CN116451131A - Radar pulse sequence identification method based on self-supervision time convolution network - Google Patents

Radar pulse sequence identification method based on self-supervision time convolution network Download PDF

Info

Publication number
CN116451131A
CN116451131A CN202310297510.9A CN202310297510A CN116451131A CN 116451131 A CN116451131 A CN 116451131A CN 202310297510 A CN202310297510 A CN 202310297510A CN 116451131 A CN116451131 A CN 116451131A
Authority
CN
China
Prior art keywords
time
sequence
pulse
network
time convolution
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310297510.9A
Other languages
Chinese (zh)
Inventor
张蔚
崔健
胡星烨
杨大川
张顼
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.)
CETC 29 Research Institute
Original Assignee
CETC 29 Research Institute
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 CETC 29 Research Institute filed Critical CETC 29 Research Institute
Priority to CN202310297510.9A priority Critical patent/CN116451131A/en
Publication of CN116451131A publication Critical patent/CN116451131A/en
Pending legal-status Critical Current

Links

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
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
    • G01S7/021—Auxiliary means for detecting or identifying radar signals or the like, e.g. radar jamming signals
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • 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
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06N3/0895—Weakly supervised learning, e.g. semi-supervised or self-supervised 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)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention discloses a radar pulse sequence identification method based on a self-supervision time convolution network, which utilizes continuous pulse parameter characteristics to identify radar radiation sources, excavates and models a large number of pulse parameter sequences through a data-driven deep learning method, firstly designs a time convolution deep neural network aiming at the characteristics of the pulse sequences, then utilizes the self-supervision time convolution network to extract the characteristics of the pulse sequences and train the network, and finally obtains a trained pulse sequence classifier for identifying the radiation source types. The feature extraction structure in the time convolution network model designed by the invention can effectively extract the change features of the input pulse sequence parameters, output the type of pulse-by-pulse/pulse sequence according to the sequence relation, and improve the recognition effect of the model.

