CN113608172A - Airborne multifunctional radar working mode identification method based on improved K nearest neighbor - Google Patents
Airborne multifunctional radar working mode identification method based on improved K nearest neighbor Download PDFInfo
- Publication number
- CN113608172A CN113608172A CN202110700893.0A CN202110700893A CN113608172A CN 113608172 A CN113608172 A CN 113608172A CN 202110700893 A CN202110700893 A CN 202110700893A CN 113608172 A CN113608172 A CN 113608172A
- Authority
- CN
- China
- Prior art keywords
- sample
- training set
- samples
- target
- test
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 111
- 238000012360 testing method Methods 0.000 claims abstract description 56
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 41
- 238000004458 analytical method Methods 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 239000002131 composite material Substances 0.000 claims abstract description 7
- 239000000523 sample Substances 0.000 claims description 123
- 239000011159 matrix material Substances 0.000 claims description 31
- 239000006185 dispersion Substances 0.000 claims description 17
- 238000012512 characterization method Methods 0.000 claims description 14
- 239000013598 vector Substances 0.000 claims description 13
- 239000013074 reference sample Substances 0.000 claims description 8
- 238000012935 Averaging Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000000354 decomposition reaction Methods 0.000 claims description 3
- 238000000528 statistical test Methods 0.000 claims description 3
- 230000000875 corresponding effect Effects 0.000 description 12
- 238000000513 principal component analysis Methods 0.000 description 8
- 238000012545 processing Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000007635 classification algorithm Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/021—Auxiliary means for detecting or identifying radar signals or the like, e.g. radar jamming signals
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Networks & Wireless Communication (AREA)
- General Engineering & Computer Science (AREA)
- Electromagnetism (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses an airborne multifunctional radar working mode identification method based on improved K nearest neighbor, which comprises the steps of acquiring a data set of multifunctional radar signals in a composite scene; preprocessing the data set to obtain a multi-dimensional characteristic representation of each sample in the data set; carrying out one-bit encoding on each sample to obtain a sample label; further obtaining a label sample set, and then randomly selecting and forming a training set and a testing set; according to the inter-class boundary, effectively eliminating the fuzzy boundary samples in the training set by using a 1-NN algorithm to obtain a target training set; and then PCA principal component contribution rate analysis is carried out, the influence of redundant features on classification recognition is reduced, and a target training set after dimension reduction is obtained. Repeatedly training a preset k nearest neighbor algorithm model, and determining an optimal value of k; and identifying the k-nearest neighbor algorithm model with k determined to be an optimal value by using a test set test. The method can reduce the time for classification and identification, and has better robustness and generalization.
Description
Technical Field
The invention belongs to the technical field of electronic warfare signal processing, and particularly relates to an airborne multifunctional radar working mode identification method based on improved K nearest neighbor.
Background
With the continuous development and innovation of equipment technology, in the present electromagnetic environment, the radar of a single system can not meet the combat requirement in a complex environment for a long time, so that the multifunctional radar becomes more and more important and is more and more widely applied. In electronic warfare, the multifunctional radar is in what working mode at present, what kind of fighting intention will be at the next moment, and plays an important role in electronic information reconnaissance, electronic support and threat warning system.
The working modes of a typical airborne multifunctional radar can be flexibly switched according to tactical requirements, the feeding phase of each radiation unit is controlled by a computer in an electric scanning mode, flexible beam emission and rapid data processing can be realized, and in addition, due to the multifunctional, multi-target processing and high self-adaption capability of the airborne multifunctional radar, the airborne multifunctional radar has multiple working modes and different threat levels, radar signal parameters in each working mode are different, how to classify and identify the working modes by using the rules and the characteristics of the signal parameters has important significance on the analysis work of electronic countermeasure information, and therefore, the correct identification of the working modes of the airborne multifunctional radar is of great importance.
