SN is a radar signal attribute of the data signal, the class set of the EKNN classifier is Ω= { L 1,…,LT},Lt is a class of radar data signals, and t∈t. For the radar data signals used for training, it is often known in class, therefore, some classified radar data signals may be used to perform optimization errors to obtain attribute weight values.
In the conventional radar radiation source signal identification and detection problem, some radar data signals from different classes (such as pulse doppler radar, synthetic aperture radar and air information radar) may have very close euclidean distances (i.e. radar signal attribute feature values are close), and some radar data signals with far euclidean distances may also belong to the same class, which means that a conventional classification mode of simply calculating the euclidean distances is likely to bring wrong decisions without considering the importance of each attribute. If the weight values of different radar signal attribute features under different categories can be calculated by processing and optimizing the radar signal data in the training data set, the false-fraction rate can be greatly reduced.
For example, the radar data signal of the air radar is acquired and classified, and the class set of the classifier at this time is Ω 1 = { passenger plane, transport plane, fighter plane }. For an empty information radar, the radar radiation source signal is extracted to form a multi-dimensional characteristic vector form, namely a data signal. It is assumed that for a certain radar signal characteristic value, the mean value of the radar signal data belonging to the passenger plane class and the transport plane class is 10, and the mean value of the dimension attribute values of the radar signal data belonging to the fighter plane class is 100. That is, the radar signal attribute is distinguished more when discriminating whether the radar radiation source signal is of the fighter class, and it is distinguished less when discriminating that a certain radar radiation source signal is of the passenger class and the conveyor class. Therefore, the classification accuracy of the data signals can be remarkably improved by reasonably considering the attribute weight value of the radar signals based on the category in the classification decision process of the data signals corresponding to the radar radiation source signals.
Consider a set of test radar data signals whose true class is L i,i＝1,...,t,Li ε Ω, where Ω is the class set of classifiers. In the initial training feature dataset, K neighbors of the data signal as training data should be found, which is mainly dependent on local information of the data signal of different classes. It is easy to know that in many cases the euclidean distance between data signals of the same class is smaller than the euclidean distance between data signals of different classes, however in practical applications, there are often cases where the euclidean distance between a certain data signal and a data signal of a different class is smaller, especially at the edge boundaries of two classes of data signals, this phenomenon is more pronounced. Thus, the classification boundary can be made more obvious by the weighting process of the attributes.
The Euclidean distance between data signals is calculated, in order to measure the similarity between the data signals and introduce a class-based radar signal attribute weight value, the Euclidean distance formula needs to be modified to overcome the previous limitation, and the modified Euclidean distance (namely the Euclidean distance between the data signals and the neighbors thereof) is shown as follows:
Where d (x, x t) is the Euclidean distance between the data signal x and its neighbor data signal x t, For the weight value of the nth radar signal attribute of category L t, x n is the nth radar signal attribute value of data signal x,N epsilon N, which is a positive integer, is the nth radar signal attribute value of data signal x t.
From the above formula, the similarity (i.e. euclidean distance) calculation of the data signal after adding the attribute weight value of the radar signal based on the category is obtained, and then through the basic idea of EKNN classifier, bba (basic belief assignment basic confidence assignment) of the data signal for each neighbor can be obtained, and the process is as follows:
In the formula, bba obtained by the target data signal about the neighbor is shown in the formula, d is the euclidean distance between the data signal x t corresponding to the radar signal to be classified and the neighbor data signal thereof, alpha, beta and gamma h are adjustable coefficients, 0< alpha <1, and preferably, alpha takes a value of 0.95, beta epsilon {1,2.. } and if the euclidean distance between a certain neighbor data signal and the data signal corresponding to the radar signal to be classified is large enough, the contribution degree of the neighbor data signal to distinguishing the data signal corresponding to the radar signal to be classified is almost 0. Conversely, if the euclidean distance between two data signals is very close, it is very likely that both are in the same class, and m (L t|xt) represents the probability that the neighbor of the data signal to be classified x i is x t, and the data signal to be classified x i belongs to L t; m (Ω|x t) represents the probability that the data signal x i belongs to Ω, and similarly to m (a|x t), a∈2 Ω\{Ω,Lt, i.e. the category other than Ω, L t in 2 Ω.