Description

Radar pulse sequence identification method based on self-supervision time convolution network
Technical Field
The invention relates to the technical field of radar radiation source pulse sequence identification, in particular to a radar pulse sequence identification method and system based on a self-supervision time convolution network.
Background
The radar radiation source signal characteristic parameters obtained by reconnaissance are utilized to identify the radiation source, and the radar radiation source signal characteristic parameters occupy an important position in electronic warfare.
The traditional radar radiation source signal identification method often needs to establish a template matching database, has high requirements on professional knowledge, and the manually established parameter database cannot ensure the identification applicable to most types of signals, and cannot give consideration to the identification precision and the identification speed. However, the conventional machine learning method needs to rely on a great deal of manual feature extraction and priori knowledge, and is difficult to deal with timing problems.
Disclosure of Invention
Aiming at the defects, the invention designs a distributed multifunctional sensor resource management method based on a function switching time sequence, designs a deep convolution network aiming at a pulse sequence, and completes radiation source identification.
In order to achieve the above purpose, the invention adopts a radar pulse sequence identification method based on a self-supervision time convolution network, which comprises the following steps:
s1: acquiring sequence data;
s2: removing abnormal pulses;
s3: self-supervision time convolution classification network design; wherein the network comprises a time convolution encoder, a time convolution decoder and a time sequence target classifier
S4: training a model; the original pulse parameter sequence is firstly processed by a time convolution encoder, extracted by a multi-layer deep neural network to obtain a high-dimensional time sequence feature vector, and then the high-dimensional vector is sent to a time convolution decoder and a time sequence target classifier
S5: model identification; the time sequence convolutional encoder and the time sequence target classifier which are trained are selected to form a time convolutional network for recognition.
Optionally, the step S1 specifically includes: pulse signals of the radar radiation source are obtained for a period of time through the radar reconnaissance equipment, and a data set is constructed by using parameter measurement data of the pulses.
Optionally, the dataset contains pulse-descriptors sequences and other target features.
Optionally, step S2 specifically includes: and judging the pulse with out-of-tolerance measurement error and abnormality according to the system characteristics and needs, and eliminating or normalizing the pulse which affects training and identification.
Optionally, in step S3, the encoder and the decoder have symmetrical network structures, which are respectively composed of an input layer, an output layer and a time convolution layer, and the multi-level deep neural network structure is used for extracting the high-dimensional time sequence characteristics of the complex pulse parameter sequence.
Optionally, in step S3, the time convolution layer is formed by stacking a plurality of time convolution residual modules, a one-dimensional cavity convolution structure design is adopted in the residual modules, the one-dimensional cavity convolution structure design is used for modeling the dependency relationship between the front and back of the time sequence information, and different residual blocks are provided with different cavity coefficients and are used for accelerating the calculation between the neural network layers.
Optionally, in step S4, after the high-dimensional vector is sent to the time sequence target classifier, a target classification result is obtained, and cross entropy loss is calculated with the tag information of the sequenceWhere n is the number of classifications, y i Is the tag value, y i ′ Is the predicted value output by the neural network; the high-dimensional vector is sent into a time convolution decoding network to reconstruct an original pulse parameter sequence, and reconstruction error loss is calculated by the original pulse parameter sequence>Where m is the number of samples in a batch, x i To input pulse parameter sequences, x i ′ To reconstruct the pulse parameter sequence.
Optionally, in step S5, after model training is completed, a trained time sequence convolutional encoder and a time sequence target classifier are selected to form a time convolutional network, and in the identification process, only the trained time convolutional encoder is used to calculate the high-dimensional characteristics of the pulse sequence, and then the time sequence target classifier is used to calculate the output probability of each type, so as to obtain the final identification result.
In order to achieve the above object, the present application further provides a radar pulse sequence identification system based on a self-supervised time convolution network, the system comprising:
the sequence data acquisition module is used for acquiring sequence data;
the abnormal pulse removing module is used for removing abnormal pulses;
the self-supervision time convolution classification network design module is used for self-supervision time convolution classification network design; wherein the network comprises a time convolution encoder, a time convolution decoder and a time sequence target classifier
The model training module is used for model training; the original pulse parameter sequence is firstly processed by a time convolution encoder, extracted by a multi-layer deep neural network to obtain a high-dimensional time sequence feature vector, and then the high-dimensional vector is sent to a time convolution decoder and a time sequence target classifier
The model identification module is used for model identification; the time sequence convolutional encoder and the time sequence target classifier which are trained are selected to form a time convolutional network for recognition.
Compared with the prior art, the invention has the beneficial effects that: compared with a single pulse identification method, the method can fully utilize time sequence associated information in the pulse parameter change process, and a radiation source identification model has more discrimination; compared with the traditional parameter matching method, the method has stronger robustness and stronger adaptability to noise, deletion and the like in the sequence; compared with the traditional sequence deep learning models such as RNN/LSTM, the method effectively avoids gradient disappearance or gradient explosion in the training process through residual connection and cavity convolution. Under the condition of long-time sequence input, the number of network parameters of the weight sharing TCN is smaller, the parallelization processing mode also enables the memory utilization rate to be lower, and the operand and the storage space are greatly reduced.
Drawings
FIG. 1 is a schematic diagram of a self-supervising time convolutional network identification process;
FIG. 2 is a schematic diagram of a self-supervising time convolution classification network;
FIG. 3 is a pulse sequence identification flow chart;
FIG. 4 is a graph of the self-supervising time convolution network identification accuracy rate variation;
fig. 5 is a graph of a self-supervising time convolution network training loss function change.
The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
The pulse parameter sequence contains rich time sequence characteristics, such as pulse width, pulse repetition interval, frequency and other information. Compared with single pulse parameters, the time sequence change characteristics can reflect the change trend of each parameter in a period of time, and the influence of dynamic errors caused by single-point parameter measurement is avoided, so that the pulse parameter sequence is used for identification, and the method is a feasible method.
In order to further extract the sequence characteristics of the pulse parameters, the invention introduces an improved self-supervision depth time convolution network into the identification of the radiation source signals, the input of the model is the original pulse parameter time sequence, and the high-dimensional characteristic extraction and classification identification processes are all carried out in the network, so that the incompleteness of manually selected characteristics is avoided. Meanwhile, the problem of overlarge calculated amount is solved by utilizing the parallel characteristic of the time convolution network, and a data set with larger sample size can be processed. The network can learn and extract pulse sequence characteristics autonomously, can adapt to various types of signals, has a simple model training mode, and does not need too much field knowledge modeling. In addition, the feature extraction structure in the time convolution network model designed by the invention can effectively extract the change features of the input pulse sequence parameters, output the type of pulse-by-pulse/pulse sequence according to the sequence relation and improve the recognition effect of the model.