Through carrying out the analysis to radar scanning envelope in 2016 people Lijunjiang etc., because there is the difference in the scanning characteristic in the actual airborne multi-functional radar with traditional PD radar, the recognition method is not accurate the reaching identification effect. Chen tour et al propose a particle swarm optimization algorithm self-adaptive optimization DPNN network, and use the trained network to identify the extracted radar phrases, but the identification accuracy under low signal-to-noise ratio needs to be improved.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides an airborne multifunctional radar working mode identification method based on improved K nearest neighbor. The technical problem to be solved by the invention is realized by the following technical scheme:
the invention provides an improved K-nearest neighbor-based airborne multifunctional radar working mode identification method, which comprises the following steps:
acquiring a data set of a multifunctional radar signal in a composite scene;
the data set comprises signals under a plurality of working modes, the signal-to-noise ratio point of each signal comprises a plurality of signals, and a plurality of samples exist in the signals under each working mode under different signal-to-noise ratios;
preprocessing the data set to obtain a multi-dimensional characteristic representation of each sample in the data set;
for each sample in the data set, carrying out one-bit encoding on the sample to obtain a sample label;
forming a label sample by using the sample label of each sample and the multi-dimensional characteristic characterization to obtain a label sample set;
randomly selecting the label sample set to form a training set and a testing set;
according to the inter-class boundary, using a 1-NN algorithm to remove the fuzzy boundary samples in the training set to obtain a target training set;
carrying out PCA principal component contribution rate analysis on the target samples in the target training set to obtain a target training set after dimensionality reduction;
repeatedly training a preset k nearest neighbor algorithm model by using a target training set, and determining an optimal value of k;
testing the k neighbor algorithm model with the k as the optimal value by using the test set, and determining the k neighbor algorithm model with the k as the optimal value as an identification model when the test accuracy reaches a set accuracy threshold;
the multifunctional radar signal is identified using the identification model.
Optionally, before performing one-bit encoding on each sample in the data set to obtain a sample tag, the identification method further includes:
the characterization of each sample in each dimension is normalized so that the characterization maps to a fixed interval.
Optionally, the removing, according to the inter-class boundary, the fuzzy boundary samples in the training set by using a 1-NN algorithm to obtain the target training set includes:
randomly dividing the training set into a first training set and a second training set;
taking the first training set as a reference sample set, enabling a 1-NN algorithm to classify a second training set, comparing a classification result of the second training set with a real class of the second training set, and eliminating samples with wrong classification in the second training set to obtain an eliminated second training set;
taking the second training set after being removed as a reference sample set, enabling a 1-NN algorithm to classify the first training set, comparing a classification result of the first training set with a real class of the first training set, and removing a sample with a wrong classification in the first training set to obtain a first training set after being removed;
and combining the first training set and the second training set into a target training set.
Optionally, the performing PCA principal component contribution rate analysis on the target samples in the target training set to obtain the reduced-dimension target training set includes:
centralizing each sample in the reduced target training set to obtain the mean value of each dimension characteristic of each sample;
calculating a covariance matrix of each sample based on the mean of each dimension feature;
each covariance matrix comprises the features of unknown values and feature vectors corresponding to the features;
performing eigenvalue decomposition on the covariance matrix, and solving an eigenvalue of the covariance matrix and a corresponding eigenvector;
sorting the eigenvalues from small to large, and selecting eigenvectors corresponding to the largest k eigenvalues as row vectors to form an eigenvector matrix;
and forming a target training set after dimensionality reduction by using the feature vector matrix corresponding to each sample.
Wherein, the samples after dimension characteristic averaging are represented as:
wherein,
representing the samples, s, after dimensional feature averagingi＝(Ci,Di,Pi,Wi,Ui,Mi,Ri,Ni)，siRepresenting a multi-dimensional characterization of each sample, CiFor smoothness, DiAs a dispersion, PiFor pulse repetition frequency, WiIs a pulse width, UiFor duty ratio, MiOf the intra-pulse modulation type, RiFor reviewing information, NiFor the number of pulses in the target dwell time, m represents the number of samples and i represents the sample number.
Optionally, the preprocessing the data set to obtain a multi-dimensional feature characterization of each sample in the data set includes:
according to the pulse amplitude sequence a of the sampleiCalculating the intermediate variable ci：
According to the intermediate variable ciCalculating smoothness C;
calculating the dispersion of each sample by using a dispersion calculation formula;
respectively determining smoothness, dispersion, pulse repetition frequency, pulse width, duty ratio, intra-pulse modulation type, retrospective information and pulse number in target residence time as the characteristics of the samples in each dimension, and obtaining the multi-dimensional characteristic representation of each sample in the data set;
wherein,
n is a pulse amplitude sequence aiThe total number of medium pulses;
the dispersion is calculated by the formula
Is a pulse amplitude sequence a according to the sampleiThe average value is obtained, where N is the pulse amplitude sequence aiThe total number of the medium pulses,
δais a pulse amplitude sequence a according to the sampleiThe standard deviation of the obtained standard deviation is calculated,
the covariance matrix calculation formula is as follows:
wherein,
x represents a dimension characteristic a or b, y represents a dimension characteristic a or b, m represents the number of samples, and i represents a sample number.
Wherein the feature value is represented by λ, the feature vector is represented by x,
where E is the identity matrix, a is the covariance matrix of the sample, λ is the eigenvalue, and x is the corresponding eigenvector.
