CN115061094A - Radar target identification method based on neural network and SVM - Google Patents
Radar target identification method based on neural network and SVM Download PDFInfo
- Publication number
- CN115061094A CN115061094A CN202210634773.XA CN202210634773A CN115061094A CN 115061094 A CN115061094 A CN 115061094A CN 202210634773 A CN202210634773 A CN 202210634773A CN 115061094 A CN115061094 A CN 115061094A
- Authority
- CN
- China
- Prior art keywords
- data
- energy density
- neural network
- training
- detected
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 23
- 230000001629 suppression Effects 0.000 claims abstract description 25
- 238000012545 processing Methods 0.000 claims abstract description 19
- 239000013598 vector Substances 0.000 claims abstract description 19
- 238000001228 spectrum Methods 0.000 claims abstract description 15
- 238000009825 accumulation Methods 0.000 claims abstract description 13
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 61
- 238000013145 classification model Methods 0.000 claims description 15
- 238000003062 neural network model Methods 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 14
- 230000001427 coherent effect Effects 0.000 claims description 12
- 210000002569 neuron Anatomy 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 2
- 238000012706 support-vector machine Methods 0.000 description 23
- 238000005516 engineering process Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 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
- 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
- G01S7/28—Details of pulse systems
- G01S7/285—Receivers
- G01S7/292—Extracting wanted echo-signals
-
- 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
- G01S7/28—Details of pulse systems
- G01S7/285—Receivers
- G01S7/292—Extracting wanted echo-signals
- G01S7/2923—Extracting wanted echo-signals based on data belonging to a number of consecutive radar periods
-
- 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
- G01S7/28—Details of pulse systems
- G01S7/285—Receivers
- G01S7/295—Means for transforming co-ordinates or for evaluating data, e.g. using computers
-
- 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
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/415—Identification of targets based on measurements of movement associated with the target
-
- 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
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
Abstract
The invention relates to a radar target recognition method based on a neural network and an SVM, and belongs to the technical field of radar target recognition. The method comprises the following steps: acquiring radar echo data; performing clutter suppression processing on radar echo data; performing wavelet packet decomposition on the data subjected to clutter suppression to obtain energy density characteristics; classifying the energy density characteristics according to the speed of the target and the signal-to-noise ratio interval of signal accumulation; inputting the classified energy density characteristics into a neural network of a corresponding interval, and converting an output result into label distance characteristics; extracting frequency domain entropy and relative RCS characteristics by using the frequency spectrum of the radar echo; extracting time domain waveform variance characteristics by using the radar echo time domain signal; combining the label distance, the frequency domain entropy, the relative RCS characteristics and the time domain waveform variance into a characteristic vector and inputting the characteristic vector into an SVM classifier; and judging a recognition result according to the output value of the SVM.
Description
Technical Field
The invention belongs to the technical field of radar target recognition, and relates to a radar target recognition method based on a neural network and an SVM.
Background
With the continuous development of information technology, information technologies such as artificial intelligence and the like are applied to automatic and behavior-intelligent weaponry, and radar target identification also becomes one of core technologies for equipment intelligence. The radar target identification can provide information such as the type and the attribute of the target, and provides an important basis for the threat degree evaluation of the battlefield target. The radar is divided into a broadband radar and a narrowband radar according to the bandwidth.
The identification of the narrow-band radar to the ground target is always a big difficulty of radar target identification, and the main reason is that the identification is influenced by a complex ground environment and low resolution, the extracted target characteristics are limited, and the identification accuracy is further influenced. The method and the device aim to solve the problem that the identification rate of the narrow-band radar to the ground target is low.
Disclosure of Invention
The invention aims to provide a radar target recognition method based on a neural network and an SVM (support vector machine), aiming at the problem of low ground target recognition rate of a narrow-band signal radar.
In order to achieve the above purpose, the present invention adopts the following technical scheme.
The radar target identification method comprises the following steps:
s1, radar echo data are obtained;
s2, performing clutter suppression processing on the radar echo data to obtain echo data subjected to clutter suppression processing, and constructing training data and data to be detected;
the training data and test data will be in accordance with q 1 :1-q 1 Dividing the ratio; q is a number of 1 The value range of (a) is 0.5 or more and 0.9 or less.
