CN114595732A - Radar radiation source sorting method based on depth clustering - Google Patents

Radar radiation source sorting method based on depth clustering Download PDF

Info

Publication number
CN114595732A
CN114595732A CN202210500222.4A CN202210500222A CN114595732A CN 114595732 A CN114595732 A CN 114595732A CN 202210500222 A CN202210500222 A CN 202210500222A CN 114595732 A CN114595732 A CN 114595732A
Authority
CN
China
Prior art keywords
radiation source
layer
radar radiation
clustering
network
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
Application number
CN202210500222.4A
Other languages
Chinese (zh)
Other versions
CN114595732B (en
Inventor
顾颖
刘江峰
王伯祥
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xi'an Shengxin Technology Co.,Ltd.
Original Assignee
Xi'an Sensing Technology Development Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xi'an Sensing Technology Development Co ltd filed Critical Xi'an Sensing Technology Development Co ltd
Priority to CN202210500222.4A priority Critical patent/CN114595732B/en
Publication of CN114595732A publication Critical patent/CN114595732A/en
Application granted granted Critical
Publication of CN114595732B publication Critical patent/CN114595732B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • 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
    • 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
    • G06F18/2134—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on separation criteria, e.g. independent component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • 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
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • 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)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention relates to a radar radiation source sorting method based on depth clustering, which comprises the following steps: preprocessing data; constructing a convolutional self-encoder network; constructing a fully connected self-encoder network; pre-training a network; fusing the characteristics; estimating the number of clusters; clustering data; network fine adjustment; and outputting a sorting result. The neural network is used for automatically extracting the intra-pulse characteristics of the radar radiation source signals, so that the artificial design and calculation of the intra-pulse characteristics are avoided, and the intelligence and the self-adaptive capacity are improved. Meanwhile, the clustering number is estimated by using a density peak clustering algorithm, so that the problem that the k-means clustering needs to preset the clustering number is solved, the prior knowledge required by the method is reduced, and the sorting of unknown radar radiation sources can be realized. In addition, the intra-pulse and inter-pulse characteristics of the radar radiation source are fused to form a combined characteristic, so that the combined characteristic for sorting simultaneously contains the intra-pulse and inter-pulse information of the radar radiation source, and the sorting accuracy of the radar radiation source is improved.