Optionally, the step of testing the K-nearest neighbor algorithm model with the K being the optimal value by using the test set, and when the test accuracy reaches a set accuracy threshold, determining the K-nearest neighbor algorithm model with the K being the optimal value as the identification model includes:
calculating a Euclidean distance between each sample in the test set and each sample in the training set;
according to the Euclidean distance, using a majority voting rule to classify and judge k samples in a test set, and judging the class of each sample into the sample class with the largest quantity;
according to the comparison result of the sample category and the real category judged by each sample in the test set, counting the test accuracy;
and when the test accuracy reaches a set accuracy threshold, determining the k-nearest neighbor algorithm model with k as an optimal value as an identification model.
Wherein, the statistical formula of the statistical test accuracy is as follows:
wherein acc is the test accuracy, IsTo identify the correct number of test set samples, TsIs the total number of samples in the test set.
1. The invention provides an airborne multifunctional radar working mode identification method based on improved K nearest neighbor, which is characterized in that on the basis of an airborne air-air scene, function alternation under the air-air and air-ground composite scene is realized simultaneously through waveform and beam agility, and the method can adapt to various working modes under different battle situation scenes;
2. the airborne multifunctional radar working mode identification method based on the improved K nearest neighbor adopts the 1-NN algorithm, and eliminates fuzzy boundary samples from a training set by focusing on the boundaries among classes of all working modes, so that the boundaries are clearer, and the classification efficiency of the algorithm can be improved;
3. the airborne multifunctional radar working mode identification method based on the improved K nearest neighbor introduces a PCA dimension reduction strategy, eliminates redundant features which reduce identification accuracy, optimizes classification accuracy and reduces operation complexity;
4. the invention provides an improved K-nearest neighbor-based airborne multifunctional radar working mode identification method, aiming at the characteristic that a data set of an airborne multifunctional radar has relatively more characteristic attributes, a K-nearest neighbor classification algorithm model is introduced for classification and identification. Compared with the prior art, the method can continuously learn new knowledge from new sample data to perform identification and classification, and has better adaptability to the classification and the identification of the airborne multifunctional radar.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
Fig. 1 is a flow chart of an airborne multifunctional radar working mode identification method based on improved K-nearest neighbor provided by the invention.
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.
As shown in fig. 1, the method for identifying the operating mode of the airborne multifunctional radar based on the improved K-nearest neighbor provided by the invention comprises the following steps:
s1, acquiring a data set of the multifunctional radar signal in the composite scene;
the data set comprises signals under a plurality of working modes, the signal-to-noise ratio point of each signal comprises a plurality of signal-to-noise ratio points, and a plurality of samples exist in the signal under each working mode under different signal-to-noise ratios;
the method comprises the steps of obtaining a typical working mode data set of the airborne multifunctional radar in an air-air and air-ground composite scene, wherein signals of the data set comprise ten typical working modes. For an air-ground scene, the data set comprises a moving target detection mode signal, a moving target tracking mode signal and a synthetic aperture radar mode signal; for the empty-empty scene, the data set comprises a speed search mode signal, a high repetition frequency ranging-while-searching mode signal, a medium repetition frequency ranging-while-searching mode signal, a search-while-tracking mode signal, a search-plus-tracking mode signal, a single-target tracking mode signal and a situation awareness mode signal. Wherein, each signal is from 5dB to 20dB, the interval is 5dB, and the total number of the signal-to-noise ratio points is 4, and the number of samples of each signal under different signal-to-noise ratios is 1000.
The carrier frequency of 10 multifunctional radar signals is set to be 1GHz, and the sampling frequency is set to be 2.4 GHz; setting the number of pulses in the target residence time of the moving target detection mode signal to be 20-550, setting the pulse width to be 2-60 us, and setting the duty ratio to be 0.1-25; setting the number of pulses in the target residence time of the moving target tracking mode signal to be 20-1000, setting the pulse width to be 2-60 us, and setting the duty ratio to be 0.1-25; setting the number of pulses in the target residence time of the synthetic aperture radar mode signal to be 150-100 k, setting the pulse width to be 3-60 us, and setting the duty ratio to be 1-25; the pulse number in the target residence time of the speed search mode signal is set to be 1500-6000, the pulse width is set to be 1-3 us, and the duty ratio is set to be 10-30; the pulse number in the high repetition frequency ranging and searching mode signal target residence time is set to be 250-2000, the pulse width is set to be 1-3 us, and the duty ratio is set to be 10-25; the pulse number in the target residence time of the medium repetition frequency ranging and searching mode signal is set to be 250-2000, the pulse width is set to be 2-4 us, and the duty ratio is set to be 10-25; setting 16-128 pulse numbers in the target residence time of the searching and tracking mode signal, setting the pulse width to be 0.81-4 us, and setting the duty ratio to be 10-30; setting 16-128 pulse numbers in the target residence time of the search and tracking mode signal, setting the pulse width to be 0.28-4 us, and setting the duty ratio to be 10-30; the pulse number in the single target tracking mode signal target residence time is set to 20000, the pulse width is set to 1-3 us, and the duty ratio is set to 0.1-5; the pulse number of the situation perception mode signal in the target residence time is set to be 1000-8000, the pulse width is set to be 1-3 us, and the duty ratio is set to be 10-30.