S3, extracting the energy density characteristics of echo data of the training data and the data to be detected after the clutter suppression processing S2;
s31, carrying out wavelet packet decomposition on the training data and the data to be detected after the clutter suppression processing of S2 to obtain a frequency sub-band coefficient;
the wavelet packet is decomposed into U layers, and the value range of U is 2-10; the frequency subband coefficients are K, and K is the power of U of 2; u is more than or equal to 2;
s32, calculating the energy of each frequency sub-band coefficient;
s33, acquiring energy density characteristics;
s33, the energy density features are K and are the same as the number of the frequency subband coefficients;
s4, carrying out interval classification on the energy density characteristics of the training data in the S3 according to the target speed and the signal coherent accumulation signal-to-noise ratio to obtain a plurality of classification intervals, and constructing a network training set and a network test set for each interval in each interval;
the number of the classification intervals is P, and P is more than or equal to 5; the network training set and the network testing set enable all energy density characteristics to be according to q 2 :1-q 2 Dividing the ratio; q is a number of 2 The value range of (a) is 0.5 or more and 0.9 or less.
S5, training a neural network according to the energy density characteristics of the classification intervals obtained in the S4, and obtaining a trained neural network model;
p trained neural network models are provided, and each trained neural network model comprises an input layer, a hidden layer and an output layer; the number of the input layers is K, and the number of the input layers is the same as the number of the energy density characteristics and the number of the frequency sub-band coefficients and is K; the number of the neurons of the hidden layer is more than or equal to 20; the hidden layer and the output layer both use a Tanh function as an activation function; the loss function of the neural network is a mean square error function;
s6, inputting the energy density characteristics of each classification interval obtained in the S4 training data into a neural network model corresponding to a target speed and signal coherent accumulation signal-to-noise ratio interval to obtain an output vector, converting the output vector into coordinates, calculating the distance between the converted coordinates and target label coordinates, and obtaining label distance characteristics;
the tag distance characteristic is marked as d ═ d i }; the value range of i is 1 to the number of target categories, and the number of the target categories is recorded as m; the value range of m is more than or equal to 2 and less than or equal to 5;
s7, calculating the frequency spectrums of the echo data of the training data and the data to be detected after clutter suppression processing in S2, and acquiring frequency domain entropy and relative RCS characteristics, specifically:
s71, calculating the frequency spectrums of the echo data of the training data and the data to be detected after clutter suppression processing in S2 to obtain a frequency spectrum module value sequence;
the frequency spectrum module value sequence is marked as X ═ X 1 ,x 2 ,…x N }; n is the length of the frequency spectrum module value sequence and the length is the power of 2;
s72, calculating frequency domain entropy and relative RCS characteristics;
the frequency domain entropy is marked as H; the relative RCS characteristic is marked as sigma;
s8, using the time domain echo data of the training data and the data to be detected after clutter suppression processing in S2 to obtain time domain waveform variance characteristics, specifically:
s81, calculating the module values and time domain module value sequences of the echo data of the training data and the data to be detected after clutter suppression in S2;
the time domain module value sequence is marked as Y ═ Y 1 ,y2,…y N }; n is the length of the time domain modulus value sequence, the length is the power of 2 and is the same as the length of the frequency spectrum modulus value sequence;
s82, calculating time domain waveform variance characteristics;
the time domain waveform variance characteristic is recorded as S;
s9, combining the distance label features obtained by the S6 training data, the frequency domain entropy and the relative RCS features obtained by the S7 training data and the waveform variance features obtained by the S8 training data into feature vectors, and constructing a training set and a test set;
the feature vector is denoted as XX, and XX ═ d i H, σ, S }; the training set and the test set use all the feature vectors according to q 3 :1-q 3 Dividing the ratio of (A) to (B); q is a number of 3 The value range of (a) is not less than 0.5 but not more than 0.9.