Description

Radar radiation source sorting method based on depth clustering
Technical Field
The invention belongs to the field of signal processing, and further relates to a radar radiation source sorting method based on depth clustering in the technical field of radar radiation source sorting. The invention can be used for sorting the received radar radiation source signals in electronic information reconnaissance, electronic support and threat warning systems.
Background
Radar source sorting refers to separating signals of each radar from intercepted densely overlapped pulse streams, and is an important component of radar reconnaissance. The radar reconnaissance system has the primary tasks of sorting and identifying each radar signal from staggered pulse streams to obtain model information of different radiation sources, and then analyzing information such as working modes, threat levels and the like of each radar radiation source according to identification results. The accuracy of radar radiation source sorting directly influences the accuracy of radar radiation source identification, and further influences the accuracy of information acquired by an upper-layer system, so that the radar radiation source sorting has important significance for radar reconnaissance and even the whole electronic countermeasure system.
In the traditional radar radiation source sorting method, a pulse description word consisting of carrier frequency, pulse width, arrival time, arrival angle and amplitude of a radar radiation source is used as a characteristic parameter to analyze the radar radiation source. However, with the development of electronic technology, the number of radar radiation sources is increased due to the wide application of electromagnetic devices such as radars, the electromagnetic environment is increasingly complex, and meanwhile, due to the application of the complex system radars, the modulation patterns of radar transmission signals are diversified, and the anti-interference and anti-reconnaissance capabilities are improved. Therefore, the traditional method for sorting by using single and stable characteristic parameters is difficult to process massive and overlapped signals, the research can process massive complex data, and the radar signal sorting method suitable for the existing electromagnetic environment becomes a problem to be solved urgently in the field of radar reconnaissance.
The university of major graduates has proposed a radar signal sorting method based on deep migration learning in the patent document "a radar signal sorting method based on deep migration learning" (patent application No. 202110359057.0, application publication No. CN 113030958A). The method is based on carrier frequency and pulse width two-dimensional characteristic parameters, radar pulse sequences received from different areas are converted into pictures to carry out radar radiation source sorting, images of other areas are trained firstly by adopting a target detection method based on a fast RCNN network to obtain the weight of a model, and then the model is transferred to the current area to realize the sorting of the radar radiation sources. The method can accurately detect the overlapped radar radiation sources, improves the sorting accuracy, but only utilizes the two-dimensional characteristics of the carrier frequency and the pulse width of the radiation sources to sort, is not sufficient for information mining of signals, and is not applicable any more if the carrier frequencies and the pulse widths of different radiation sources are highly similar.
The patent document applied by the university of people liberation military air force engineering of China "a radar signal sorting method based on nuclear cluster support vector clustering" (patent application No. 201811464561.1, application publication No. CN 109613486A) provides a radar signal sorting method based on nuclear cluster support vector clustering. The method comprises the steps of firstly utilizing conventional parameters to perform clustering pre-sorting on interleaved signals, then selecting intra-pulse data corresponding to missed selection pulses, extracting intra-pulse modulation characteristics beneficial to signal sorting, then utilizing a characteristic selection algorithm to select key characteristics of radar radiation source signals, adopting a proposed kernel cluster-based support vector clustering method to perform clustering sorting on the selected characteristics, and finally combining two sorting results to complete final sorting. The method overcomes the problems that the traditional method is limited in application range, complex in calculation, not beneficial to engineering realization and the like, but the method needs manual extraction of intra-pulse features, the sorting result is greatly influenced by manually designed features, the manual extraction of the features is time-consuming and labor-consuming, and the intelligence degree of the method is low.
Disclosure of Invention
Technical problem to be solved
In order to avoid the defects of the prior art, the invention provides a radar radiation source sorting method based on deep clustering, and aims to solve the problems that the existing radar radiation source sorting method needs manual feature extraction, the sorting accuracy is low, and the like.
Technical scheme
A radar radiation source sorting method based on depth clustering is characterized by comprising the following steps:
step 1: preprocessing a radar radiation source data set to obtain a carrier frequency, a pulse width and a signal time-frequency diagram;
step 2: constructing a convolutional self-encoder network;
the convolutional self-encoder network comprises 16 layers, and the structure of the convolutional self-encoder network sequentially comprises the following layers: first convolution layer → first pooling layer → second convolution layer → second pooling layer → third convolution layer → third pooling layer → Flatten layer → first fully-connected layer → second fully-connected layer → Reshape layer → first upsampling layer → fourth convolution layer → second upsampling layer → fifth convolution layer → third upsampling layer → sixth convolution layer, wherein the first 8 layers constitute the encoder and the second 8 layers constitute the decoder;
the number of convolution kernels in the first convolution layer to the sixth convolution layer is set to be 16, 8, 8, 8, 16 and 1 respectively, and the sizes of the convolution kernels are all set to be 3
Figure 39822DEST_PATH_IMAGE001
Step length is set to 1, the activation function is a ReLU function, the filling mode selects and uses 'same' filling, the first to third pooling layers all adopt a maximum pooling mode, and the size of the pooling area core is set to 2
Figure 160225DEST_PATH_IMAGE002
Step length is set to be 1, the number of neurons of the first full connection layer and the second full connection layer is 32 and 512 respectively, the activation function is a ReLU function, and the size of an up-sampling window of the first up-sampling layer to the third up-sampling layer is 2