S2, preprocessing the data set to obtain the multi-dimensional characteristic representation of each sample in the data set;
identification of operating mode for airborne multifunction radar, slave dataThe formalized expression of the working mode can be obtained in a centralized manner, each sample is mainly characterized by eight-dimensional information such as smoothness, dispersion, pulse repetition frequency, pulse width, duty ratio, intra-pulse modulation type, back illumination and pulse number in target residence time, and the expression S is one sample and can be expressed as follows: si＝(Ci,Di,Pi,Wi,Ui,Mi,Ri,Ni) In which C isiFor smoothness, DiAs a dispersion, PiFor pulse repetition frequency, WiIs a pulse width, UiFor duty ratio, MiOf the intra-pulse modulation type, RiFor reviewing information, NiThe number of pulses in the target dwell time.
Before a sample label is obtained by encoding the sample by one bit for each sample in the data set, the present invention needs to normalize the feature representation of each sample in each dimension so that the feature representation is mapped in a fixed interval.
In the radar information actually detected, some characteristic parameters are fixed values, some characteristic parameters are a range, and the characteristic parameters are often not on the same data level. Performing maximum-minimum normalization processing on all characteristic parameters, and mapping the parameter values to [0,1]Interval, which is abstracted as i values for each attribute, assuming it has i values in the sample spacevThe normalization process is as follows:
as an optional embodiment of the present invention, the preprocessing the data set to obtain the multi-dimensional feature characterization of each sample in the data set includes:
s21, according to the pulse amplitude sequence a of the samplesiCalculating the intermediate variable ci：
S22, according to the intermediate variable ciCalculating smoothness C;
s23, calculating the dispersion of each sample by using a dispersion calculation formula;
s24, respectively determining smoothness, dispersion, pulse repetition frequency, pulse width, duty ratio, intra-pulse modulation type, retroillumination information and pulse number in target residence time as the characteristics of the samples in each dimension, and obtaining the multi-dimensional characteristic representation of each sample in the data set;
wherein,
n is a pulse amplitude sequence aiThe total number of medium pulses;
the dispersion is calculated by the formula
Is a pulse amplitude sequence a according to the sampleiThe average value is obtained, where N is the pulse amplitude sequence aiThe total number of the medium pulses,
δais a pulse amplitude sequence a according to the sampleiThe standard deviation of the obtained standard deviation is calculated,
s3, for each sample in the data set, carrying out one-bit encoding on the sample to obtain a sample label;
one bit encoding is performed on each sample of the data set, i.e. a new set of each sample can be expressed as:
Si＝(Ci,Di,Pi,Wi,Ui,Mi,Ri,Ni,labeli)
wherein, labeli＝{0,1,2,3,4,5,6,7,8,9}。
S4, forming the label sample by the sample label of each sample and the multi-dimensional characteristic characterization to obtain a label sample set;
forming the label sample by using the sample label of each sample and the multi-dimensional characteristic characterization, and obtaining a label sample set with the length of 9, wherein the label sample set is used for subsequent work pattern recognition, and the dimension is 10000 × 9;
s5, randomly selecting and forming a training set and a testing set from the label sample set;
in the data set of 4 signal-to-noise ratios of 10 typical working mode signals obtained by the invention, in 40000 samples in total, 8000 samples can be randomly extracted from each signal-to-noise ratio sample set to serve as a training set, 1000 samples serve as a verification set, and 1000 samples serve as a test set.
S6, removing fuzzy boundary samples in the training set by using a 1-NN algorithm according to the inter-class boundary to obtain a target training set;
s7, carrying out PCA principal component contribution rate analysis on the target samples in the target training set to obtain a target training set after dimensionality reduction;
for the multifunctional radar working mode, the increase of the characteristics of the data set is not positively correlated with the recognition rate, the recognition accuracy rate is reduced by partial redundant characteristics, and the input data sample set is Z ═ S1,S2,…,SmAnd K is the dimension of the low-dimensional space.