In this embodiment, as shown in fig. 1, in order to fully extract the time sequence characteristics of the pulse parameter sequence and reduce the influence of random errors and detection errors on single pulse parameter identification, the invention designs a deep convolution network for the pulse sequence to complete radiation source identification. The specific implementation steps are as follows:
acquisition of sequence data: firstly, detecting pulse signals of a radar radiation source obtained for a period of time through radar reconnaissance equipment, constructing a data set by utilizing parameter measurement data of the pulses, wherein the data set generally comprises Pulse Description Word Sequences (PDWs), and other characteristics can be added according to requirements;
removing abnormal pulses: and judging the pulses with out-of-tolerance and abnormal measurement errors according to the system characteristics and requirements, removing or normalizing the pulses which affect training and identification, optimizing the quality of a pulse sequence, and reducing the influence of the abnormal pulses on the sequence.
Self-supervision time convolution classification network design: a self-supervised temporal convolution classification network with the architecture shown in fig. 2 was designed for training of classifier models.
The network includes a time convolution encoder, a time convolution decoder, and a timing target classifier. The encoder and decoder have symmetrical network structures, which are respectively composed of an input layer, an output layer and a time convolution layer, and the multi-level deep neural network structure is used for extracting high-dimensional time sequence characteristics of a pulse parameter sequence. The time convolution layer is formed by stacking a plurality of time convolution residual modules, and a one-dimensional cavity convolution structure design is adopted in the residual modules and is used for modeling the dependency relationship of the time sequence information. Different residual blocks are provided with different hole coefficients for accelerating calculation between neural network layers. The cavity convolution changes the size of the convolution kernel by sampling the convolved input at intervals, which is equivalent to skipping part of the input, so that the TCN can obtain a large receptive field with a relatively small layer to extract information, and complex network parameters and calculation amount are effectively reduced. The change of the PDW sequence in a period of time is utilized to jointly determine the identification output of the time sequence target classifier. The time sequence target classifier consists of an input layer, a full-connection layer and an output layer and is used for carrying out dimension reduction classification on high-dimension pulse sequence features and outputting a final classification result.
Model training: the original pulse parameter sequence is firstly subjected to a time convolution encoder, extracted by a multi-layer deep neural network to obtain a high-dimensional time sequence feature vector, and then the high-dimensional time sequence feature vector is sent into two branch networks to cooperatively perform supervised training. The first branch is input into a time sequence target classifier to obtain a target classification result, and the target classification result and the label information of the sequence are used for calculating cross entropy lossWhere n is the number of classifications, y i Is the tag value, y i ′ Is the predicted value output by the neural network. The other branch is introduced into a time convolution decoding network to reconstruct the original pulse parameter sequence, and the reconstruction error loss is calculated by the original pulse parameter sequence>Where m is the number of samples in a batch, x i To input pulse parameter sequences, x i ′ To reconstruct the pulse parameter sequence. The branch circuit further enhances the depicting capability of the extracted high-dimensional features on the original time sequence information mainly through a self-supervision mode. The loss functions of the two branches are combined into the whole trainingLoss function loss=l of the training network CE +L MSE And through the inverse gradient propagation of the neural network, the overall loss function is minimized, and a converged model is obtained through repeated iterative training.
Model identification: after model training is completed, a trained time sequence convolution encoder and a time sequence target classifier are selected to form a time convolution network, and fig. 3 is a process and a flow chart for identifying the whole pulse sequence:
in the identification process, only a trained time convolution encoder is used for calculating the high-dimensional characteristics of the pulse sequence, and then the time sequence target classifier is used for calculating the output probability of each type, so that the final identification result is obtained.
The invention adopts a pulse sequence identification method based on a time convolution network. The method designs a self-supervision time convolution classification network, can fully extract the change characteristics of a pulse Parameter (PDW) sequence in the time dimension, implicitly describes the change rule of the PDW in time in a data driving mode, solves the excessive dependence on parameter values in a single-pulse identification method, makes up the defect that the single-pulse identification method cannot model the change rule of time sequences, and avoids the influence of single-point detection errors on identification results.
Meanwhile, in the double-branch training process adopted by the method, the robustness of feature extraction is improved through double supervision training of the pulse sequence and the data tag. The special causal hole convolution structure in the designed time convolution network outputs the information containing rich long-term tracking, can obtain longer sequence dependency relationship, and effectively models the change of PDW. Meanwhile, compared with a cyclic neural network RNN (RNN-sequenced) processing architecture, the weight sharing characteristic of the convolutional network can regard PDW sequence data as a whole, and all data in the PDW sequence data can be processed in parallel, so that the calculation time of a model is greatly reduced.
In some specific examples, the invention provides a pulse sequence identification method based on a self-supervision time convolution network, which can effectively and rapidly identify the type of a pulse signal by utilizing pulse parameter sequence data.
In order to prove the effectiveness of the method, the embodiment verifies actual measurement data of the method in an electronic reconnaissance project, firstly, 10 radar actual measurement data of different types are used for manufacturing a data set according to the method of the step (1) and the step (2), and the total sample set is divided into a training set and a testing set, wherein the training set accounts for 75% and the testing set accounts for 25%. The training set contains 30000 PDW pulse data, including PRI, frequency, pulse width, and other main parameter information.
Then, a self-supervision time convolution depth neural network is developed and designed according to the network structure shown in fig. 2, the number of residual blocks of the time convolution layer is set to be 4, the time sequence feature vector is 64 dimensions, and the pulse sequence length is 25. The network model was trained using a number of pulse data parameter features, training 200 epochs until the network converged. The change of the recognition accuracy and the loss function in the self-supervision time convolution network training process are shown in fig. 4 and 5.
After training, the test set is processed according to the processing flow of fig. 1, recognition results are counted, and meanwhile, a plurality of mainstream machine learning and deep learning models are utilized to carry out comparison tests on the same test set. The recognition effect of the various models is shown in table 1.
The pulse sequence recognition method based on the self-supervision time convolution network has good recognition performance.
In summary, the method provided by the invention can fully extract the change characteristics of the pulse parameter sequence on time sequence, and accordingly, the quick and accurate type identification is realized, and the effectiveness of the invention is proved.
Table 1 main stream intelligent learning algorithm recognition accuracy rate comparison table
The method utilizes continuous pulse parameter characteristics to identify radar radiation sources, a large number of pulse parameter sequences are mined and modeled through a data-driven deep learning method, a time convolution deep neural network is designed according to the characteristics of the pulse sequences, then the self-supervision time convolution network is utilized to extract the characteristics of the pulse sequences and train the network, and finally a trained pulse sequence classifier is obtained and used for identifying the radiation source types.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.
The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) as described above, comprising instructions for causing a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to perform the method according to the embodiments of the present invention.
The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the scope of the invention, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.