Figure 533437DEST_PATH_IMAGE003
2；
And step 3: constructing a fully-connected self-encoder network:
the fully-connected self-encoder network comprises 6 layers, and the structure of the fully-connected self-encoder network is as follows in sequence: the first hidden layer → the second hidden layer → the third hidden layer → the fourth hidden layer → the fifth hidden layer → the output layer, wherein, the first 3 layers compose the encoder, the last three layers compose the decoder;
setting the number of the neurons of the first hidden layer to 8, 8, 16, 8 and 8 respectively, wherein the activation function is a ReLU function, the number of the neurons of the output layer is 2, and the activation function is a sigmoid function;
and 4, step 4: network pre-training:
inputting the signal time-frequency diagram into a convolutional self-encoder network for pre-training to obtain a feature extraction radar radiation source intra-pulse feature extraction encoder network, and inputting the carrier frequency and the pulse width into a fully-connected self-encoder network for pre-training to obtain a radar radiation source inter-pulse feature extraction encoder network;
and 5: feature fusion:
splicing the outputs of the radar radiation source intra-pulse feature extraction encoder network and the inter-pulse feature extraction encoder network corresponding to each sample in the radar radiation source data set to obtain the joint features corresponding to the samples;
step 6: and (3) estimating the number of clusters:
estimating the clustering number of the radar radiation source data set by using a density peak clustering algorithm;
and 7: data clustering:
clustering samples in a radar radiation source data set to obtain a clustering label corresponding to each sample, and defining the clustering label as a label of the sample;
and 8: network fine adjustment:
splicing the outputs of the two encoder networks, inputting the spliced outputs into a full connection layer and an output layer, transforming the networks into a classification network, training the classification network by using a pseudo label supervision network, and realizing fine adjustment of network parameters, wherein the number of neurons of the full connection layer is set to be 32, an activation function is a ReLU function, the number of nodes of the output layer is equal to the number of clusters obtained by estimation in the step 6, and the activation function is a softmax function;
and step 9: and (3) outputting a sorting result:
and (5) repeatedly executing the step (5) to the step (8) for multiple times, and taking the last data clustering result as a final radar radiation source sorting result.
The further technical scheme of the invention is as follows: the data preprocessing in step 1 is specifically as follows:
(1a) intercepting radar radiation source signals within a period of time by using a reconnaissance receiver to obtain a radar radiation source data set;
(1b) measuring the carrier frequency and the pulse width of each radar radiation source signal sample in the radar radiation source data set, and normalizing the carrier frequency and the pulse width of each sample in the radar radiation source data set by using a min-max normalization method;
(1c) and calculating by using short-time Fourier transform to obtain a time-frequency graph of each sample in the radar radiation source data set, and performing min-max normalization and downsampling on the time-frequency graph to obtain a signal time-frequency graph.
The further technical scheme of the invention is as follows: the step 4 is as follows:
inputting the signal time-frequency diagram into a convolutional self-encoder network, iteratively updating the weight of the convolutional self-encoder network for N times by utilizing an Adam algorithm, discarding a decoder part of a trained convolutional self-encoder, and obtaining a pre-trained radar radiation source intra-pulse feature extraction encoder network;
inputting the carrier frequency and the pulse width into a fully-connected self-encoder network, iteratively updating the weight of the network for N times by using an Adam algorithm, discarding a decoder part of the trained fully-connected self-encoder, and obtaining a pre-trained radar radiation source inter-pulse characteristic extraction encoder network.
The further technical scheme of the invention is as follows: and 7, clustering the samples in the radar radiation source data set by adopting a k-means clustering algorithm.
Advantageous effects
In the prior art, the intra-pulse characteristics of radar radiation source signals need to be manually extracted, the prior knowledge such as the known clustering number is needed, and the characteristic mining of the signals is not sufficient. The method uses a self-encoder to extract the intra-pulse and inter-pulse characteristics of radar radiation source signals, fuses the inter-pulse and intra-pulse characteristics to form combined characteristics, and then adopts a strategy of clustering and network fine adjustment to perform network fine adjustment alternately, thereby improving the sorting accuracy; on the other hand, in order to reduce the dependence on the prior knowledge, the invention estimates the clustering number by using a density peak value clustering algorithm, provides input parameter information for a depth clustering algorithm and improves the intelligence of the method. Compared with the prior art, the invention has the following advantages:
firstly, because the neural network is utilized to automatically extract the intra-pulse characteristics of the radar radiation source signals, the artificial design and calculation of the intra-pulse characteristics are avoided, the manual intervention is reduced, and the intelligence and the self-adaptive capability of the invention are improved.
Secondly, the invention utilizes the density peak value clustering algorithm to estimate the clustering number, solves the problem that the k-means clustering needs to preset the clustering number, reduces the prior knowledge required by the invention, and can realize the sorting of unknown radar radiation sources.
Thirdly, because the intra-pulse and inter-pulse characteristics of the radar radiation source are fused to form a combined characteristic, the combined characteristic for sorting simultaneously contains the intra-pulse and inter-pulse information of the radar radiation source, and the sorting accuracy of the radar radiation source can be improved; on the other hand, the invention combines clustering and feature extraction, alternately carries out network fine adjustment and sample clustering, enables the network to learn the features which are easy to distinguish by a clustering algorithm, and improves the sorting accuracy.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, wherein like reference numerals are used to designate like parts throughout.
FIG. 1 is a flow chart of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The following describes the implementation steps of the present invention with reference to fig. 1.
