CN113361639A - Deep learning-based radiation source signal multi-model comprehensive classification method - Google Patents

Deep learning-based radiation source signal multi-model comprehensive classification method Download PDF

Info

Publication number
CN113361639A
CN113361639A CN202110751828.0A CN202110751828A CN113361639A CN 113361639 A CN113361639 A CN 113361639A CN 202110751828 A CN202110751828 A CN 202110751828A CN 113361639 A CN113361639 A CN 113361639A
Authority
CN
China
Prior art keywords
radiation source
model
signal
models
deep learning
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
CN202110751828.0A
Other languages
Chinese (zh)
Inventor
王佳铭
洪鼎
赵兴海
臧勤
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.)
724th Research Institute of CSIC
Original Assignee
724th Research Institute of CSIC
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 724th Research Institute of CSIC filed Critical 724th Research Institute of CSIC
Priority to CN202110751828.0A priority Critical patent/CN113361639A/en
Publication of CN113361639A publication Critical patent/CN113361639A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • 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
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention relates to a deep learning-based radiation source signal multi-model comprehensive classification method, which realizes intelligent identification of radiation source radar signal types through steps of data cleaning and normalization, unsupervised clustering, neural network learning and the like according to acquired PDW data of various radar radiation source signals. By constructing a comprehensive model based on a deep convolutional neural network and a long-time and short-time memory network, the method can adapt to various complex radar radiation source signals, high-precision and intelligent identification of the radiation source signal types is realized, and meanwhile, the model has the capabilities of self optimization and multi-scene generalization.