Claims (9)

1. A method for identifying a radar pulse sequence based on a self-supervised time convolution network, the method comprising the steps of:
s1: acquiring sequence data;
s2: removing abnormal pulses;
s3: self-supervision time convolution classification network design; wherein the network comprises a time convolution encoder, a time convolution decoder and a time sequence target classifier
S4: training a model; the original pulse parameter sequence is firstly processed by a time convolution encoder, extracted by a multi-layer deep neural network to obtain a high-dimensional time sequence feature vector, and then the high-dimensional vector is sent to a time convolution decoder and a time sequence target classifier
S5: model identification; the time sequence convolutional encoder and the time sequence target classifier which are trained are selected to form a time convolutional network for recognition.
2. The method for identifying a radar pulse sequence based on a self-supervised time convolution network as set forth in claim 1, wherein the step S1 specifically includes: pulse signals of the radar radiation source are obtained for a period of time through the radar reconnaissance equipment, and a data set is constructed by using parameter measurement data of the pulses.
3. The method for radar pulse train identification based on a self-supervised time convolution network as defined in claim 2, wherein the data set contains pulse descriptor sequences and other target features.
4. The method for identifying a radar pulse sequence based on a self-supervised time convolution network as set forth in claim 1, wherein the step S2 specifically includes: and judging the pulse with out-of-tolerance measurement error and abnormality according to the system characteristics and needs, and eliminating or normalizing the pulse which affects training and identification.
5. The method for identifying radar pulse sequences based on self-supervised time convolution network as defined in claim 1, wherein in step S3, the encoder and decoder have symmetrical network structures respectively composed of an input layer, an output layer and a time convolution layer, and a multi-level deep neural network structure is used for extracting high-dimensional time sequence characteristics of the complex pulse parameter sequences.
6. The method for identifying radar pulse sequences based on a self-supervision time convolution network according to claim 5, wherein in the step S3, the time convolution layer is formed by stacking a plurality of time convolution residual modules, a one-dimensional cavity convolution structure design is adopted in the residual modules for modeling the dependency relationship between the front and the back of the time sequence information, and different cavity coefficients are set for different residual blocks for accelerating the calculation between the neural network layers.
7. The method for recognizing radar pulse sequence based on self-monitoring time convolution network as defined in claim 1, wherein in step S4, after high-dimensional vector is sent into time sequence target classifier, target classification result is obtained, and cross entropy loss is calculated with label information of the sequenceWhere n is the number of classifications, y i Is the tag value, y i ′ Is the predicted value output by the neural network; the high-dimensional vector is sent into a time convolution decoding network to reconstruct an original pulse parameter sequence, and reconstruction error loss is calculated by the original pulse parameter sequence>Where m is the number of samples in a batch, x i To input pulse parameter sequences, x i ′ To reconstruct the pulse parameter sequence.
8. The method for recognizing radar pulse sequence based on self-supervision time convolution network as defined in claim 1, wherein in step S5, after model training is completed, a trained time sequence convolution encoder and a time sequence target classifier are selected to form a time convolution network, during recognition, only the trained time convolution encoder is used for calculating high-dimensional characteristics of pulse sequence, and then the time sequence target classifier is used for calculating output probabilities of various types to obtain final recognition results.
9. A radar pulse sequence identification system based on a self-supervising time convolution network, the system comprising:
the sequence data acquisition module is used for acquiring sequence data;
the abnormal pulse removing module is used for removing abnormal pulses;
the self-supervision time convolution classification network design module is used for self-supervision time convolution classification network design; wherein the network comprises a time convolution encoder, a time convolution decoder and a time sequence target classifier
The model training module is used for model training; the original pulse parameter sequence is firstly processed by a time convolution encoder, extracted by a multi-layer deep neural network to obtain a high-dimensional time sequence feature vector, and then the high-dimensional vector is sent to a time convolution decoder and a time sequence target classifier
The model identification module is used for model identification; the time sequence convolutional encoder and the time sequence target classifier which are trained are selected to form a time convolutional network for recognition.
CN202310297510.9A 2023-03-24 2023-03-24 Radar pulse sequence identification method based on self-supervision time convolution network Pending CN116451131A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310297510.9A CN116451131A (en) 2023-03-24 2023-03-24 Radar pulse sequence identification method based on self-supervision time convolution network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310297510.9A CN116451131A (en) 2023-03-24 2023-03-24 Radar pulse sequence identification method based on self-supervision time convolution network