Step 1, data preprocessing.
And intercepting radar radiation source signals within a period of time by using a reconnaissance receiver to obtain a radar radiation source data set.
And measuring the carrier frequency and the pulse width of each radar radiation source signal sample in the radar radiation source data set, and normalizing the carrier frequency and the pulse width of each sample in the radar radiation source data set by using a min-max normalization method.
Calculating by using short-time Fourier transform to obtain a time-frequency diagram of each sample in the radar radiation source data set, and performing min-max normalization and downsampling on the time-frequency diagram to obtain a time-frequency diagram with the size of 64
Figure 256542DEST_PATH_IMAGE004
64 signal time-frequency diagram.
The min-max is normalized as follows:
Figure 133232DEST_PATH_IMAGE005
wherein, the data of the radar radiation source after normalization processing is represented, the data of the radar radiation source before normalization processing is represented,
Figure 158956DEST_PATH_IMAGE006
and
Figure 488307DEST_PATH_IMAGE007
respectively representing maximum and minimum operations.
And 2, constructing a convolutional self-encoder network.
A16-layer convolutional self-encoder network is built, and the structure sequentially comprises the following steps: the first convolution layer → the first pooling layer → the second convolution layer → the second pooling layer → the third pooling layer → the Flatten layer → the first fully-connected layer → the second fully-connected layer → the Reshape layer → the first upsampled layer → the fourth convolution layer → the second upsampled layer → the fifth convolution layer → the third upsampled layer → the sixth convolution layer, wherein the first 8 layers constitute the encoder and the second 8 layers constitute the decoder.
The number of convolution kernels in the first convolution layer to the sixth convolution layer is set to be 16, 8, 8, 8, 16 and 1 respectively, and the sizes of the convolution kernels are all set to be 3
Figure 749524DEST_PATH_IMAGE008
Step length is set to 1, the activation function is a ReLU function, the filling mode selects and uses 'same' filling, the first to third pooling layers all adopt a maximum pooling mode, and the size of the pooling area core is set to 2
Figure 746298DEST_PATH_IMAGE002
Step length is set to be 1, the number of neurons of the first full connection layer and the second full connection layer is 32 and 512 respectively, the activation function is a ReLU function, and the size of an up-sampling window of the first up-sampling layer to the third up-sampling layer is 2
Figure 801979DEST_PATH_IMAGE002
2。
The mathematical model of the ReLU function is represented as follows:
Figure 353046DEST_PATH_IMAGE009
wherein,f(x)representing input values of a networkxResponse after the activation function ReLU.
And 3, constructing a fully-connected self-encoder network.
A6-layer full-connection self-encoder network is built, and the structure of the network is as follows in sequence: first hidden layer → second hidden layer → third hidden layer → fourth hidden layer → fifth hidden layer → output layer, wherein, the first 3 layers compose the encoder, and the last three layers compose the decoder.
The number of the neurons of the first hidden layer, the second hidden layer, the third hidden layer and the fifth hidden layer is respectively set to be 8, 8, 16, 8 and 8, the activation function is a ReLU function, the number of the neurons of the output layer is set to be 2, and the activation function is a sigmoid function.
The mathematical model of the ReLU function is represented as follows:
Figure 152375DEST_PATH_IMAGE009
wherein,f(x)representing input values of a networkxResponse after the activation function ReLU.
The mathematical model of the sigmiod function is represented as follows:
Figure 738077DEST_PATH_IMAGE010
wherein,g(z) Representing the response of the input value of the network after passing through the activation function sigmoid,e （。）expressed as natural constantseBottom exponential operation.
And 4, pre-training the network.
Inputting a signal time-frequency diagram of a sample in a radar radiation source data set into a convolutional self-encoder network, iteratively updating the weight of the network 100 times by using an Adam algorithm, discarding a decoder part of a trained convolutional self-encoder, and obtaining a pre-trained radar radiation source intra-pulse characteristic extraction encoder network.
Inputting the carrier frequency and the pulse width of a sample in a radar radiation source data set into a fully-connected self-encoder network, iteratively updating the weight of the network 100 times by using an Adam algorithm, discarding a decoder part of a trained fully-connected self-encoder, and obtaining a pre-trained radar radiation source inter-pulse characteristic extraction encoder network.
The Adam algorithm is as follows:
Figure 495818DEST_PATH_IMAGE011
wherein,grepresenting loss function at current iteration
Figure 799760DEST_PATH_IMAGE012
The gradient of (a) of (b) is,
Figure 160638DEST_PATH_IMAGE013
a value-assignment operation is represented and,
Figure 397584DEST_PATH_IMAGE014
the gradient operator is represented by a gradient operator,
Figure 936013DEST_PATH_IMAGE012
representing the loss function of the current iteration when the iterative training is carried out on the radar radiation source identification network,
Figure 258410DEST_PATH_IMAGE015
representing the weight of the current iteration when the radar radiation source identification network is iteratively trained,mrepresents a gradient with an initial value of 0gIs estimated by the first moment of (a) of (b),
Figure 665120DEST_PATH_IMAGE016
representing the exponential decay rate of the first moment estimate at a value of 0.9,
Figure 366360DEST_PATH_IMAGE017
represents a gradient with an initial value of 0gIs estimated by the second order moment of (a),
Figure 465903DEST_PATH_IMAGE018
representing the exponential decay rate of the second moment estimate at a value of 0.999,Tit is shown that the transpose operation,
Figure 10017DEST_PATH_IMAGE019
representing a parameter with a preset value of 0.001,
Figure 95785DEST_PATH_IMAGE020
expressed as a value of 10-8Is constant.
And 5, fusing the characteristics.
And splicing the outputs of the radar radiation source intra-pulse feature extraction encoder network and the inter-pulse feature extraction encoder network corresponding to each sample in the radar radiation source data set to obtain the joint features corresponding to the samples.
And 6, estimating the number of clusters.
And estimating the clustering number of the radar radiation source data set by using a density peak value clustering algorithm based on the joint characteristics of the samples.
The specific steps of estimating the cluster number of the radar radiation source data sets by using the density peak clustering algorithm are as follows:
