WO2023092923A1 - 复合干扰信号识别方法和系统 - Google Patents

复合干扰信号识别方法和系统 Download PDF

Info

Publication number
WO2023092923A1
WO2023092923A1 PCT/CN2022/084693 CN2022084693W WO2023092923A1 WO 2023092923 A1 WO2023092923 A1 WO 2023092923A1 CN 2022084693 W CN2022084693 W CN 2022084693W WO 2023092923 A1 WO2023092923 A1 WO 2023092923A1
Authority
WO
WIPO (PCT)
Prior art keywords
domain
features
neural network
interference signal
input
Prior art date
Application number
PCT/CN2022/084693
Other languages
English (en)
French (fr)
Inventor
边丽蘅
刘思田
张军
Original Assignee
北京理工大学
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 北京理工大学 filed Critical 北京理工大学
Publication of WO2023092923A1 publication Critical patent/WO2023092923A1/zh

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
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield 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
    • G06N3/045—Combinations of networks
    • 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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/22—Source localisation; Inverse modelling

Definitions

  • the present disclosure relates to the field of satellite navigation signal processing, in particular to a method and system for identifying satellite navigation composite interference signals.
  • the navigation satellites of the Beidou system are located at an altitude of 10,000 meters. When the satellite signals reach the ground after long-distance transmission, the signal power is already very weak. As long as there is a little interference in the frequency band where the satellite signals are located, the normal operation of the navigation signal receiver will be affected. Therefore It is necessary to monitor the complex electromagnetic environment of various receivers in the satellite navigation system.
  • the satellite navigation interference monitoring technology mainly includes the direction finding and positioning technology of the interference source, interference detection and alarm technology, interference signal spectrum feature extraction and interference type identification technology.
  • the characteristics of the interference signal are nothing more than extracted from the time domain, frequency domain, and time-frequency domain of the signal.
  • Demirkiran et al. use the short-time Fourier transform method to extract features and classify and identify single-tone interference, multi-tone interference, and chirp interference.
  • Yang Xiaoming and others extracted features from the Welch periodogram method and fractional Fourier transform domain to identify the interference signal in the direct sequence spread spectrum system.
  • Classification algorithms are generally carried out on the premise of feature extraction. Most of the algorithms use existing and commonly used traditional machine learning classification algorithms such as support vector machines, decision trees, and BP neural networks.
  • Angelov et al. studied unsupervised learning and used cluster analysis to classify and identify interference signals in the uplink of 3G networks.
  • the present disclosure aims to solve one of the technical problems in the related art at least to a certain extent.
  • the purpose of this disclosure is to solve the problem of fast and accurate identification of satellite navigation composite interference signals, and propose a composite interference signal identification method and system.
  • the first aspect of the present disclosure proposes a composite interference signal identification method, the method comprising:
  • the extracted domain features of different dimensions are input into the feature fusion layer and the fully connected layer of the pre-trained deep learning neural network model to obtain the classification and recognition results of the composite interference signal.
  • the composite interference signal identification method of the embodiment of the present disclosure preprocesses the composite interference signal, extracts the multi-domain features of the signal, respectively inputs the pre-trained deep learning neural network, and outputs the classification and recognition results of the composite interference signal.
  • the disclosure improves the classification and identification accuracy of the composite interference signal, and the calculation process is simple, convenient and quick.
  • the preprocessing of the compound interference signal to be identified includes:
  • Absolute value processing, normalization processing, filtering and denoising and signal down-conversion processing are performed on the composite interference signal to be identified.
  • the multi-domain includes: time domain, frequency domain, time-frequency domain and spatial domain features.
  • the pre-trained deep learning neural network model before inputting the obtained multi-domain features into the pre-trained deep learning neural network model based on their corresponding dimensions, it also includes:
  • the pre-trained deep learning neural network model includes: LSTM layer, Attention layer, one-dimensional sequence feature extraction module and multi-dimensional sequence feature extraction module, feature fusion layer and fully connected layer;
  • the pre-trained deep learning neural network model is obtained by training an initial deep learning neural network based on obtaining multi-domain features of composite interference signals in a historical period.
  • the obtained multi-domain features are respectively input into the pre-trained deep learning neural network model based on their corresponding dimensions
  • the one-dimensional sequence features are input into the one-dimensional sequence feature extraction module
  • the multi-dimensional sequence features are input into the The multi-dimensional sequence feature extraction module then extracts domain features of different dimensions, including:
  • Time-frequency domain and space domain multi-dimensional sequence features of the complex interference signal to be identified into the multi-dimensional sequence feature extraction module of the pre-trained deep learning neural network model respectively, and obtain the time-frequency domain and space domain features.