S10, training a Gaussian kernel SVM classification model by using the feature vector combined in the S9 to obtain the Gaussian kernel SVM classification model;
s10, the scale range of the Gaussian kernel in the Gaussian kernel SVM classification model is more than or equal to 0.5 and less than or equal to 1; the output values of the Gaussian kernel SVM classification model are m, wherein m is the number of target categories; the method comprises the steps that an M-fold cross verification method is adopted to verify a Gaussian kernel SVM classification model by training the Gaussian kernel SVM classification model, wherein M is greater than or equal to 5 and less than or equal to 20;
s11, sequentially inputting the K energy density characteristics of the data to be detected extracted by S3, the frequency domain entropy and the relative RCS characteristics of the data to be detected extracted by S7 and the time domain waveform variance characteristics of the data to be detected extracted by S8 into an S5 trained neural network model and an S10 trained Gaussian kernel SVM classification model to obtain a final recognition result, wherein the final recognition result is specifically as follows:
s111, inputting K energy density characteristics of echo data to be detected into a trained neural network model of S5 according to a target speed and a signal coherent accumulation signal-to-noise ratio classification interval to obtain an output result;
s112, converting the output result of the S111 into a label distance characteristic;
and S113, inputting the tag distance characteristic of S112, the frequency domain entropy and the relative RCS characteristic of the data to be detected extracted by S7, and the time domain waveform variance characteristic of the data to be detected extracted by S8 into a Gaussian kernel SVM model of S10 to obtain a final recognition result.
Advantageous effects
Compared with the conventional single-classifier target identification method, the radar target identification method based on the neural network and the SVM has the following beneficial effects:
the conventional single classifier identification method is easy to identify the ground target by mistake; the method adopts a serial structure of two classifiers, so that misrecognized data can be corrected, and the accuracy of target recognition is improved.
Drawings
FIG. 1 is a flow chart of an implementation of a radar target identification method based on a neural network and an SVM of the present invention;
FIG. 2 is a classification interval of energy density features according to target speed and signal coherent accumulation SNR in an embodiment;
FIG. 3 is a comparison graph of the recognition results of the present invention method with a conventional single neural network recognition method and a single SVM classifier recognition method.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the embodiments are a part of the embodiments of the present invention, but not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, belong to the scope of the present invention.
Example 1
The method and the device aim at identifying the data of the moving target of the single coherent accumulation frame acquired by the ground radar, and the moving target can comprise a vehicle, a person, an unmanned aerial vehicle and the like. The data of the embodiment is obtained by a radar on a certain ground in an outfield field experiment. The radar equipment acquires data of two types of vehicles, which are respectively represented by the vehicle 1 and the vehicle 2, and the two types of data are identified by adopting the target identification method.
This example illustrates a specific implementation of the method of the present invention, as shown in FIG. 1.
The radar target identification method based on the neural network and the SVM comprises the following steps:
and S1, receiving data of the X-band radar in a coherent accumulation frame, wherein one coherent processing time comprises 128 repetition periods.
And S2, performing clutter suppression processing on the data of the S1. And obtaining echo data after clutter suppression processing, and constructing training data and data to be detected according to the ratio of 0.7: 0.3.
And S3, extracting energy density characteristics of the training data and the data to be detected after the S2 clutter suppression. Wherein, the extraction of the energy density characteristic comprises the following substeps:
s31, carrying out 4-layer wavelet packet decomposition on the training data and the data to be detected after S2 clutter suppression to obtain 16 frequency sub-bands, wherein the coefficient in each frequency sub-band is a i,k Where i is the number of the frequency sub-band and k is the number of the coefficient in the frequency sub-band.
S32, calculating energy E of each frequency sub-band i ：
S3.3, obtaining 16 energy density characteristics e i ：
S4, classifying the 16 energy density characteristics of the training data in the S3 according to a target speed and a signal coherent accumulation signal-to-noise ratio interval, wherein the specific classification method is as shown in figure 2, firstly dividing the training data into an interval of 0-10 km/h (including 10km/h) and an interval above 10km/h according to the speed, and then dividing the training data into 5 intervals of 13 dB-16 dB (including 16dB), 16 dB-20 dB (including 20dB), 20 dB-25 dB (including 25dB), 25 dB-30 dB (including 30dB) and 30dB according to the signal coherent accumulation signal-to-noise ratio in the two speed intervals. The energy density characteristic is divided into 10 intervals, and a network training set and a network test set are constructed by training data of each interval according to the proportion of 0.7: 0.3.