firstly, calculating the distance between every two samples in a radar radiation source data set according to the following formula:
Figure 41744DEST_PATH_IMAGE021
wherein, dist: (f i ，f j ) Indicating first in radar radiation source data setiA sample andjthe distance between the individual samples of the sample,f i denotes the firstiThe joint feature vector of each sample is then calculated,f j is shown asjThe joint feature vector of each sample is then calculated,f iu is shown asiThe first of the joint features of the samplesuThe value of the characteristic is compared with the value of the characteristic,f ju is shown asjThe first of the joint features of the samplesuThe value of the characteristic is used as the characteristic value,
Figure 312188DEST_PATH_IMAGE022
which means that the square root operation is performed,
Figure 218964DEST_PATH_IMAGE023
it is indicated that the summing operation is performed,
Figure 967478DEST_PATH_IMAGE024
representing an absolute value taking operation;
secondly, arranging the sample distances obtained in the first step in ascending order, and taking the sample distance with the rank less than or equal to the top 2% and the maximum value as a truncation distanced c ；
Thirdly, calculating the local density of each sample according to the following formula:
Figure 908889DEST_PATH_IMAGE025
wherein,
Figure 350234DEST_PATH_IMAGE026
indicating first in radar source data setiLocal density of samples, F for the radar radiation source data setA joint feature matrix;
fourthly, calculating the center offset distance of each sample according to the following formula:
Figure 603361DEST_PATH_IMAGE027
wherein,
Figure 296511DEST_PATH_IMAGE028
indicating first in radar radiation source data setiA center offset of a sample;
the fifth step, respectively using the local density
Figure 685904DEST_PATH_IMAGE026
And center offset distance
Figure 563730DEST_PATH_IMAGE028
According to each sample, for the horizontal and vertical coordinates
Figure 179519DEST_PATH_IMAGE026
And
Figure 535414DEST_PATH_IMAGE028
and marking all sample points on the two-dimensional plane to obtain a decision graph of the data set, observing and counting the number of samples distributed at the upper right corner in the decision graph, wherein the number of the samples is an estimated value of the clustering data.
And 7, clustering the data.
Based on the joint characteristics of the samples, clustering the samples in the radar radiation source data set by using a k-means clustering algorithm to obtain a clustering label corresponding to each sample, and defining the clustering label as the label of the sample.
And 8, fine-tuning the network.
And (3) splicing the outputs of the two encoder networks, inputting the spliced outputs into a full connection layer and an output layer, transforming the networks into a classification network, training by using a pseudo label supervision network, and realizing fine adjustment of network parameters, wherein the number of neurons of the full connection layer is set to be 32, an activation function is a ReLU function, the number of nodes of the output layer is equal to the number of clusters obtained by estimation in the step 6, and the activation function is a softmax function.
And 9, outputting a sorting result.
And (5) repeatedly executing the step 5 to the step 8 for multiple times (15 times), and taking the data clustering result of the last time as a final radar radiation source sorting result.
The effect of the present invention is further explained by combining the simulation experiment as follows:
1. simulation conditions are as follows:
the hardware platform of the simulation experiment of the invention is as follows: the processor is Intel (R) core (TM) i9-9980HK, the main frequency is 2.40GHZ, and the memory is 64 GB.
The software platform of the simulation experiment of the invention is as follows: WINDOWS 10 operating system, MATLAB R2018a, Keras.
2. Simulation content and result analysis:
the radar radiation source samples to be sorted used in the simulation experiment are generated by MATLAB simulation, 10 different types of radar signals are simulated together, the specific parameters are shown in Table 1, and the jitter rate of all jitter parameters is 10%. The sampling frequency of the receiving system is 2.5GHz, 500 samples are simulated under the signal-to-noise ratios of 10 dB, 8 dB, 6 dB, 4 dB, 2dB, 0 dB and-2 dB of each type of radiation source, so that 35000 samples are generated in total, and 5000 samples are generated under each signal-to-noise ratio.
Figure 44893DEST_PATH_IMAGE030
Simulation experiment 1: cluster number estimation
The method utilizes sample data generated by simulation to estimate the clustering number of 5000 samples under each signal-to-noise ratio by using density peak clustering. 20 replicates were performed at each signal-to-noise ratio, and the mode of the 20 results was taken as the result of the final cluster number estimation, as shown in table 2 below. As can be seen from Table 2, when the SNR is greater than or equal to 2dB, the cluster number can be correctly estimated, and when the SNR is 0 or-2 dB, the cluster number estimated by the density peak algorithm is 14, and the estimation is wrong. From this, it can be concluded that the number of clusters estimated when the signal-to-noise ratio is high for the density peak clusters is relatively reliable.
TABLE 2 estimation of cluster population at different SNR
Signal to noise ratio/dB -2 0 2 4 6 8 10
Number of clusters 14 14 10 10 10 10 10
Simulation experiment 2: radar radiation source depth clustering sorting under different signal-to-noise ratios
In order to further verify the performance of the sorting method based on the deep clustering, experiments are carried out under the condition of different signal-to-noise ratios, the total number of correctly sorted samples of 10 different types of signals under each signal-to-noise ratio is counted, then the total number of correctly sorted samples of 10 different types of signals under each signal-to-noise ratio is divided by the total number of samples 5000 under each signal-to-noise ratio to obtain the correct rate of radar radiation source sorting under each signal-to-noise ratio, and all calculation results are drawn into a table 3.
TABLE 3 sorting accuracy List under different SNR
Signal to noise ratio/dB 10 8 6 4 2 0 -2
Accuracy rate 94.99% 90.51% 90.86% 87.8% 96.65% 90.01% 84.05%
As can be seen from Table 1, the invention is less affected by noise, and the sorting accuracy is higher than 80% under any signal-to-noise ratio condition, and the sorting accuracy can reach 84.05% even under the condition of-2 dB. Therefore, the invention effectively inhibits the influence of noise, extracts proper intra-pulse and inter-pulse characteristics and improves the performance of radar radiation source sorting.
While the invention has been described with reference to specific embodiments, the invention is not limited thereto, and various equivalent modifications or substitutions can be easily made by those skilled in the art within the technical scope of the present disclosure.