S8, repeatedly training a preset k nearest neighbor algorithm model by using a target training set, and determining an optimal value of k;
and for the airborne multifunctional radar working mode signal, eight-dimensional characteristic representation is adopted, a PCA (principal component analysis) dimension reduction strategy is adopted, the eight-dimensional characteristics are compared with the identification accuracy of the data set reduced to different dimensions (one-dimensional-seven-dimensional), the average value of results of ten tests is calculated, and the identification accuracy is compared. The optimum value of k is obtained.
According to experimental verification, when the low-dimensional space dimension K is 7, the identification accuracy is highest, namely the low-dimensional space dimension K is selected to be 7.
S9, testing the k neighbor algorithm model with the k as the optimal value by using a test set, and determining the k neighbor algorithm model with the k as the optimal value as an identification model when the test accuracy reaches a set accuracy threshold;
and S10, using the identification model to identify the multifunctional radar signal.
The invention discloses an airborne multifunctional radar working mode identification method based on improved K nearest neighbor, which comprises the steps of acquiring a data set of multifunctional radar signals in a composite scene; preprocessing the data set to obtain a multi-dimensional characteristic representation of each sample in the data set; carrying out one-bit encoding on each sample to obtain a sample label; further obtaining a label sample set, and then randomly selecting and forming a training set and a testing set; according to the inter-class boundary, effectively eliminating the fuzzy boundary samples in the training set by using a 1-NN algorithm to obtain a target training set; and then PCA principal component contribution rate analysis is carried out, the influence of redundant features on classification recognition is reduced, and a target training set after dimension reduction is obtained. Repeatedly training a preset k nearest neighbor algorithm model, and determining an optimal value of k; and identifying the k-nearest neighbor algorithm model with k determined to be an optimal value by using a test set test. The method can reduce the time for classification and identification, and has better robustness and generalization.
As an optional implementation manner of the present invention, according to the inter-class boundary, using a 1-NN algorithm to remove the fuzzy boundary samples in the training set, and obtaining the target training set includes:
s61, randomly dividing the training set into a first training set and a second training set;
s62, using the first training set as a reference sample set, enabling a 1-NN algorithm to classify the second training set, comparing the classification result of the second training set with the real class of the second training set, and eliminating the sample with the wrong classification in the second training set to obtain the second training set after elimination;
s63, taking the second training set after being removed as a reference sample set, enabling the 1-NN algorithm to classify the first training set, comparing the classification result of the first training set with the real class of the first training set, removing the sample with the wrong classification in the first training set, and obtaining the first training set after being removed;
and S64, forming the first training set and the second training set into a target training set.
The process of obtaining the target training set is as follows: training set data are randomly scrambled and then are equally divided into two parts ZlAnd Zr(ii) a First of all with ZlFor reference sample set ZrUsing the 1-NN algorithm, apply ZrComparing the classification result with the real classification, eliminating the sample set with wrong classification, and expressing the eliminated data set as Z'r(ii) a And then with Z'rFor reference sample set ZlAdopting a 1-NN algorithm, and expressing the data set after rejecting the corresponding classification error sample set as Z'l(ii) a Obtaining a target training set of Z ═ Z'l+Z'r。
As an optional implementation manner of the present invention, performing PCA principal component contribution rate analysis on a target sample in a target training set to obtain a reduced-dimension target training set includes:
s71, centralizing each sample in the reduced-dimension target training set to obtain the mean value of each dimension characteristic of each sample;
wherein, the samples after dimension characteristic averaging are represented as:
wherein,
denotes a sample after dimension feature averaging, si ═ Ci,Di,Pi,Wi,Ui,Mi,Ri,Ni)，siRepresenting a multi-dimensional characterization of each sample, CiFor smoothness, DiAs a dispersion, PiFor pulse repetition frequency, WiIs a pulse width, UiFor duty ratio, MiOf the intra-pulse modulation type, RiFor reviewing information, NiFor the number of pulses in the target dwell time, m represents the number of samples and i represents the sample number.
S72, calculating a covariance matrix of each sample based on the mean value of each dimension characteristic;
each covariance matrix comprises the features of unknown values and feature vectors corresponding to the features; the covariance matrix calculation formula is:
wherein,
x represents a dimension characteristic a or b, y represents a dimension characteristic a or b, m represents the number of samples, and i represents a sample number. From the above matrix, it can be seen that the two dimensional feature variables are the variances of the matrix diagonal, and the other elements are the covariances of the variables a and b, which are unified into one matrix.