  • Time-frequency domain and spatial domain features are the time-frequency domain and spatial domain features.
  • the input of the extracted domain features of different dimensions into the feature fusion layer and the fully connected layer of the pre-trained deep learning neural network model, to obtain the classification and recognition results of the composite interference signal including:
  • the time-domain and frequency-domain features and the time-frequency domain and space-domain features of the down-sampling of the attention weight obtained greater than the preset weight threshold are input into the feature fusion layer of the deep learning neural network model trained in advance to carry out feature fusion;
  • the fused feature data is input into the fully connected layer of the pre-trained deep learning neural network model to obtain the classification recognition result of the compound interference signal to be recognized.
  • the training process of the pre-trained deep learning neural network model includes:
  • the multi-domain features of the composite interference signal in the preprocessed historical period wherein the multi-domain features include: time domain, frequency domain, time-frequency domain and space domain features;
  • the time domain and frequency domain of the composite interference signal in the historical period after the preprocessing are respectively input into the initial deep learning neural network model, and the cross entropy is used as the loss function of the model, and the adaptive matrix estimation Adam optimization algorithm is used for all
  • the above model is trained to obtain a trained deep learning neural network model.
  • the one-dimensional sequence feature extraction module or the multi-dimensional sequence feature extraction module includes: a recurrent neural network and a convolutional neural network.
  • the second aspect of the present disclosure proposes a composite interference signal identification system, including:
  • An acquisition module configured to preprocess the composite interference signal to be identified, and acquire the multi-domain features of the preprocessed composite interference signal to be identified;
  • the extraction module is used to input the obtained multi-domain features into the pre-trained deep learning neural network model based on their corresponding dimensions, input the one-dimensional sequence features to the one-dimensional sequence feature extraction module, and input the multi-dimensional sequence features to the multi-dimensional sequence features The extraction module, and then extract the domain features of different dimensions;
  • the identification module is used to input the extracted domain features of different dimensions into the feature fusion layer and the fully connected layer of the pre-trained deep learning neural network model to obtain the classification and identification results of the composite interference signal.
  • the composite interference signal identification system of the embodiment of the present disclosure preprocesses the composite interference signal, extracts the time domain features, frequency domain features, time-frequency domain features and space domain features of the signal, respectively inputs the pre-trained deep learning neural network, and outputs Classification and identification results of composite interference signals.
  • the disclosure improves the classification and identification accuracy of the composite interference signal, and the calculation process is simple, convenient and fast.
  • a third aspect of the present disclosure provides an electronic device, including:
  • a memory a processor, and a computer program stored in the memory and operable on the processor.
  • the processor executes the program, the method as described in any one of the above-mentioned first aspects is implemented.
  • a fourth aspect of the present disclosure provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the method described in any one of the above first aspects.
  • a fifth aspect of the present disclosure provides a computer program product, including a computer program, and when the computer program is executed by a processor, the method described in any one of the above first aspects is implemented.
  • the disclosure improves the classification and identification accuracy of the composite interference signal, and the calculation process is simple, convenient and fast.
  • FIG. 1 is a flowchart of a composite interference signal identification method according to an embodiment of the present disclosure
  • FIG. 2 is a flowchart of a method for identifying satellite navigation composite interference signals based on a convolutional neural network and a long-short-term memory network according to Embodiment 1 of the present disclosure
  • FIG. 3 is a flow chart of a method for identifying complex interference signals based on a convolutional neural network and a long-short-term memory network according to Embodiment 2 of the present disclosure
  • FIG. 4 is a schematic structural diagram of a composite interference signal identification system according to an embodiment of the present disclosure.
  • FIG. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure.
  • FIG. 1 is a flowchart of a method for identifying satellite navigation composite interference signals according to an embodiment of the disclosure.
  • the satellite navigation composite interference signal identification method includes the following steps:
  • Step 1 Preprocessing the composite interference signal to be identified, and obtaining the multi-domain features of the preprocessed composite interference signal to be identified;
  • Step 2 Input the obtained multi-domain features into the pre-trained deep learning neural network model based on their corresponding dimensions, input the one-dimensional sequence features to the one-dimensional sequence feature extraction module, and input the multi-dimensional sequence features to the multi-dimensional sequence feature extraction module , and then extract domain features of different dimensions;
  • Step 3 Input the extracted domain features of different dimensions into the feature fusion layer and fully connected layer of the pre-trained deep learning neural network model to obtain the classification and recognition results of the composite interference signal.