Claims (4)

1. A radar radiation source sorting method based on depth clustering is characterized by comprising the following steps:
step 1: preprocessing a radar radiation source data set to obtain a carrier frequency, a pulse width and a signal time-frequency diagram;
step 2: constructing a convolutional self-encoder network;
the convolutional self-encoder network comprises 16 layers, and the structure of the convolutional self-encoder network sequentially comprises the following layers: first convolution layer → first pooling layer → second convolution layer → second pooling layer → third convolution layer → third pooling layer → Flatten layer → first fully-connected layer → second fully-connected layer → Reshape layer → first upsampling layer → fourth convolution layer → second upsampling layer → fifth convolution layer → third upsampling layer → sixth convolution layer, wherein the first 8 layers constitute the encoder and the second 8 layers constitute the decoder;
the number of convolution kernels in the first convolution layer to the sixth convolution layer is set to be 16, 8, 8, 8, 16 and 1 respectively, and the sizes of the convolution kernels are all set to be 3
Figure 866049DEST_PATH_IMAGE001
Step length is set to 1, the activation function is a ReLU function, the filling mode selects and uses 'same' filling, the first to third pooling layers all adopt a maximum pooling mode, and the size of the pooling area core is set to 2
Figure 191857DEST_PATH_IMAGE002
Step length is set to be 1, the number of neurons of the first full connection layer and the second full connection layer is 32 and 512 respectively, the activation function is a ReLU function, and the size of an up-sampling window of the first up-sampling layer to the third up-sampling layer is 2
Figure 712968DEST_PATH_IMAGE003
2；
And step 3: constructing a fully-connected self-encoder network:
the fully-connected self-encoder network comprises 6 layers, and the structure of the fully-connected self-encoder network sequentially comprises the following layers: the first hidden layer → the second hidden layer → the third hidden layer → the fourth hidden layer → the fifth hidden layer → the output layer, wherein, the first 3 layers compose the encoder, the last three layers compose the decoder;
setting the number of neurons of the first hidden layer, the second hidden layer, the third hidden layer and the fourth hidden layer as 8, 8, 16, 8 and 8 respectively, wherein the activation function is a ReLU function, the number of neurons of the output layer is set as 2, and the activation function is a sigmoid function;
and 4, step 4: network pre-training:
inputting the signal time-frequency diagram into a convolutional self-encoder network for pre-training to obtain a feature extraction radar radiation source intra-pulse feature extraction encoder network, and inputting the carrier frequency and the pulse width into a fully-connected self-encoder network for pre-training to obtain a radar radiation source inter-pulse feature extraction encoder network;
and 5: feature fusion:
splicing the outputs of the radar radiation source intra-pulse feature extraction encoder network and the inter-pulse feature extraction encoder network corresponding to each sample in the radar radiation source data set to obtain the joint features corresponding to the samples;
step 6: and (3) estimating the number of clusters:
estimating the clustering number of the radar radiation source data set by using a density peak clustering algorithm;
and 7: data clustering:
clustering samples in a radar radiation source data set to obtain a clustering label corresponding to each sample, and defining the clustering label as a label of the sample;
and 8: network fine adjustment:
splicing the outputs of the two encoder networks, inputting the spliced outputs into a full connection layer and an output layer, transforming the networks into a classification network, training the classification network by using a pseudo label supervision network, and realizing fine adjustment of network parameters, wherein the number of neurons of the full connection layer is set to be 32, an activation function is a ReLU function, the number of nodes of the output layer is equal to the number of clusters obtained by estimation in the step 6, and the activation function is a softmax function;
and step 9: and (3) outputting a sorting result:
and (5) repeatedly executing the step (5) to the step (8) for multiple times, and taking the last data clustering result as a final radar radiation source sorting result.
2. The radar radiation source sorting method based on depth clustering according to claim 1, wherein the data preprocessing in step 1 is as follows:
1 a: intercepting radar radiation source signals within a period of time by using a reconnaissance receiver to obtain a radar radiation source data set;
1 b: measuring the carrier frequency and the pulse width of each radar radiation source signal sample in the radar radiation source data set, and normalizing the carrier frequency and the pulse width of each sample in the radar radiation source data set by using a min-max normalization method;
1 c: and calculating by using short-time Fourier transform to obtain a time-frequency graph of each sample in the radar radiation source data set, and performing min-max normalization and downsampling on the time-frequency graph to obtain a signal time-frequency graph.
3. The radar radiation source sorting method based on depth clustering according to claim 1, wherein the step 4 is as follows:
inputting the signal time-frequency diagram into a convolutional self-encoder network, iteratively updating the weight of the convolutional self-encoder network for N times by utilizing an Adam algorithm, discarding a decoder part of a trained convolutional self-encoder, and obtaining a pre-trained radar radiation source intra-pulse feature extraction encoder network;
inputting the carrier frequency and the pulse width into a fully-connected self-encoder network, iteratively updating the weight of the network for N times by using an Adam algorithm, discarding a decoder part of the trained fully-connected self-encoder, and obtaining a pre-trained radar radiation source inter-pulse characteristic extraction encoder network.
4. The method for sorting radar radiation sources based on deep clustering according to claim 1, wherein step 7 is implemented by clustering samples in a radar radiation source data set by using a k-means clustering algorithm.
CN202210500222.4A 2022-05-10 2022-05-10 Radar radiation source sorting method based on depth clustering Active CN114595732B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210500222.4A CN114595732B (en) 2022-05-10 2022-05-10 Radar radiation source sorting method based on depth clustering