And S5, training the neural network by using the energy density characteristics of the training data of each interval in the S4 to obtain 10 neural network models.
Specifically, a neural network with two hidden layers and one output layer is built, wherein the number of neurons in the hidden layers is 35 and 20 in sequence, Tanh functions are used as activation functions in the hidden layers and the output layers, and loss functions of the neural network are mean square error functions. The output tag settings are [1,0] and [0,1] for car 1 and car 2, respectively. Finally, a 16 × 35 × 20 × 2 neural network structure can be obtained, and the neural network structure of each interval is the same.
Specifically, the expression of the Tanh function is as follows:
wherein alpha is the linear operation result of the neuron.
Specifically, the mean square error loss function expression is:
wherein, B is the label vector, and f (x) is the output vector of the output layer.
S6, inputting the energy density characteristics of the training data of each interval in the S4 into the neural network of the corresponding interval to obtain an output vector [ x, y]And is given in the form of coordinates (x, y). Using the coordinates, the distance d to the coordinates (1,0) and (0,1) is calculated 1 And d 2 Where (1,0) represents the tag coordinates of car 1 and (0,1) represents the tag coordinates of car 2.
And S7, extracting frequency domain entropy and relative RCS characteristics of the training data and the data to be detected after clutter suppression in the S2. Wherein, the extraction of the frequency domain entropy and the relative RCS characteristics comprises the following substeps:
s71, calculating the frequency spectrums of the echo data of the training data and the data to be detected after the clutter suppression processing in the S2 to obtain a frequency spectrum module value sequence X ═ X 1 ,x 2 ,…x 128 }。
S72, calculating frequency domain entropy H:
wherein
S73, calculating relative RCS characteristics:
σ＝P r R 4
wherein, P r R is the target distance, which is the radar echo power. P r The calculation formula of (2) is as follows:
and S8, extracting time domain waveform variance characteristics of the training data and the data to be detected after clutter suppression in S2. Wherein the extraction of the time domain waveform variance features comprises the following sub-steps:
s81, obtaining a time domain module value sequence Y ═ Y 1 ,y 2 ,…,y 128 }。
S82, calculating time domain waveform variance S:
wherein, the first and the second end of the pipe are connected with each other,
for the average modulus, the calculation formula is:
s9, combining the distance label feature obtained by the S6 training data, the frequency domain entropy and relative RCS feature obtained by the S7 training data and the waveform variance feature obtained by the S8 training data into a feature vector XX, and constructing a training set and a testing set in a ratio of 0.7: 0.3.
The feature vector XX is specifically:
XX＝{d 1 ,d 2 ,H,σ,S}
s10, training the Gaussian kernel SVM model by using the feature vector XX combined by the S9.
Specifically, a gaussian kernel function SVM is constructed, a kernel scale is set to be 0.61, and two output labels 1 and 2 are set, wherein 1 represents vehicle 1, and 2 represents vehicle 2. The SVM model is trained according to the proportion of the training set to the test set in a 7:3 mode by the data set, and 10-fold cross validation is adopted in training.
S11, sequentially inputting the 16 energy density features of the data to be detected extracted by the S3, the frequency domain entropy and the relative RCS features of the data to be detected extracted by the S7 and the time domain waveform variance features of the data to be detected extracted by the S8 into the S5 trained neural network model and the S10 trained Gaussian kernel SVM classification model to obtain a final recognition result.
Comprising the following substeps:
and S111, inputting the 16 energy density characteristics of the echo data to be detected extracted in the step S3 into the neural network model trained in the step S6 according to the target speed and the signal-coherent accumulation signal-to-noise ratio classification interval to obtain an output result (x, y).