According to the above, K groups of basic trust function values of the data signals to be classified about the neighbors thereof can be obtained byThe prediction classification result of the data signal can be obtained by using a classical DS fusion rule, wherein m (|x 1) represents a basic trust function value obtained by the data signal to be classified for x 1 about the neighbor, the right side of the equal sign represents fusion of the obtained K groups of basic trust function values, and the result on the left side of the equal sign is obtained, that is, m (|) represents a fusion result obtained by a certain sample to be classified according to the neighbor.
The attribute weight matrix can be optimally solved by comparing the error between the predicted classification value and the true classification value of the classifier.
Omega ij represents the weight value of the feature of the j-th radar signal attribute in category i, which should satisfy omega ij > 0, i=1, M, j=1, N, M represents the number of categories of the set of data signals, N represents the number of radar signal attributes of the set of categories.
Solving a radar signal attribute weight matrix by adopting the following method:
in this section, a leave-one-out approach is taken in the training dataset. Assuming that the training data set used at this time is Γ= { (x 1,L1),(x2,L2),..(xt,Lt)}.xt) is a data signal corresponding to a radar signal of the class L t, Γ '=Γ\ (x t,Lt) is a new training data set, meaning that in this training data set, data signal x t is removed, K neighbors of x t are found in Γ' and the corresponding basic trust function values are fused, and the class prediction probability vector T t＝[Tt(1),...,Tt(M)]T,Tt(1)、…、Tt (M) of x t can be obtained to represent the number 0 or 1, for example, one training data set has three classes 1.2.3, the class of a certain data signal is 1, then t= [1, 0] T, that is, the corresponding class position is 1, and the rest position is 0. Again because the true class of x t is L t, then the desired prediction output value is Tt＝[Tt(1),...,Tt(M)]T,Tt(Lt)＝1,Tt(Lc)＝0,Lc≠Lt.
Referring to the example above, L t represents the class to which the data signal should be. Then for data signal x t, its output error is then m t(·)-Tt||＝εt,mt (·) representing the fusion classification result of the prediction, where i·i represents the euclidean distance.
For all data signals in class L t, the corresponding error values can be found, while the objective of the final optimization is to make the sum of the errors as small as possible. If the predicted output of the data signal x t is very close to the true value, that is to say the error epsilon t is very small.
In radar radiation source signal identification, it is always desirable to fuse the classification result with the true value of the data signal as close as possible. In the fusion process, if the fusion classification result of a data signal is inconsistent with the true class label of the data signal, that is, the fusion classification result does not distribute the maximum probability to the true class of the data signal, the classification difficulty of the data signal is considered to be high. In practice, the degree of inconsistency between such fused classification results and the true class of data signals is referred to as evidence distance. These data signals with larger evidence distances are also more sensitive to training of source weights. Sample distance coefficients are introduced here to measure the inconsistency between classification results and true categories.
If the evidence distance of a certain data signal is larger, it is more important in the optimization process, and the sample is considered to be harder to classify, that is, the sample is more sensitive to training, and the weight is larger in the optimization process, that is, a larger importance coefficient is obtained. Conversely, if the evidence distance of this data signal is small, it is stated that its importance coefficient should be small. The sample distance coefficient λ t = [0,1] is defined as:
Where d t E [0,1] refers to the Euclidean distance between the predicted result of the classifier and the true class of the data signal. V > 0 is a penalty factor reflecting the extent to which the evidence distance affects the data signal distance coefficient, the greater its value, the greater the influence of the evidence distance on the sample distance coefficient. In this example, it was found from a plurality of experiments that in most cases, it is appropriate to set the value of v to [0.1,0.3 ].
Based on the above, an optimization objective function is defined as:
Wherein, For a vector consisting of weight values for each radar signal attribute in the L t class, lambda t is the euclidean distance coefficient, q is the total number of data signals in the training dataset,The base trust function value for data signal x t and its neighbor data signal x k (i.e., m (|x 1)),Tt is the class truth value for data signal x t).
The radar signal attribute weight value with the class L t can be obtained by minimizing the objective function by using the fmincon optimization method, and the like, and finally, the radar signal attribute weight value in all the classes can be obtained.