  • the pre-trained deep learning neural network model includes: LSTM layer, Attention layer, one-dimensional sequence feature extraction module and multidimensional sequence feature extraction module, feature fusion layer and fully connected layer;
  • the pre-trained deep learning neural network model is obtained by training an initial deep learning neural network based on obtaining multi-domain features of composite interference signals in a historical period.
  • the satellite navigation composite interference signal identification method of the embodiment of the present disclosure preprocesses the composite interference signal, extracts the time domain features, frequency domain features, time-frequency domain features and space domain features of the signal, and inputs them into the pre-trained deep learning neural network respectively. , and output the classification and identification results of the composite interference signal.
  • the disclosure improves the classification and identification accuracy of satellite navigation composite interference signals, and the calculation process is simple, convenient and fast.
  • Fig. 2 is a flowchart of a method for identifying satellite navigation composite interference signals based on a convolutional neural network and a long-short-term memory network according to Embodiment 1 of the present disclosure.
  • the method for identifying satellite navigation composite interference signals based on convolutional neural network and long-short-term memory network includes:
  • the complex baseband signal is firstly subjected to power normalization processing, and since the interference signal is a complex signal, the absolute value processing of the signal is used as the input of the network model in the preprocessing stage .
  • the parameters are input into the neural network respectively, including:
  • time-domain features and frequency-domain features of the signal are 1-dimensional feature vectors, they must first pass through the LSTM layer, while the time-frequency domain features and spatial domain features are 2-dimensional feature vectors, so they do not need to go through the LSTM layer, and can directly pass through the corresponding convolution.
  • Neural Networks Since the time-domain features and frequency-domain features of the signal are 1-dimensional feature vectors, they must first pass through the LSTM layer, while the time-frequency domain features and spatial domain features are 2-dimensional feature vectors, so they do not need to go through the LSTM layer, and can directly pass through the corresponding convolution.
  • LSTM is a special recurrent neural network that solves the problem of gradient disappearance in traditional RNN.
  • the noise interference in historical information can be weakened through the forget gate, and important parts of historical information can be selected through the input gate.
  • the forget gate and The input gate determines the state information of the current neural network layer, and the output gate determines which information will be output.
  • the pure convolutional neural network cannot effectively analyze the characteristics and historical information of time series signals.
  • the LSTM network will save the learned important features as long-term memory, and selectively retain, update or forget the saved long-term memory according to the learning.
  • the features whose weight is always small in iterations will be regarded as short-term memory and forgotten. This mechanism makes important features transfer as the number of iterations increases, making the network perform well in classification problems that deal with long-term dependent sequences. . Therefore, adding the LSTM network can improve the accuracy of interference signal identification.
  • the input of the attention mechanism layer is the output of LSTM, which is responsible for assigning the corresponding attention weight to the features learned in the LSTM layer.
  • the Attention mechanism uses the human brain to identify the characteristics of things, focuses on the salient features of things, ignores other unimportant details, and puts limited energy and resources in important positions to improve work efficiency.
  • the Attention mechanism is used to assign attention weights to the feature vectors extracted by LSTM, and those more significant feature sets are highly summarized and then input into the convolutional neural network to generate classification results.
  • the introduction of the Attention mechanism can make the model always focus on those more significant features during training, which can not only reduce the parameters during training to improve training efficiency, but also improve the classification accuracy of the model.
  • the fusion of features can achieve the effect of feature complementarity, which can effectively improve the accuracy of signal recognition, and channel compression can more effectively calculate channel attention, and the aggregation of spatial information usually uses average pooling Or the method of max pooling.
  • the final output layer is the Softmax fully connected layer, which can effectively output the classification results.
  • the classification results are effectively output, including:
  • the convolutional neural network propagates forward through multiple operations such as convolution and pooling, and trains the network through error backpropagation.
  • the convolutional neural network is useful for regular structured data. Strong feature extraction ability, but the feature extraction ability for time series changes is weak, so before the time domain features and frequency domain features of the signal enter the convolutional neural network, it needs to go through the LSTM network and the Attention layer to form the characteristics of the interference signal.