S112, converting the output result (x, y) of the S111 into a label distance characteristic d 1 And d 2 。
S113, the label distance characteristic d obtained in the S112 1 And d 2 And S7, extracting the frequency domain entropy characteristic H and the relative RCS characteristic sigma of the data to be detected, and S8, extracting the time domain waveform variance characteristic S of the data to be detected, and inputting the time domain waveform variance characteristic S into the Gaussian kernel SVM model trained in S10. If 1 is output, the vehicle is judged as 1; if 2 is outputted, it is determined as vehicle 2.
The identification results of the data to be measured of the vehicles 1 and 2 obtained according to the above method are shown in table 1:
TABLE 1 identification results of data to be tested
The recognition rate of the vehicle 1 reaches 93.39%, and the recognition rate of the vehicle 2 reaches 92.01%. The average recognition rate of the test set reaches 92.71%. Compared with the recognition results of the conventional single neural network recognition method and the single SVM classifier recognition method, as shown in FIG. 3, the recognition rate of the vehicle 1 is respectively improved by 9.04% and 0.89% compared with the recognition method of the single neural network and the recognition method of the single SVM; the recognition rate of the vehicle 2 is respectively improved by 9.66% and 6.94%. The recognition accuracy of the two targets can be effectively improved.
The above description is an embodiment of the present invention, and the present invention should not be limited to the disclosure of the embodiment and the drawings. Equivalents and modifications may be made without departing from the spirit of the disclosure, which is to be considered as within the scope of the invention.
Claims (10)
1. A radar target identification method based on a neural network and an SVM is characterized in that: the method comprises the following steps:
s1, radar echo data are obtained;
s2, performing clutter suppression processing on the radar echo data to obtain echo data subjected to clutter suppression processing, and constructing training data and data to be detected;
s3, extracting the energy density characteristics of echo data of training data and data to be detected after the clutter suppression processing of S2, specifically:
s31, carrying out wavelet packet decomposition on the training data and the data to be detected after the clutter suppression processing of S2 to obtain a frequency sub-band coefficient;
s32, calculating the energy of each frequency sub-band coefficient;
s33, acquiring energy density characteristics;
s4, carrying out interval classification on the energy density characteristics of the training data in the S3 according to the target speed and the signal coherent accumulation signal-to-noise ratio to obtain a plurality of classification intervals, and constructing a network training set and a network test set for each interval in each interval;
s5, training a neural network according to the energy density characteristics of the classification intervals obtained in the S4, and obtaining a trained neural network model;
s6, inputting the energy density characteristics of each classification interval obtained in the S4 training data into a neural network model corresponding to a target speed and signal coherent accumulation signal-to-noise ratio interval to obtain an output vector, converting the output vector into coordinates, calculating the distance between the converted coordinates and target label coordinates, and obtaining label distance characteristics;
s7, calculating the frequency spectrums of the echo data of the training data and the data to be detected after clutter suppression processing in S2, and acquiring frequency domain entropy and relative RCS characteristics;
s8, using the time domain echo data of the training data and the data to be detected after clutter suppression processing in S2 to obtain time domain waveform variance characteristics, specifically:
s81, calculating the module values and time domain module value sequences of the echo data of the training data and the data to be detected after clutter suppression in S2;
s82, calculating time domain waveform variance characteristics;
s9, combining the distance label features obtained from the S6 training data, the frequency domain entropy and relative RCS features obtained from the S7 training data and the waveform variance features obtained from the S8 training data into feature vectors, and constructing a training set and a test set;
s10, training a Gaussian kernel SVM classification model by using the feature vector combined in the S9 to obtain the Gaussian kernel SVM classification model;
s11, sequentially inputting the K energy density features of the data to be detected extracted from S3, the frequency domain entropy and the relative RCS features of the data to be detected extracted from S7 and the time domain waveform variance features of the data to be detected extracted from S8 into an S5-trained neural network model and an S10-trained Gaussian kernel SVM classification model to obtain a final recognition result.
2. The radar target recognition method of claim 1, wherein: the training data and the test data of S2 are according to q 1 :1-q 1 Dividing the ratio; q is a number of 1 The value range of (a) is 0.5 or more and 0.9 or less.