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210500222.4A CN114595732B (en) 2022-05-10 2022-05-10 Radar radiation source sorting method based on depth clustering

Publications (2)

Publication Number Publication Date
CN114595732A true CN114595732A (en) 2022-06-07
CN114595732B CN114595732B (en) 2022-07-29

Family

ID=81821192

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210500222.4A Active CN114595732B (en) 2022-05-10 2022-05-10 Radar radiation source sorting method based on depth clustering

Country Status (1)

Country Link
CN (1) CN114595732B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115166678A (en) * 2022-09-06 2022-10-11 中国人民解放军海军工程大学 Complex-scene radar radiation source pulse signal sorting and cognitive countermeasure system
CN115169421A (en) * 2022-09-06 2022-10-11 中国人民解放军海军工程大学 Behavior intention cognitive system for new system radar radiation source target
CN115470873A (en) * 2022-11-14 2022-12-13 中国人民解放军国防科技大学 Radar radiation source identification method and system
CN117129947A (en) * 2023-10-26 2023-11-28 成都金支点科技有限公司 Planar transformation method radar signal identification method based on mininet
CN117972536A (en) * 2024-04-01 2024-05-03 成都大学 Pulse classification method and system

Citations (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110705618A (en) * 2019-09-24 2020-01-17 北京理工大学 Self-encoder network optimization method serving clustering tasks
CN110751121A (en) * 2019-10-28 2020-02-04 北京理工大学 Unsupervised radar signal sorting method based on clustering and SOFM
US20200125959A1 (en) * 2019-12-19 2020-04-23 Intel Corporation Autoencoder Neural Network for Signal Integrity Analysis of Interconnect Systems
CN111239692A (en) * 2020-01-10 2020-06-05 西安电子科技大学 PRI (pulse repetition index) combined intra-pulse information radiation source signal identification method based on deep learning
CN111693945A (en) * 2020-06-19 2020-09-22 清华大学 Method and system for identifying modulation mode between pulses, computer equipment and storage medium
CN112098957A (en) * 2020-09-15 2020-12-18 西安电子科技大学 Complex radar radiation source identification method based on one-dimensional self-walking convolution neural network
CN112906737A (en) * 2021-01-14 2021-06-04 无锡国芯微电子系统有限公司 Method for clustering and identifying based on density features based on multiple radiation sources
CN113156391A (en) * 2021-04-25 2021-07-23 电子科技大学 Radar signal multi-dimensional feature intelligent sorting method
CN113469236A (en) * 2021-06-25 2021-10-01 江苏大学 Deep clustering image recognition system and method for self-label learning
CN113721219A (en) * 2021-10-08 2021-11-30 中国电子科技集团公司第三十八研究所 Radar signal sorting method and system based on multi-parameter clustering
CN113759323A (en) * 2021-07-19 2021-12-07 西安电子科技大学 Signal sorting method and device based on improved K-Means combined convolution self-encoder
CN113971440A (en) * 2021-10-21 2022-01-25 中国人民解放军空军航空大学 Unsupervised radar signal sorting method based on deep clustering
US20220129758A1 (en) * 2020-10-27 2022-04-28 Raytheon Company Clustering autoencoder

Patent Citations (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110705618A (en) * 2019-09-24 2020-01-17 北京理工大学 Self-encoder network optimization method serving clustering tasks
CN110751121A (en) * 2019-10-28 2020-02-04 北京理工大学 Unsupervised radar signal sorting method based on clustering and SOFM
US20200125959A1 (en) * 2019-12-19 2020-04-23 Intel Corporation Autoencoder Neural Network for Signal Integrity Analysis of Interconnect Systems
CN111239692A (en) * 2020-01-10 2020-06-05 西安电子科技大学 PRI (pulse repetition index) combined intra-pulse information radiation source signal identification method based on deep learning
CN111693945A (en) * 2020-06-19 2020-09-22 清华大学 Method and system for identifying modulation mode between pulses, computer equipment and storage medium
CN112098957A (en) * 2020-09-15 2020-12-18 西安电子科技大学 Complex radar radiation source identification method based on one-dimensional self-walking convolution neural network
US20220129758A1 (en) * 2020-10-27 2022-04-28 Raytheon Company Clustering autoencoder
CN112906737A (en) * 2021-01-14 2021-06-04 无锡国芯微电子系统有限公司 Method for clustering and identifying based on density features based on multiple radiation sources
CN113156391A (en) * 2021-04-25 2021-07-23 电子科技大学 Radar signal multi-dimensional feature intelligent sorting method
CN113469236A (en) * 2021-06-25 2021-10-01 江苏大学 Deep clustering image recognition system and method for self-label learning
CN113759323A (en) * 2021-07-19 2021-12-07 西安电子科技大学 Signal sorting method and device based on improved K-Means combined convolution self-encoder
CN113721219A (en) * 2021-10-08 2021-11-30 中国电子科技集团公司第三十八研究所 Radar signal sorting method and system based on multi-parameter clustering
CN113971440A (en) * 2021-10-21 2022-01-25 中国人民解放军空军航空大学 Unsupervised radar signal sorting method based on deep clustering