In summary, for the data signal y to be classified, firstly, it is assumed that y belongs to class a, then, the weight of the radar signal attribute in class a is obtained by substituting, Γ is used as a training data set, a EKNN classifier is used for classifying to obtain a group of classification results, then, by analogy, it is assumed that the sample belongs to class B, C, … and class M, the operations are repeated to obtain a group of M basic trust function values, finally, several groups of evidence obtained are fused by using classical DS rules to obtain a fused classification result, and finally, the class with the largest confidence value of the data signal is selected as the classification result of the radar radiation source signal according to the result.
According to the method, the distance measurement is improved based on the EKNN classifier by providing the attribute refinement weighting method based on the labels, and aiming at the situations that the attribute weights of all radar signals are the same in use and the contribution degree is the same in the classification process of the conventional classifier, the output result of the classifier is more accurate. In the optimization process, a mode of minimizing errors is adopted, and meanwhile, evidence distances are introduced, so that samples which are not well classified occupy a larger proportion in the optimization, and the optimization accuracy is improved. For the data signals to be classified, the data signals are assumed to belong to n groups of evidence obtained by each class and are fused, so that the robustness of the training effect is greatly improved.
Aiming at the situations that the weights of attribute values of different dimensions in the same category are different and the contribution degree of the attribute of the same dimension to the category is different in different categories, the invention provides an attribute weighting method based on the category. And solving the unknown attribute weight matrix based on the category in an optimized mode by changing the similarity measurement between samples. This makes the classification boundaries between data signals clearer, especially for different classes of data signals at the classification edge, which tend to be misclassified due to closer distances, by changing the attribute weights such that the data signal is closer to the same data signal as the own category, classification accuracy is improved.
When the data signals are identified, the invention performs DS fusion on all obtained evidences by assuming that the data signals belong to each type. The step comprehensively considers the situation that the target sample belongs to each class, and the robustness of the identification result is greatly improved through the DS fusion method.
Another technical scheme of the invention is as follows: a DS rule-based multi-attribute information weighted fusion target recognition device, as shown in FIG. 3, comprises:
The extraction and conversion module 110 is configured to obtain a radar radiation source signal and extract a corresponding data signal; wherein the data signal comprises radar signal attributes of at least two dimensions; a classification module 120, configured to determine a classification result of the data signal based on EKNN classifier, using the data signal as input information; when the radar signals are classified by the EKNN classifier, a weight value is given to the radar signal attribute in the class corresponding to each data signal in the EKNN classifier; the first determining module 130 determines the type of radar radiation source signal according to the classification result of the data signal.
Further, the classification module includes a weight value determination module, the weight value determination module including:
The first construction module is used for constructing a training data set, wherein the training data set comprises a plurality of data signals corresponding to radar radiation source signals of different types and categories corresponding to each data signal; the computing module is used for respectively computing the basic trust function fusion value of each data signal and all the neighbors in the training data set; wherein the basic trust function fusion value comprises a weight value; the second construction module is used for constructing an optimization objective function according to the basic trust function fusion value, solving the optimization objective function and obtaining the Euclidean distance between the data signal and the neighbor of the data signal; and the second determining module is used for determining the weight value of the radar signal attribute in the category corresponding to the data signal according to the Euclidean distance.
It should be noted that, because the content of information interaction and execution process between the modules and the embodiment of the method of the present invention are based on the same concept, specific functions and technical effects thereof may be referred to in the method embodiment section, and details thereof are not repeated herein.
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional modules is illustrated, and in practical application, the above-described functional allocation may be performed by different functional modules according to needs, i.e. the internal structure of the apparatus is divided into different functional modules to perform all or part of the above-described functions. The functional modules in the embodiments may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit, where the integrated units may be implemented in a form of hardware or a form of a software functional unit. In addition, the specific names of the functional modules are only for distinguishing from each other, and are not used for limiting the protection scope of the present invention. The specific working process of the modules in the above system may refer to the corresponding process in the foregoing method embodiment, which is not described herein again.
Another technical scheme of the invention is as follows: as shown in fig. 4, the multi-attribute information weighted fusion target recognition device based on the DS rule includes a memory 31, a processor 32, and a computer program 33 stored in the memory and capable of running on the processor, wherein the processor 32 implements a multi-attribute information weighted fusion target recognition method based on the DS rule according to any one of the above when executing the computer program.