Description

Deep learning-based radiation source signal multi-model comprehensive classification method
Technical Field
The invention belongs to the field of data processing in the field of passive detection, relates to PDW data preprocessing and artificial intelligence analysis technology, and can be applied to intelligent analysis and identification of radiation source radar signals.
Background
The radar signal sorting is one of important contents of radar anti-reconnaissance signal processing, and is also a premise and a basis for radar feature extraction, identification and threat assessment. The radar countermeasure receiver outputs to the signal processing system a densely overlapping stream of pulses (i.e., full pulses, each pulse represented in the form of a pulse description word PDW). Signal sorting is the process of separating each radar pulse train from such a randomly overlapping pulse stream. The signal sorting is realized by utilizing the correlation of the same radar signal parameter and the difference of different radar signal parameters, the characteristic parameters which can be generally used for sorting comprise instantaneous parameters and secondary parameters, and the instantaneous parameters are parameters which can be obtained by one-time measurement and comprise a signal arrival Direction (DOA), a pulse arrival Time (TOA), a carrier frequency (RF), a Pulse Amplitude (PA), a Pulse Width (PW) and the like; the secondary parameters are parameters that can be obtained after a plurality of measurements, and include a Pulse Repetition Interval (PRI), an Antenna Scanning Period (ASP), and the like.
Traditional signal sorting algorithms such as an extended correlation method, a difference histogram method, a PRI transformation method and a TOA folding sorting method are all analyzed aiming at one or more signal characteristics, and after sorting results are generated, signal source types are determined in a mode of comparing with known data, so that the method has a plurality of limitations: firstly, signal parameters of different radars are overlapped to different degrees and can be distinguished only by comprehensive analysis; secondly, the traditional algorithm needs to adjust parameters repeatedly according to each application scene, and the generalization performance is weak; thirdly, the signal source devices of the same type have different hardware conditions, and the parameters of the transmitted signals fluctuate, which causes difficulty in data comparison; finally, the hardware cost and the time cost required by comparison are correspondingly improved along with the gradual expansion of the database.
Disclosure of Invention
In order to solve the problems existing in the traditional method, the invention provides a deep learning-based radiation source signal multi-model comprehensive classification method, which comprises the following 4 steps:
step 1, converting the characteristics of different dimensions into a unified scale range in a batch normalization mode, and avoiding the phenomenon that the wrong characteristics are learned by a network due to too great numerical value difference;
step 2, distinguishing radar signals of different radiation sources by an unsupervised clustering method, and further rejecting error data;
step 3, judging whether the signal has a threat by using a full-connection neural network, if no threat exists, directly outputting an identification result, and if a threat signal exists, performing step 4 to further identify;
and 4, the distinguished samples are sent into 18 layers of residual convolutional neural networks and long and short time memory networks one by one to carry out feature extraction and analysis, and the model synthesis method is used for integrating the output of the two models to identify the type of the radar signal of the radiation source.
Further, the method for integrating the model in the step 4 comprises the following steps: and (4) synthesizing the judgment of the two models by adopting a secondary integration method, taking the classification confidence coefficient vectors output by the two models as input parameters, and comprehensively learning the commonality and difference between the two complex models in the step (4) by using a full-connection network to output a final classification result.
In conclusion, through the secondary integration of two algorithms, the multi-model comprehensive classification method for the radar signals of the radiation source is constructed, the recognition precision of more than 95% is achieved in a 10-classification task, the defects of multi-dimensional parameter information splitting, poor environment adaptability and insufficient anti-interference capability of a traditional method are overcome, and the method has the characteristics of strong model optimization capability, good universality and the like.
Drawings
Fig. 1 is a flow chart of a radiation source signal classification system.
Fig. 2 is a schematic diagram of a residual convolutional neural network structure.
Fig. 3 is a schematic diagram of a deep fully-connected network structure.
FIG. 4 is a schematic diagram of a long term memory network.
Detailed Description
The invention is further explained by the embodiments in conjunction with the drawings. The invention provides a deep learning-based radiation source signal multi-model comprehensive classification method, which is explained in the following implementation process by combining with an embodiment:
step 1, transforming the features of different dimensions to a uniform scale range in a batch normalization mode: performing visual analysis and cleaning screening on pulse arrival Time (TOA), carrier frequency (RF), Pulse Amplitude (PA), Pulse Width (PW) and Pulse Repetition Interval (PRI) of signal source PDW data by a visual method; the characteristics of different dimensions are converted into a uniform scale range in a batch normalization mode, so that the phenomenon that the wrong characteristics are learned by a network due to too great numerical value difference is avoided. The data normalization adopts a mean standard deviation method, and the formula is as follows:
Figure BDA0003144975980000021
where μ, σ represents the mean and standard deviation of all pdw data, all data are re-normalized each time new data is added to the training.
Step 2, distinguishing signals of different signal sources by an unsupervised clustering method: and distinguishing the aliased PDW signals through a kmeans mean value clustering algorithm, and if a certain sample signal in a clustering result is too few, rejecting the aliasing PDW signals as noise. Let X be X, which includes n objects, X ═ X1，X2，X3…，XnEach object having attributes of m dimensions. The Kmeans algorithm first initializes k cluster centers and then calculates the euclidean distance of each object to each center as shown in the following equation:
Figure BDA0003144975980000022
in the above formula, XiDenotes the ith object, CjDenotes the jth cluster center, XitThe t-th attribute, C, representing the ith objectjtThe t-th attribute representing the j-th cluster center.
Step 3, judging whether the target has a threat by using a full-connection neural network: and performing two-classification processing on each sample through a three-layer fully-connected neural network to identify whether the target is threatened. If no threat exists, directly outputting an identification result, and if a threat signal exists, performing step 4 to further identify; the fully connected model structure is shown in figure 3. The full-connection network has three layers, each layer comprises 500 parameters and 1500 parameters, the output layer has two parameters which respectively correspond to the probability of threat or no threat of a sample, the sum of the two probability values is equal to 1, and the parameter with the larger probability is taken as the output of the module.
And 4, sending the distinguished samples into 18 layers of residual convolutional neural networks and long-time and short-time memory networks one by one for feature extraction and analysis, and integrating the outputs of the two models to identify the type of a signal source.
The target intelligent identification is carried out through a convolutional neural network and a long-term and short-term memory network, and models of the convolutional neural network and the long-term and short-term memory network are shown in attached figures 3 and 4. In the embodiment, an 18-layer one-dimensional residual convolutional neural network is built, compared with the traditional data analysis method, the 18-layer neural network has stronger characteristic analysis and expression capability, and the residual structure is formed by adding cross-layer connections in different levels of the network. Taking the residual block in FIG. 2 as an example, let the input be xlThe forward propagation calculation is F (x)l，ωl) Then the output x of this residual blockl′＝F(xl，ωl)+xlThen the corresponding back propagation calculation is shown as follows:
Figure BDA0003144975980000031
the 18-layer convolutional neural network comprises 33161024 parameters, the long-term memory network comprises 2560 parameters, and the final output is a vector of 1 × 10, corresponding to the probability of 10 signal types. And splicing the output vectors of the two models into a 2 x 10 matrix, inputting the other fully connected model, calculating, outputting a 1 x 10 matrix, performing softmax transformation on the matrix, and taking the category corresponding to the highest probability value as final output.