  • the effective extraction is finally performed with the downsampled time-frequency domain features and spatial domain features for feature fusion, and then the classification is realized through the convolutional layer and the Softmax fully connected layer, thereby forming an overall model for interference signal recognition.
  • the fully connected layer normalizes the obtained feature vector through the Softmax classifier to generate a classification probability and outputs it.
  • FIG. 3 is a flow chart of a complex interference signal identification method based on a convolutional neural network and a long-short-term memory network according to Embodiment 2 of the present disclosure.
  • the convolutional neural network and long short-term memory network are based on complex interference signal identification methods, including:
  • the length of the signal received by the receiving end is 1x1024, that is, the number of sampling points of one bit signal at the receiving end is 1024, and the interference signal with an interference-to-noise ratio (JNR) of -5dB ⁇ 14dB is taken as the receiving signal, and the JNR interval is 1dB.
  • JNR interference-to-noise ratio
  • the one-dimensional time-domain features and frequency-domain features first pass through the LSTM layer.
  • the LSTM layer in the network structure is responsible for receiving sample data and To learn the feature information of the sample, the number of neurons is 128, L2 regularization is adopted, the L2 norm is set to 0.001, and the LSTM network is set as a bidirectional network;
  • the Attention layer is responsible for assigning attention weights to the feature set learned from the LSTM layer, using the sigmoid function, and the scoring function is a product matrix;
  • the convolutional neural network structure passed is as follows: the number of convolution kernels in the first layer is 32, and the convolution kernel size is 1*10; Resize the one-dimensional vector into a two-dimensional tensor and input it to the second layer.
  • the number of convolution kernels in the second layer is 32, and the size of the convolution kernel is 3*3; the number of convolution kernels in the third layer is 64, and the volume
  • the product kernel size is 5*5;
  • the convolutional neural network structure of the two-dimensional feature vector time-frequency domain feature and spatial domain feature is as follows: the number of convolution kernels in the first layer is 32, and the size of the convolution kernel is 3*3; the number of convolution kernels in the second layer is 64, the convolution kernel size is 5*5.
  • the four feature vectors of the convolutional neural network are fused through the method of the channel attention mechanism, and then the convolution kernel is used to compress the channel, which can calculate the channel attention more effectively.
  • the neural network training process includes:
  • this embodiment also provides a satellite navigation composite interference signal identification system 10 , the system 10 includes: an acquisition module 100 , an extraction module 200 , and an identification module 300 .
  • the acquisition module 100 is configured to preprocess the composite interference signal to be identified, and acquire the multi-domain features of the preprocessed composite interference signal to be identified;
  • the extraction module 200 is used to input the obtained multi-domain features into the pre-trained deep learning neural network model respectively based on their corresponding dimensions, input the one-dimensional sequence features to the one-dimensional sequence feature extraction module, and input the multi-dimensional sequence features to the multi-dimensional sequence Feature extraction module, and then extract domain features of different dimensions;
  • the identification module 300 is configured to input the extracted domain features of different dimensions into the feature fusion layer and the fully connected layer of the pre-trained deep learning neural network model to obtain the classification and identification result of the composite interference signal.
  • the pre-trained deep learning neural network model includes: LSTM layer, Attention layer, one-dimensional sequence feature extraction module and multi-dimensional sequence feature extraction module, feature fusion layer and fully connected layer.
  • the pre-trained deep learning neural network model is obtained by training an initial deep learning neural network based on obtaining multi-domain features of composite interference signals in a historical period.
  • the satellite navigation composite interference signal identification system of the embodiment of the present disclosure preprocesses the composite interference signal, extracts the time domain features, frequency domain features, time-frequency domain features and space domain features of the signal, and inputs them into the pre-trained deep learning neural network respectively. , and output the classification and identification results of the composite interference signal.
  • the disclosure improves the classification and identification accuracy of satellite navigation composite interference signals, and the calculation process is simple, convenient and quick.
  • the preprocessing of the composite interference signal to be identified includes:
  • Absolute value processing, normalization processing, filtering and denoising and signal down-conversion processing are performed on the composite interference signal to be identified.
  • the multi-domain includes: time domain, frequency domain, time-frequency domain and space domain features.
  • the pre-trained deep learning neural network model before inputting the obtained multi-domain features based on their corresponding dimensions to the pre-trained deep learning neural network model, it also includes:
  • the pre-trained deep learning neural network model includes: LSTM layer, Attention layer, one-dimensional sequence feature extraction module and multi-dimensional sequence feature extraction module, feature fusion layer and fully connected layer;
  • the pre-trained deep learning neural network model is obtained by training an initial deep learning neural network based on obtaining multi-domain features of composite interference signals in a historical period.