S73, performing eigenvalue decomposition on the covariance matrix, and solving the eigenvalue of the covariance matrix and the corresponding eigenvector;
the feature value is denoted as λ, the feature vector is denoted as x,
where E is the identity matrix, a is the covariance matrix of the sample, λ is the eigenvalue, and x is the corresponding eigenvector.
S74, sorting the eigenvalues from small to large, and selecting the eigenvectors corresponding to the largest k eigenvalues as row vectors to form an eigenvector matrix;
and sorting the eigenvalues from small to large, and selecting the eigenvectors corresponding to the largest K eigenvalues as row vectors to form an eigenvector matrix respectively to obtain the sample data after dimension reduction. By searching different low-dimensional space dimensions, the classification and identification efficiency is highest when the dimension is 7.
And S75, forming a target training set after dimension reduction by using the feature vector matrix corresponding to each sample.
As an optional embodiment of the present invention, the testing the K-nearest neighbor algorithm model with the optimal value K using the test set, and when the test accuracy reaches the set accuracy threshold, determining the K-nearest neighbor algorithm model with the optimal value K as the identification model includes:
s91, calculating the Euclidean distance between each sample in the test set and each sample in the training set;
s92, classifying and judging k samples in the test set by using a majority voting rule according to the Euclidean distance, and judging the class of each sample as the sample class with the largest quantity;
s93, counting the test accuracy according to the comparison result of the sample type and the real type judged by each sample in the test set;
the statistical formula for the statistical test accuracy is as follows:
wherein acc is the test accuracy, IsTo identify the correct number of test set samples, TsIs the total number of samples in the test set.
And S94, when the test accuracy reaches the set accuracy threshold, determining the k-nearest neighbor algorithm model with k as an optimal value as the identification model.
Wherein, the accurate threshold value can be obtained through experiments.
And when all the samples in the test set are classified and identified, comparing the real classes of the samples, and calculating the classification and identification accuracy of the test set. Under the condition that the signal-to-noise ratio is 10dB, the classification and identification accuracy rate is higher than 95% through multiple test experiments.
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 (10)
1. An airborne multifunctional radar working mode identification method based on improved K nearest neighbor is characterized by comprising the following steps:
acquiring a data set of a multifunctional radar signal in a composite scene;
the data set comprises signals under a plurality of working modes, the signal-to-noise ratio point of each signal comprises a plurality of signals, and a plurality of samples exist in the signals under each working mode under different signal-to-noise ratios;
preprocessing the data set to obtain a multi-dimensional characteristic representation of each sample in the data set;
for each sample in the data set, carrying out one-bit encoding on the sample to obtain a sample label;
forming a label sample by using the sample label of each sample and the multi-dimensional characteristic characterization to obtain a label sample set;
randomly selecting the label sample set to form a training set and a testing set;
according to the inter-class boundary, using a 1-NN algorithm to remove the fuzzy boundary samples in the training set to obtain a target training set;
carrying out PCA principal component contribution rate analysis on the target samples in the target training set to obtain a target training set after dimensionality reduction;
repeatedly training a preset k nearest neighbor algorithm model by using a target training set, and determining an optimal value of k;
testing the k neighbor algorithm model with the k as the optimal value by using the test set, and determining the k neighbor algorithm model with the k as the optimal value as an identification model when the test accuracy reaches a set accuracy threshold;
the multifunctional radar signal is identified using the identification model.
2. The method of claim 1, wherein before encoding each sample in the data set with one bit to obtain a sample tag, the method further comprises:
the characterization of each sample in each dimension is normalized so that the characterization maps to a fixed interval.
3. The method according to claim 1, wherein the using a 1-NN algorithm to remove the fuzzy boundary samples in the training set according to the inter-class boundary to obtain the target training set comprises:
randomly dividing the training set into a first training set and a second training set;
taking the first training set as a reference sample set, enabling a 1-NN algorithm to classify a second training set, comparing a classification result of the second training set with a real class of the second training set, and eliminating samples with wrong classification in the second training set to obtain an eliminated second training set;
taking the second training set after being removed as a reference sample set, enabling a 1-NN algorithm to classify the first training set, comparing a classification result of the first training set with a real class of the first training set, and removing a sample with a wrong classification in the first training set to obtain a first training set after being removed;
and combining the first training set and the second training set into a target training set.