3. The radar target recognition method of claim 1, wherein: s31, decomposing the wavelet packet into U layers, wherein the value range of U is 2-10; the frequency subband coefficients are K, and K is the power of U of 2; u is more than or equal to 2; and S33, the energy density is characterized by K, and the number of the energy density characteristics is the same as that of the frequency subband coefficients.
4. The radar target recognition method of claim 1, wherein: s4, the number of the classification intervals is P, and P is more than or equal to 5; the network training set and the network testing set enable all energy density characteristics to be according to q 2 :1-q 2 Dividing the ratio; q is a number of 2 The value range of (a) is 0.5 or more and 0.9 or less.
5. The radar target recognition method according to claim 1 or 3, characterized in that: s5, the number of the trained neural network models is P, and each trained neural network model comprises an input layer, a hidden layer and an output layer; the number of the input layers is K, and the number of the input layers is the same as the number of the energy density characteristics and the number of the frequency sub-band coefficients and is K; the number of the neurons of the hidden layer is more than or equal to 20; the hidden layer and the output layer both use a Tanh function as an activation function; the loss function of the neural network is a mean square error function.
6. The radar target recognition method of claim 1, wherein: the label distance characteristic of S6 is marked as d ═ d i }; the value range of i is 1 to the number of target categories, and the number of the target categories is recorded as m; the value range of m is more than or equal to 2 and less than or equal to 5.
7. The radar target recognition method of claim 6, wherein: s7, specifically, the method comprises the following steps:
s71, calculating the frequency spectrums of the echo data of the training data and the data to be detected after clutter suppression processing in S2 to obtain a frequency spectrum module value sequence;
the frequency spectrum module value sequence is marked as X ═ X 1 ，x 2 ，...x N }; n is the length of the frequency spectrum module value sequence and the length is the power of 2;
s72, calculating frequency domain entropy and relative RCS characteristics;
the frequency domain entropy is marked as H; the relative RCS characteristic is denoted as sigma.
8. The radar target recognition method of claim 7, wherein: s81, the time-domain modulus sequence is denoted as Y ═ Y 1 ，y 2 ，...y N }; n is the length of the time domain module value sequence, the length is the power of 2 and is the same as the length of the frequency spectrum module value sequence; and S82, the time domain waveform variance is characterized by being marked as S.
S9, denoted as XX, and XX ═ d i H, σ, S }; the training set and the test set use all the feature vectors according to q 3 ∶1-q 3 Dividing the ratio of (A) to (B); q is a number of 3 The value range of (a) is 0.5 or more and 0.9 or less.
9. The radar target recognition method according to claim 1, characterized in that: s10, the scale range of the Gaussian kernel in the Gaussian kernel SVM classification model is more than or equal to 0.5 and less than or equal to 1; the output values of the Gaussian kernel SVM classification model are m, wherein m is the number of target categories; and verifying the Gaussian kernel SVM classification model by adopting an M-fold cross verification method for the training Gaussian kernel SVM classification model, wherein M is greater than or equal to 5 and less than or equal to 20.