  • the extraction module 200 includes:
  • the first extraction unit is used to input the time domain and frequency domain one-dimensional sequence features after the distribution of attention weight into the one-dimensional sequence feature extraction module of the pre-established deep learning neural network model respectively, and extract the attention weight greater than Time-domain and frequency-domain characteristics of preset weight thresholds;
  • the second extraction unit is used to input the time-frequency domain and space domain multi-dimensional sequence features of the composite interference signal to be identified into the multi-dimensional sequence feature extraction module of the pre-trained deep learning neural network model respectively, and obtain the time-frequency domain
  • the time-frequency domain and spatial domain features are downsampled with the spatial domain features.
  • the identification module 300 includes:
  • the fusion unit is used to input the time-domain and frequency-domain features whose attention weight is greater than the preset weight threshold and the down-sampled time-frequency domain and space-domain features into the feature fusion layer of the pre-trained deep learning neural network model Perform feature fusion;
  • the recognition unit is configured to input the fused feature data into the fully connected layer of the pre-trained deep learning neural network model to obtain the classification recognition result of the compound interference signal to be recognized.
  • the training process of the pre-trained deep learning neural network model includes:
  • the multi-domain features of the composite interference signal in the preprocessed historical period wherein the multi-domain features include: time domain, frequency domain, time-frequency domain and space domain features;
  • the time domain and frequency domain of the composite interference signal in the historical period after the preprocessing are respectively input into the initial deep learning neural network model, and the cross entropy is used as the loss function of the model, and the adaptive matrix estimation Adam optimization algorithm is used for all
  • the above model is trained to obtain a trained deep learning neural network model.
  • the one-dimensional sequence feature extraction module or the multi-dimensional sequence feature extraction module includes: a recurrent neural network and a convolutional neural network.
  • FIG. 5 is a schematic structural diagram of an electronic device 1000 provided by an embodiment of the present disclosure.
  • the electronic device 1000 includes: a processor 1001 , a memory 1002 and a communication interface 1003 .
  • the communication interface 1003 is used to implement data exchange with external devices, and the processor 1001 can call the computer program stored in the memory 1002 to realize:
  • the extracted domain features of different dimensions are input into the feature fusion layer and the fully connected layer of the pre-trained deep learning neural network model to obtain the classification and recognition results of the composite interference signal.
  • the electronic device 1000 described in the embodiment of the present application can execute the description of the satellite navigation composite interference signal identification method in the embodiments corresponding to Figures 1 to 3 above, and can also execute the embodiment corresponding to Figure 4 above