4. The identification method according to claim 1, wherein the performing PCA principal component contribution rate analysis on the target samples in the target training set to obtain the reduced-dimension target training set comprises:
centralizing each sample in the reduced target training set to obtain the mean value of each dimension characteristic of each sample;
calculating a covariance matrix of each sample based on the mean of each dimension feature;
each covariance matrix comprises the features of unknown values and feature vectors corresponding to the features;
performing eigenvalue decomposition on the covariance matrix, and solving an eigenvalue of the covariance matrix and a corresponding eigenvector;
sorting the eigenvalues from small to large, and selecting eigenvectors corresponding to the largest k eigenvalues as row vectors to form an eigenvector matrix;
and forming a target training set after dimensionality reduction by using the feature vector matrix corresponding to each sample.
5. The identification method according to claim 4, wherein the dimension-averaged sample is represented by:
wherein,
representing the samples, s, after dimensional feature averagingi＝(Ci,Di,Pi,Wi,Ui,Mi,Ri,Ni)，siRepresenting a multi-dimensional characterization of each sample, CiFor smoothness, DiAs a dispersion, PiFor pulse repetition frequency, WiIs a pulse width, UiFor duty ratio, MiOf the intra-pulse modulation type, RiFor reviewing information, NiFor the number of pulses in the target dwell time, m represents the number of samples and i represents the sample number.
6. The identification method of claim 5, wherein the preprocessing the data set to obtain the multi-dimensional feature characterization of each sample in the data set comprises:
according to the pulse amplitude sequence a of the sampleiCalculating the intermediate variable ci：
According to the intermediate variable ciCalculating smoothness C;
calculating the dispersion of each sample by using a dispersion calculation formula;
respectively determining smoothness, dispersion, pulse repetition frequency, pulse width, duty ratio, intra-pulse modulation type, retrospective information and pulse number in target residence time as the characteristics of the samples in each dimension, and obtaining the multi-dimensional characteristic representation of each sample in the data set;
wherein,
n is a pulse amplitude sequence aiThe total number of medium pulses;
the dispersion is calculated by the formula
Is a pulse amplitude sequence a according to the sampleiThe average value is obtained, where N is the pulse amplitude sequence aiThe total number of the medium pulses,
δais a pulse amplitude sequence a according to the sampleiThe standard deviation of the obtained standard deviation is calculated,
7. the identification method according to claim 4, wherein the covariance matrix is calculated as:
wherein,
x represents a dimension characteristic a or b, y represents a dimension characteristic a or b, m represents the number of samples, and i represents a sample number.
8. The identification method according to claim 4, characterized in that said feature value is denoted λ, the feature vector is denoted x,
where E is the identity matrix, a is the covariance matrix of the sample, λ is the eigenvalue, and x is the corresponding eigenvector.
9. The identification method according to claim 1, wherein the step of testing the K-nearest neighbor algorithm model with the K being the optimal value by using the test set, and when the test accuracy reaches a set accuracy threshold, the step of determining the K-nearest neighbor algorithm model with the K being the optimal value as the identification model comprises the steps of:
calculating a Euclidean distance between each sample in the test set and each sample in the training set;
according to the Euclidean distance, using a majority voting rule to classify and judge k samples in a test set, and judging the class of each sample into the sample class with the largest quantity;
according to the comparison result of the sample category and the real category judged by each sample in the test set, counting the test accuracy;
and when the test accuracy reaches a set accuracy threshold, determining the k-nearest neighbor algorithm model with k as an optimal value as an identification model.