Claims (2)

1. A radiation source signal multi-model comprehensive classification method based on deep learning is characterized by comprising the following steps:
step 1, changing the characteristics of different dimensions into a uniform scale range in a batch normalization mode;
step 2, distinguishing radar signals of different radiation sources by an unsupervised clustering method, and further rejecting error data;
step 3, judging whether the signal has a threat by using a full-connection neural network, if no threat exists, directly outputting an identification result, and if a threat signal exists, performing step 4 to further identify;
and 4, the distinguished samples are sent into 18 layers of residual convolutional neural networks and long and short time memory networks one by one to carry out feature extraction and analysis, and the model synthesis method is used for integrating the output of the two models to identify the type of the radar signal of the radiation source.
2. The deep learning-based radiation source signal multi-model comprehensive classification method according to claim 1, characterized in that: the method for synthesizing the model in the step 4 comprises the following steps: and (4) synthesizing the judgment of the two models by adopting a secondary integration method, taking the classification confidence coefficient vectors output by the two models as input parameters, and comprehensively learning the commonality and difference between the two complex models in the step (4) by using a full-connection network to output a final classification result.
CN202110751828.0A 2021-07-02 2021-07-02 Deep learning-based radiation source signal multi-model comprehensive classification method Pending CN113361639A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110751828.0A CN113361639A (en) 2021-07-02 2021-07-02 Deep learning-based radiation source signal multi-model comprehensive classification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110751828.0A CN113361639A (en) 2021-07-02 2021-07-02 Deep learning-based radiation source signal multi-model comprehensive classification method

Publications (1)

Publication Number Publication Date
CN113361639A true CN113361639A (en) 2021-09-07

Family

ID=77538008

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110751828.0A Pending CN113361639A (en) 2021-07-02 2021-07-02 Deep learning-based radiation source signal multi-model comprehensive classification method

Country Status (1)

Country Link
CN (1) CN113361639A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113962261A (en) * 2021-10-21 2022-01-21 中国人民解放军空军航空大学 Depth network model for radar signal sorting

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113962261A (en) * 2021-10-21 2022-01-21 中国人民解放军空军航空大学 Depth network model for radar signal sorting
CN113962261B (en) * 2021-10-21 2024-05-14 中国人民解放军空军航空大学 Deep network model construction method for radar signal sorting

Similar Documents

Publication Publication Date Title
CN113156391B (en) Radar signal multi-dimensional feature intelligent sorting method
Petrov et al. Radar emitter signals recognition and classification with feedforward networks
CN110109095B (en) Target feature assisted multi-source data association method
CN113050797A (en) Method for realizing gesture recognition through millimeter wave radar
Xiao et al. Specific emitter identification of radar based on one dimensional convolution neural network
CN112684427A (en) Radar target identification method based on serial quadratic reinforcement training
CN114742102A (en) NLOS signal identification method and system
CN115438708A (en) Classification and identification method based on convolutional neural network and multi-mode fusion
Nuhoglu et al. Image segmentation for radar signal deinterleaving using deep learning
CN112990125B (en) Method for judging whether radiation source radar belongs to target platform
CN113361639A (en) Deep learning-based radiation source signal multi-model comprehensive classification method
CN113887583A (en) Radar RD image target detection method based on deep learning under low signal-to-noise ratio
CN112213697B (en) Feature fusion method for radar deception jamming recognition based on Bayesian decision theory
CN113064489A (en) Millimeter wave radar gesture recognition method based on L1-Norm
CN116340533B (en) Satellite-borne electromagnetic spectrum big data intelligent processing system based on knowledge graph
Li et al. Sea/land clutter recognition for over-the-horizon radar via deep CNN
Chen et al. Function recognition of multi-function radar via cnn-gru neural network
Jordanov et al. Sets with incomplete and missing data—NN radar signal classification
CN115481667A (en) UWB fuze target identification method based on multi-scale UDA and feature learning
CN112666528B (en) Multi-station radar system interference identification method based on convolutional neural network
Petrov et al. Identification of radar signals using neural network classifier with low-discrepancy optimisation
Zhang et al. An Incremental Recognition Method for MFR Working Modes Based on Deep Feature Extension in Dynamic Observation Scenarios
CN115620172B (en) Intelligent comprehensive identification method for marine ship target based on cross-domain multi-feature
Hughes et al. Automatic target recognition: Problems of data separability and decision making
CN116908806B (en) Multi-dimensional feature target detection method based on self-supervised learning

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
CB02 Change of applicant information
CB02 Change of applicant information

Address after: 210003 No. 346, Zhongshan North Road, Jiangsu, Nanjing

Applicant after: 724 Research Institute of China Shipbuilding Corp.

Address before: 210003 No. 346, Zhongshan North Road, Jiangsu, Nanjing

Applicant before: 724TH RESEARCH INSTITUTE OF CHINA SHIPBUILDING INDUSTRY Corp.