10. The radar target recognition method according to claim 1 or 3, characterized in that: s11, specifically, the method comprises the following steps:
s111, inputting K energy density characteristics of echo data to be detected into a neural network model trained in S5 according to a target speed and a signal coherent accumulation signal-to-noise ratio classification interval to obtain an output result;
s112, converting the output result of the S111 into a label distance characteristic;
and S113, inputting the tag distance characteristic of S112, the frequency domain entropy and the relative RCS characteristic of the data to be detected extracted by S7, and the time domain waveform variance characteristic of the data to be detected extracted by S8 into a Gaussian kernel SVM model of S10 to obtain a final recognition result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210634773.XA CN115061094A (en) | 2022-06-06 | 2022-06-06 | Radar target identification method based on neural network and SVM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210634773.XA CN115061094A (en) | 2022-06-06 | 2022-06-06 | Radar target identification method based on neural network and SVM |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115061094A true CN115061094A (en) | 2022-09-16 |
Family
ID=83200982
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210634773.XA Pending CN115061094A (en) | 2022-06-06 | 2022-06-06 | Radar target identification method based on neural network and SVM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115061094A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117331047A (en) * | 2023-12-01 | 2024-01-02 | 德心智能科技(常州)有限公司 | Human behavior data analysis method and system based on millimeter wave radar |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108256488A (en) * | 2018-01-19 | 2018-07-06 | 中国人民解放军陆军装甲兵学院 | A kind of radar target identification method based on micro-Doppler feature extraction and deep learning |
| DE102018122560A1 (en) * | 2018-09-14 | 2020-03-19 | Autel Robotics Co., Ltd. | TARGET CLASSIFICATION METHOD AND RELATED DEVICE |
-
2022
- 2022-06-06 CN CN202210634773.XA patent/CN115061094A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108256488A (en) * | 2018-01-19 | 2018-07-06 | 中国人民解放军陆军装甲兵学院 | A kind of radar target identification method based on micro-Doppler feature extraction and deep learning |
| DE102018122560A1 (en) * | 2018-09-14 | 2020-03-19 | Autel Robotics Co., Ltd. | TARGET CLASSIFICATION METHOD AND RELATED DEVICE |
Non-Patent Citations (2)
| Title |
|---|
| HE MIAO等: "Performance Comparison of Wideband Radar Target Detection Methods", ADVANCED MATERIALS RESEARCH, 10 September 2015 (2015-09-10) * |
| 曾卫民;李桂祥;赵伏贤;: "一种基于支持向量机的无源雷达目标识别方法", 现代防御技术, no. 01, 15 February 2010 (2010-02-15) * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117331047A (en) * | 2023-12-01 | 2024-01-02 | 德心智能科技(常州)有限公司 | Human behavior data analysis method and system based on millimeter wave radar |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107728142B (en) | Radar high-resolution range profile target identification method based on two-dimensional convolutional network | |
| CN109597043B (en) | Radar signal identification method based on quantum particle swarm convolutional neural network | |
| CN107728143B (en) | Radar high-resolution range profile target identification method based on one-dimensional convolutional neural network | |
| CN107830996B (en) | Fault diagnosis method for aircraft control surface system | |
| CN112882009B (en) | Radar micro Doppler target identification method based on amplitude and phase dual-channel network | |
| CN108051781A (en) | A kind of radar signal operating mode recognition methods based on DBN model | |
| CN111722199A (en) | Radar signal detection method based on convolutional neural network | |
| CN112560803A (en) | Radar signal modulation identification method based on time-frequency analysis and machine learning | |
| CN110109110B (en) | HRRP target identification method based on priori optimal variation self-encoder | |
| CN106950475A (en) | A kind of local discharge signal extracting method and device based on wavelet transformation | |
| CN109711314A (en) | Radar emitter signal classification method based on Fusion Features and SAE | |
| CN111368930B (en) | Radar human body posture identification method and system based on multi-class spectrogram fusion and hierarchical learning | |
| CN115061094A (en) | Radar target identification method based on neural network and SVM | |
| CN113095354B (en) | Unknown radar target identification method based on radiation source characteristic subspace knowledge | |
| CN111983569B (en) | Radar interference suppression method based on neural network | |
| CN114487129A (en) | Flexible material damage identification method based on acoustic emission technology | |
| CN113640768A (en) | Low-resolution radar target identification method based on wavelet transformation | |
| CN111458688B (en) | Three-dimensional convolution network-based radar high-resolution range profile target recognition method | |
| CN112904299A (en) | Radar high-resolution range profile open set target identification method based on deep intra-class division | |
| CN113486917A (en) | Radar HRRP small sample target identification method based on metric learning | |
| CN116804735A (en) | Radar target recognition method based on wavelet transformation and neural network | |
| CN110555457A (en) | Engine lubricating oil wear signal characteristic processing system and method | |
| CN103439696B (en) | Method for steadily recognizing ground moving objects under low signal to noise ratio condition | |
| CN115616503A (en) | Radar interference signal type identification method based on convolutional neural network model | |
| CN114970638A (en) | Radar radiation source individual open set identification method and system |
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 |