Publications (1)

Publication Number Publication Date
CN116451131A true CN116451131A (en) 2023-07-18

Family

ID=87129414

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310297510.9A Pending CN116451131A (en) 2023-03-24 2023-03-24 Radar pulse sequence identification method based on self-supervision time convolution network

Country Status (1)

Country Link
CN (1) CN116451131A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117972536A (en) * 2024-04-01 2024-05-03 成都大学 Pulse classification method and system

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117972536A (en) * 2024-04-01 2024-05-03 成都大学 Pulse classification method and system

Similar Documents

Publication Publication Date Title
Kwon et al. Beta shapley: a unified and noise-reduced data valuation framework for machine learning
CN111079683B (en) Remote sensing image cloud and snow detection method based on convolutional neural network
CN107832778B (en) Same target identification method based on spatial comprehensive similarity
CN106570464A (en) Human face recognition method and device for quickly processing human face shading
CN111027576B (en) Cooperative significance detection method based on cooperative significance generation type countermeasure network
CN112016313B (en) Spoken language element recognition method and device and warning analysis system
CN111368920A (en) Quantum twin neural network-based binary classification method and face recognition method thereof
CN112307820B (en) Text recognition method, device, equipment and computer readable medium
CN113065525B (en) Age identification model training method, face age identification method and related device
CN113469088A (en) SAR image ship target detection method and system in passive interference scene
CN110956309A (en) Flow activity prediction method based on CRF and LSTM
CN111046979A (en) Method and system for discovering badcase based on small sample learning
CN114048468A (en) Intrusion detection method, intrusion detection model training method, device and medium
CN116451131A (en) Radar pulse sequence identification method based on self-supervision time convolution network
CN114821340A (en) Land utilization classification method and system
CN115759254A (en) Question-answering method, system and medium based on knowledge-enhanced generative language model
CN115952724A (en) Method, system, equipment and medium for predicting residual life of aircraft engine
CN116756594A (en) Method, system, equipment and medium for detecting abnormal points of power grid data
CN116977834B (en) Method for identifying internal and external images distributed under open condition
CN114741487B (en) Image-text retrieval method and system based on image-text semantic embedding
CN114357160B (en) Early rumor detection method and device based on generated propagation structural features
CN107067034B (en) Method and system for rapidly identifying infrared spectrum data classification
CN115375966A (en) Image countermeasure sample generation method and system based on joint loss function
CN113920124A (en) Brain neuron iterative segmentation method based on segmentation and error guidance
CN116912920B (en) Expression recognition method and device

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