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
DIEGO MAREZ等: ""Radar emitter and activity identification using deep clustering methods"", 《PROCEEDINGS OF SPIE》 *
YONGCHUN LI等: ""A Novel Cluster-Sorting Method for Radar Signals Based on Graph Autoencoder"", 《2021 INTERNATIONAL CONFERENCE ON CONTROL, AUTOMATION AND INFORMATION SCIENCES (ICCAIS)》 *
吉胜斌: ""基于QT的雷达辐射源信号侦察处理系统软件的设计与实现"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
李雪琼: ""基于机器学习的雷达辐射源分选与识别技术研究"", 《中国博士学位论文全文数据库 信息科技辑》 *
郝一飞: ""基于机器学习的雷达威胁分析识别"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
黄颖坤: ""深度特征选择方法及其在辐射源信号识别中的应用研究"", 《万方数据企业知识服务平台》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115166678A (en) * 2022-09-06 2022-10-11 中国人民解放军海军工程大学 Complex-scene radar radiation source pulse signal sorting and cognitive countermeasure system
CN115169421A (en) * 2022-09-06 2022-10-11 中国人民解放军海军工程大学 Behavior intention cognitive system for new system radar radiation source target
CN115166678B (en) * 2022-09-06 2022-12-20 中国人民解放军海军工程大学 Complex-scene radar radiation source pulse signal sorting and cognitive countermeasure system
CN115470873A (en) * 2022-11-14 2022-12-13 中国人民解放军国防科技大学 Radar radiation source identification method and system
CN117129947A (en) * 2023-10-26 2023-11-28 成都金支点科技有限公司 Planar transformation method radar signal identification method based on mininet
CN117129947B (en) * 2023-10-26 2023-12-26 成都金支点科技有限公司 Planar transformation method radar signal identification method based on mininet
CN117972536A (en) * 2024-04-01 2024-05-03 成都大学 Pulse classification method and system

Also Published As

Publication number Publication date
CN114595732B (en) 2022-07-29

Similar Documents

Publication Publication Date Title
CN114595732B (en) Radar radiation source sorting method based on depth clustering
CN109993280B (en) Underwater sound source positioning method based on deep learning
CN109597043B (en) Radar signal identification method based on quantum particle swarm convolutional neural network
CN109086700B (en) Radar one-dimensional range profile target identification method based on deep convolutional neural network
CN108696331B (en) Signal reconstruction method based on generation countermeasure network
CN113050042B (en) Radar signal modulation type identification method based on improved UNet3+ network
CN110222748B (en) OFDM radar signal identification method based on 1D-CNN multi-domain feature fusion
CN112949387B (en) Intelligent anti-interference target detection method based on transfer learning
CN112036239B (en) Radar signal working mode identification method and system based on deep learning network
Li et al. IncepTCN: A new deep temporal convolutional network combined with dictionary learning for strong cultural noise elimination of controlled-source electromagnetic data
CN110417694A (en) A kind of modulation mode of communication signal recognition methods
CN114019467B (en) Radar signal identification and positioning method based on MobileNet model transfer learning
CN113657491A (en) Neural network design method for signal modulation type recognition
CN108171119B (en) SAR image change detection method based on residual error network
CN113109782B (en) Classification method directly applied to radar radiation source amplitude sequence
CN110516728A (en) Polarization SAR terrain classification method based on denoising convolutional neural networks
CN116482618B (en) Radar active interference identification method based on multi-loss characteristic self-calibration network
CN114814776B (en) PD radar target detection method based on graph attention network and transfer learning
CN115114949A (en) Intelligent ship target identification method and system based on underwater acoustic signals
CN112749633A (en) Separate and reconstructed individual radiation source identification method
CN114973019A (en) Deep learning-based geospatial information change detection classification method and system
CN105809200A (en) Biologically-inspired image meaning information autonomous extraction method and device
CN116340846A (en) Aliasing modulation signal identification method for multi-example multi-label learning under weak supervision
CN116304561A (en) Ocean target magnetic anomaly signal denoising and detecting method based on deep migration learning
CN113343924B (en) Modulation signal identification method based on cyclic spectrum characteristics and generation countermeasure network

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
CP01 Change in the name or title of a patent holder

Address after: Floor 5, block B, productivity building, No. 3, dianzi West Street, high tech Zone, Xi'an, Shaanxi 710065

Patentee after: Xi'an Shengxin Technology Co.,Ltd.

Address before: Floor 5, block B, productivity building, No. 3, dianzi West Street, high tech Zone, Xi'an, Shaanxi 710065

Patentee before: XI'AN SENSING TECHNOLOGY DEVELOPMENT CO.,LTD.

CP01 Change in the name or title of a patent holder
CP02 Change in the address of a patent holder

Address after: Room 810, Building C, 8th Floor, Chuangye Building, No. 16 Gaoxin 1st Road, Xi'an City, Shaanxi Province, 710065

Patentee after: Xi'an Shengxin Technology Co.,Ltd.

Address before: Floor 5, block B, productivity building, No. 3, dianzi West Street, high tech Zone, Xi'an, Shaanxi 710065

Patentee before: Xi'an Shengxin Technology Co.,Ltd.

CP02 Change in the address of a patent holder