Another technical scheme of the invention is as follows: a computer readable storage medium storing a computer program which when executed by a processor implements a multi-attribute information weighted fusion target recognition method based on a DS rule of any one of the above.
The computer readable medium may include at least: any entity or device capable of carrying computer program code to a camera device/terminal equipment, recording medium, computer Memory, read-Only Memory (ROM), random access Memory (RAM, randomAccess Memory), electrical carrier signals, telecommunications signals, and software distribution media. Such as a U-disk, removable hard disk, magnetic or optical disk, etc. In some jurisdictions, computer readable media may not be electrical carrier signals and telecommunications signals in accordance with legislation and patent practice.
In the foregoing embodiments, the descriptions of the embodiments are emphasized, and in part, not described or illustrated in any particular embodiment, reference is made to the related descriptions of other embodiments.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
Verification example:
As shown in Table 1, the invention is a basic information table of simulation verification data set used in the verification process of the invention, the validity and accuracy of the invention are proved by 6 groups of experimental data, wherein the former three groups adopt UCI data sets, the latter 3 groups are experimental data for simulating various radar radiation source signals, and ten times of experiments are repeated.
Table 1 basic information of data set used in verification procedure
| Data set | Number of sample categories | Number of attributes | Number of samples |
| 1 | 4 | 6 | 871 |
| 2 | 3 | 5 | 151 |
| 3 | 3 | 6 | 106 |
| 4 | 5 | 4 | 600 |
| 5 | 4 | 4 | 400 |
| 6 | 3 | 4 | 320 |
The control methods used in this example were KNN, EKNN and AF. Where KNN, EKNN are classical classifiers and AF is an average method (i.e. when it is assumed that the data signal to be classified is based on the average of the n groups of evidence obtained after belonging to each class, a decision is made). The K value (i.e., the number of neighbors) is chosen from 2 to 10. Table 2 shows experimental results, from which it can be seen that, under most conditions, the multi-attribute information refinement weighted fusion recognition method (DS) provided by the invention has the advantages of effectively improving accuracy and obtaining better classification accuracy compared with other three methods.
TABLE 2 classification results for different methods
As can be seen from table 2, when using the samples of datasets 1 and 3, the classification accuracy of the method is superior to EKNN classifier, KNN classifier and AF classification method, regardless of the number of neighbors selected. When the sample of the dataset 2 is used, the classification accuracy of the method is slightly lower than that of a EKNN classifier only when the number of neighbors is 2 and 3, and the classification accuracy of the rest neighbor numbers is better than that of a EKNN classifier, a KNN classifier and an AF classification method. When using samples of dataset 4, dataset 5 and dataset 6, the classification accuracy of the method is better than EKNN classifier, KNN classifier and AF classification methods.
Claims (6)
1. A DS rule-based multi-attribute information weighted fusion target identification method is characterized by comprising the following steps:
Acquiring radar radiation source signals and extracting corresponding data signals; wherein the data signal comprises radar signal attributes of at least two dimensions;
Determining a classification result of the data signal based on EKNN classifier by taking the data signal as input information; when the EKNN classifier is used for classifying, a weight value is given to the radar signal attribute in the class corresponding to each data signal in the EKNN classifier;
determining the type of the radar radiation source signal according to the classification result of the data signal;
the weight value calculating method comprises the following steps:
Constructing a training data set, wherein the training data set comprises a plurality of data signals corresponding to radar radiation source signals of different types and categories corresponding to each data signal;
Respectively calculating a basic trust function fusion value of each data signal and all neighbors of each data signal in a training data set; wherein the basic trust function fusion value comprises the weight value;
Constructing an optimization objective function according to the basic trust function fusion value, and solving the optimization objective function to obtain Euclidean distance between the data signal and the neighbor of the data signal;
determining a weight value of the radar signal attribute in the category corresponding to the data signal according to the Euclidean distance;
The optimization objective function is as follows:
Wherein, For a vector consisting of weight values for each radar signal attribute in the L t class, lambda t is the euclidean distance coefficient, q is the total number of data signals in the training dataset,For the basic trust function value of data signal x t and its neighbor data signal x k, T t is the class true value of data signal x t;
Lambda t = [0,1] is defined as:
wherein d t E [0,1] refers to Euclidean distance between the prediction result of the classifier and the real class of the data signal, v > 0 is a penalty factor reflecting the influence degree of the evidence distance on the distance coefficient of the data signal, and the larger the value of the penalty factor is, the larger the influence of the evidence distance on the distance coefficient of the sample is;
The calculation method of the Euclidean distance between the data signal and the adjacent neighbor comprises the following steps:
Where d (x, x t) is the Euclidean distance between the data signal x and its neighbor data signal x t, For the weight value of the nth radar signal attribute of category L t, x n is the nth radar signal attribute value of data signal x,Is the nth radar signal attribute value of data signal x t.