10. The identification method of claim 9, wherein the statistical formula for the statistical test accuracy is:
wherein acc is the test accuracy, IsTo identify the correct number of test set samples, TsIs the total number of samples in the test set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110700893.0A CN113608172B (en) | 2021-06-23 | 2021-06-23 | Airborne multifunctional radar working mode identification method based on improved K nearest neighbor |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110700893.0A CN113608172B (en) | 2021-06-23 | 2021-06-23 | Airborne multifunctional radar working mode identification method based on improved K nearest neighbor |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113608172A true CN113608172A (en) | 2021-11-05 |
| CN113608172B CN113608172B (en) | 2023-12-29 |
Family
ID=78303649
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110700893.0A Active CN113608172B (en) | 2021-06-23 | 2021-06-23 | Airborne multifunctional radar working mode identification method based on improved K nearest neighbor |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113608172B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114463294A (en) * | 2022-01-24 | 2022-05-10 | 北京唯迈医疗设备有限公司 | Automatic metering control method and system for X-ray machine |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101894269A (en) * | 2010-07-16 | 2010-11-24 | 西安电子科技大学 | Multi-classifier system-based synthetic aperture radar automatic target recognition method |
| CN107194433A (en) * | 2017-06-14 | 2017-09-22 | 电子科技大学 | A kind of Radar range profile's target identification method based on depth autoencoder network |
| CN107390193A (en) * | 2017-06-23 | 2017-11-24 | 西安电子科技大学 | Frequency modulated continuous wave radar Aircraft Targets sorting technique based on the fusion of more range cells |
| CN108509935A (en) * | 2018-04-12 | 2018-09-07 | 电子科技大学 | A kind of radar operation mode recognition methods based on random forests algorithm |
| US20200074230A1 (en) * | 2018-09-04 | 2020-03-05 | Luminar Technologies, Inc. | Automatically generating training data for a lidar using simulated vehicles in virtual space |
| CN110954872A (en) * | 2019-12-17 | 2020-04-03 | 西安电子科技大学 | Multi-layer perceptron MLP-based phased array radar working mode identification method |
-
2021
- 2021-06-23 CN CN202110700893.0A patent/CN113608172B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101894269A (en) * | 2010-07-16 | 2010-11-24 | 西安电子科技大学 | Multi-classifier system-based synthetic aperture radar automatic target recognition method |
| CN107194433A (en) * | 2017-06-14 | 2017-09-22 | 电子科技大学 | A kind of Radar range profile's target identification method based on depth autoencoder network |
| CN107390193A (en) * | 2017-06-23 | 2017-11-24 | 西安电子科技大学 | Frequency modulated continuous wave radar Aircraft Targets sorting technique based on the fusion of more range cells |
| CN108509935A (en) * | 2018-04-12 | 2018-09-07 | 电子科技大学 | A kind of radar operation mode recognition methods based on random forests algorithm |
| US20200074230A1 (en) * | 2018-09-04 | 2020-03-05 | Luminar Technologies, Inc. | Automatically generating training data for a lidar using simulated vehicles in virtual space |
| CN110954872A (en) * | 2019-12-17 | 2020-04-03 | 西安电子科技大学 | Multi-layer perceptron MLP-based phased array radar working mode identification method |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114463294A (en) * | 2022-01-24 | 2022-05-10 | 北京唯迈医疗设备有限公司 | Automatic metering control method and system for X-ray machine |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113608172B (en) | 2023-12-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Petrov et al. | Radar emitter signals recognition and classification with feedforward networks | |
| CN110954872B (en) | Multi-layer perceptron MLP-based phased array radar working mode identification method | |
| CN108051781A (en) | A kind of radar signal operating mode recognition methods based on DBN model | |
| CN111060878B (en) | LFM radar working mode real-time classification method and device suitable for single pulse | |
| Mottier et al. | Deinterleaving and clustering unknown radar pulses | |
| CN112213697B (en) | Feature fusion method for radar deception jamming recognition based on Bayesian decision theory | |
| CN113608172A (en) | Airborne multifunctional radar working mode identification method based on improved K nearest neighbor | |
| Matuszewski | Application of clustering methods for recognition of technical objects | |
| Nguyen et al. | A parallel neural network-based scheme for radar emitter recognition | |
| Ren et al. | Radar emitter signal classification based on mutual information and fuzzy support vector machines | |
| Jordanov et al. | Sets with incomplete and missing data—NN radar signal classification | |
| CN113625242B (en) | Radar signal sorting method based on potential distance graph combined PCA and improved cloud model | |
| CN112564834B (en) | Intelligent cognition and interference method and system for wireless communication system | |
| Zhu et al. | Radar signal pre-sorting based on mean-shift algorithm | |
| Wang et al. | Radar emitter classification based on a multiperspective collaborative clustering method and radar characteristic spectrum | |
| CN110880012B (en) | Inter-pulse agile radar radiation source frequency information correlation method for multi-reconnaissance platform | |
| Ruijia et al. | Radar emitter recognition in airborne RWR/ESM based on improved K nearest neighbor algorithm | |
| Wang et al. | A PRI estimation and signal deinterleaving method based on density-based clustering | |
| Liu et al. | Individual Intelligent Recognition Method Based on Fingerprint Features of Radar Emitter | |
| CN113381827B (en) | Electromagnetic signal spectrum sensing method based on deep clustering network | |
| CN114594427B (en) | Radar active interference online identification method based on posterior probability support vector machine | |
| KR102678770B1 (en) | Method and Apparatus for Features Extraction for Target Identification | |
| El Hasnaouy et al. | Comparison of Feature Extraction Methods for Automated Target Recognition by Reducing Speckle Noise in SAR Data | |
| Zhou et al. | Algorithm for Radar Signal Pre-sorting Based on Correlation Distance Differential | |
| Mei et al. | Radar high resolution range profile open set recognition of UAV target |
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 |