2. The method for identifying a multi-attribute information weighted fusion target based on DS rules according to claim 1, wherein obtaining the euclidean distance between the data signal and its neighbors comprises:
determining a basic trust function value of the data signal and each neighbor thereof according to the solution of the optimized objective function;
and determining Euclidean distance between the data signal and each neighbor according to the basic trust function value.
3. The method of claim 2, wherein determining the classification result of the data signal based on EKNN classifier comprises:
Assuming the data signal as any category in a category set, and determining sub-classification results of the data signal by using EKNN classifiers respectively until the categories in the category set are traversed to obtain a plurality of sub-classification results; in the process of determining the sub-classification result of the data signal by using EKNN classifier, a weight value is given to each radar signal attribute of the class;
And fusing the sub-classification results to determine the classification result of the data signal.
4. The utility model provides a multi-attribute information weighted fusion target identification device based on DS rule which characterized in that includes:
the extraction and conversion module is used for acquiring radar radiation source signals and extracting corresponding data signals; wherein the data signal comprises radar signal attributes of at least two dimensions;
A classification module for taking the data signal as input information, and determining a classification result of the data signal based on a EKNN classifier; when the EKNN classifier is used for classifying, a weight value is given to the radar signal attribute in the class corresponding to each data signal in the EKNN classifier;
the first determining module is used for determining the type of the radar radiation source signal according to the classification result of the data signal; the classification module comprises a weight value determination module, and the weight value determination module comprises:
The first construction module is used for constructing a training data set, wherein the training data set comprises a plurality of data signals corresponding to radar radiation source signals of different types and categories corresponding to each data signal;
the computing module is used for respectively computing the basic trust function fusion value of each data signal and all the adjacent neighbors in the training data set; wherein the basic trust function fusion value comprises the weight value;
The second construction module is used for constructing an optimization objective function according to the basic trust function fusion value, solving the optimization objective function and obtaining Euclidean distance between the data signal and the neighbor of the data signal;
the second determining module is used for determining a weight value of the radar signal attribute in the category corresponding to the data signal according to the Euclidean distance;
The optimization objective function is as follows:
Wherein, For a vector consisting of weight values for each radar signal attribute in the L t class, lambda t is the euclidean distance coefficient, q is the total number of data signals in the training dataset,For the basic trust function value of data signal x t and its neighbor data signal x k, T t is the class true value of data signal x t;
Lambda t = [0,1] is defined as:
wherein d t E [0,1] refers to Euclidean distance between the prediction result of the classifier and the real class of the data signal, v > 0 is a penalty factor reflecting the influence degree of the evidence distance on the distance coefficient of the data signal, and the larger the value of the penalty factor is, the larger the influence of the evidence distance on the distance coefficient of the sample is;
The calculation method of the Euclidean distance between the data signal and the adjacent neighbor comprises the following steps:
Where d (x, x t) is the Euclidean distance between the data signal x and its neighbor data signal x t, For the weight value of the nth radar signal attribute of category L t, x n is the nth radar signal attribute value of data signal x,Is the nth radar signal attribute value of data signal x t.
5. A multi-attribute information weighted fusion target recognition device based on DS rules, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements a multi-attribute information weighted fusion target recognition method based on DS rules as claimed in any one of claims 1 to 3 when executing the computer program.
6. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements a multi-attribute information weighted fusion target recognition method based on DS rules as claimed in any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011168979.5A CN112307935B (en) | 2020-10-28 | 2020-10-28 | Multi-attribute information weighted fusion target identification method based on DS rule |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011168979.5A CN112307935B (en) | 2020-10-28 | 2020-10-28 | Multi-attribute information weighted fusion target identification method based on DS rule |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112307935A CN112307935A (en) | 2021-02-02 |
| CN112307935B true CN112307935B (en) | 2024-10-15 |
Family
ID=74331624
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011168979.5A Active CN112307935B (en) | 2020-10-28 | 2020-10-28 | Multi-attribute information weighted fusion target identification method based on DS rule |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112307935B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118604753B (en) * | 2024-05-23 | 2024-11-19 | 中国人民解放军军事科学院系统工程研究院 | A signal-level radar effectiveness evaluation method and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107341447A (en) * | 2017-06-13 | 2017-11-10 | 华南理工大学 | A kind of face verification mechanism based on depth convolutional neural networks and evidence k nearest neighbor |
| CN110084263A (en) * | 2019-03-05 | 2019-08-02 | 西北工业大学 | A kind of more frame isomeric data fusion identification methods based on trust |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2013003905A1 (en) * | 2011-07-06 | 2013-01-10 | Fred Bergman Healthcare Pty Ltd | Improvements relating to event detection algorithms |
| US9299010B2 (en) * | 2014-06-03 | 2016-03-29 | Raytheon Company | Data fusion analysis for maritime automatic target recognition |
| US11087184B2 (en) * | 2018-09-25 | 2021-08-10 | Nec Corporation | Network reparameterization for new class categorization |
| CN110109095B (en) * | 2019-04-30 | 2022-10-28 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Target feature assisted multi-source data association method |
| CN111126504A (en) * | 2019-12-27 | 2020-05-08 | 西北工业大学 | Multi-source incomplete information fusion image target classification method |
-
2020
- 2020-10-28 CN CN202011168979.5A patent/CN112307935B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107341447A (en) * | 2017-06-13 | 2017-11-10 | 华南理工大学 | A kind of face verification mechanism based on depth convolutional neural networks and evidence k nearest neighbor |
| CN110084263A (en) * | 2019-03-05 | 2019-08-02 | 西北工业大学 | A kind of more frame isomeric data fusion identification methods based on trust |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112307935A (en) | 2021-02-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Pei et al. | SAR automatic target recognition based on multiview deep learning framework | |
| Zhang et al. | Local density adaptive similarity measurement for spectral clustering | |
| US10908252B2 (en) | Method and device for identifying radar emission modes | |
| CN110647912A (en) | Fine-grained image recognition method and device, computer equipment and storage medium | |
| CN114594440B (en) | Radar high-resolution one-dimensional range profile target recognition method and system based on dual parallel networks | |
| CN112437053B (en) | Intrusion detection method and device | |
| CN113435108B (en) | Battlefield target grouping method based on improved whale optimization algorithm | |
| CN111126504A (en) | Multi-source incomplete information fusion image target classification method | |
| CN106599927B (en) | The Target cluster dividing method divided based on Fuzzy ART | |
| CN116165611A (en) | Precise and intelligent identification method, system, equipment and terminal of radar radiation source model | |
| CN114117141A (en) | Self-adaptive density clustering method, storage medium and system | |
| CN112307935B (en) | Multi-attribute information weighted fusion target identification method based on DS rule | |
| CN119961758B (en) | Intelligent recognition method of low-altitude small and micro UAV targets based on track characteristics | |
| US6754390B2 (en) | Fusing outputs from multiple detection/classification schemes | |
| CN111783088B (en) | Malicious code family clustering method and device and computer equipment | |
| CN114067224A (en) | Unmanned aerial vehicle cluster target number detection method based on multi-sensor data fusion | |
| CN118365863A (en) | Radiation source individual open set identification method based on similarity weighting | |
| CN114445700B (en) | Evidence fusion target identification method for unbalanced SAR image data | |
| CN108106500B (en) | Missile target type identification method based on multiple sensors | |
| CN112115768A (en) | Radar radiation source identification method facing complex electromagnetic environment | |
| Zhu et al. | Research on online learning of radar emitter recognition based on Hull Vector | |
| CN116166965B (en) | Radar radiation source identification method based on feature fusion and integrated learning | |
| CN113269217A (en) | Radar target classification method based on Fisher criterion | |
| CN113447907B (en) | Radar sorting system control method and radar sorting system | |
| Liu et al. | Synthetic aperture radar image target recognition based on improved fusion of R-FCN and SRC |
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 |