  • the description of the satellite navigation composite interference signal identification system 10 will not be repeated here.
  • the description of the beneficial effect of adopting the same method will not be repeated here.
  • first and second are used for descriptive purposes only, and cannot be interpreted as indicating or implying relative importance or implicitly specifying the quantity of indicated technical features.
  • the features defined as “first” and “second” may explicitly or implicitly include at least one of these features.
  • “plurality” means at least two, such as two, three, etc., unless otherwise specifically defined.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Signal Processing (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Noise Elimination (AREA)
  • Image Analysis (AREA)

Abstract

本申请公开了一种复合干扰信号识别方法和系统，方法包括：将待识别的复合干扰信号进行预处理，并获取所述预处理后的待识别复合干扰信号的多域特征；将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型，一维序列特征输入到一维序列特征提取模块，多维序列特征输入到多维序列特征提取模块，然后分别对不同维度的域特征进行提取；将提取出的不同维度的域特征输入预先训练好的深度学习神经网络模型的特征融合层和全连接层中，得到所述复合干扰信号的分类识别结果。

Description

复合干扰信号识别方法和系统
相关申请的交叉引用
本申请基于申请号为202111423556.8、申请日为2021年11月26日的中国专利申请提出，并要求该中国专利申请的优先权，该中国专利申请的全部内容在此引入本申请作为参考。
技术领域
本公开涉及卫星导航信号处理领域，尤其涉及卫星导航复合干扰信号识别方法和系统。
背景技术
北斗系统的导航卫星位于万米高空，卫星信号经过长距离的传播到达地面时，信号功率已经十分微弱，只要在卫星信号所在的频段稍加干扰，就会影响导航信号接收机的正常运行，因此需要对卫星导航系统中各种接收机复杂的电磁环境进行监测，卫星导航干扰监测技术主要包括干扰源的测向定位技术、干扰检测与告警技术、干扰信号频谱特征提取及干扰类型识别技术。
干扰信号的特征不外乎就是从信号的时域、频域、和时频域等这些方面去提取出来的。Demirkiran等人利用短时傅里叶变换的方法来进行特征提取，对单音干扰、多音干扰、线性调频干扰进行分类识别。杨小明等人从Welch周期图法和分数阶傅里叶变换域上提取特征对直接序列扩频系统中的干扰信号进行识别。分类算法一般是在特征提取的前提下进行的算法，大部分算法都是用已有的常用的传统的机器学习的分类算法如支持向量机、决策树、BP神经网络。Angelov等人研究了无监督的学习，用聚类分析在3G网络上行链路的干扰信号进行分类识别。
综合上述研究现状，干扰信号识别的大部分研究集中在不同通信系统的不同干扰信号的特征提取上，对分类算法的研究比较少，这也表明了基于特征提取的干扰识别这类方法中特征提取的重要性。对于卷积神经网络这类通用性的干扰信号分类算法，还有较大的研究空间。
目前针对北斗导航中的干扰信号的识别大多是对单一干扰信号的识别，然而实际中导航卫星接收端接收的信号大多是卫星信号、噪声和多种干扰信号交织的混合信号，对混合信号中的混合干扰的识别有助于针对性的实施干扰抑制。传统的针对混合信号的识别流程主要是先用盲源信号分离法将混合信号分离成独立的信号，再将分离后的信号放入分类器进行分类，但是这种分离方法往往要求接收信号之间相互独立，并不十分实用。
因此，针对卫星导航中的复杂多变的复合干扰信号设计一种快速且精准的分类识别器尤为重要。
发明内容
本公开旨在至少在一定程度上解决相关技术中的技术问题之一。
为此，本公开的目的在于解决针对卫星导航复合干扰信号的快速精准识别问题，提出了一种复合干扰信号识别方法和系统。
为达上述目的，本公开第一方面提出了一种复合干扰信号识别方法，所述方法包括：
将待识别的复合干扰信号进行预处理，并获取所述预处理后的待识别复合干扰信号的多域特征；
将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型，一维序列特征输入到一维序列特征提取模块，多维序列特征输入到多维序列特征提取模块，然后分别对不同维度的域特征进行提取；
将提取出的不同维度的域特征输入预先训练好的深度学习神经网络模型的特征融合层和全连接层中，得到所述复合干扰信号的分类识别结果。
本公开实施例的复合干扰信号识别方法，将复合干扰信号进行预处理，提取信号的多域特征，分别输入预先训练好的深度学习神经网络，输出对复合干扰信号的分类识别结果。本公开提高了复合干扰信号的分类识别精度，且计算过程简单、方便、快捷。
在一些实施例中，所述将待识别的复合干扰信号进行预处理，包括：
将待识别的复合干扰信号进行绝对值处理、归一化处理、滤波去噪和信号下变频处理。
在一些实施例中，所述多域包括：时域、频域、时频域和空域特征。
在一些实施例中，所述将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型前还包括：
将获取的所述待识别复合干扰信号的时域和频域特征分别输入所述预先训练好的深度学习神经网络模型的LSTM层中，分别学习得到所述待识别复合干扰信号的时域和频域特征信息；
将所述学习得到的所述待识别复合干扰信号的时域和频域特征信息分别输入预先训练好的深度学习神经网络模型的Attention层中，对学习得到所述信号的时域和频域特征信息分配对应的注意力权重，得到分配注意力权重后的时域和频域一维序列特征；
其中，所述预先训练好的深度学习神经网络模型包括：LSTM层、Attention层、一维序列特征提取模块与多维序列特征提取模块、特征融合层和全连接层；
所述预先训练好的深度学习神经网络模型是基于获取历史时段内复合干扰信号的多域特征对初始深度学习神经网络进行训练得到的。
在一些实施例中，所述将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型，一维序列特征输入到一维序列特征提取模块，多维序列特征输入到多维序列特征提取模块，然后分别对不同维度的域特征进行提取，包括：
将所述分配注意力权重后的时域和频域一维序列特征分别输入预先建立的深度学习神经网络模型的一维序列特征提取模块中，提取出注意力权重大于预设的权重阈值的时域和频域特征；
将获取的所述待识别复合干扰信号的时频域和空域多维序列特征分别输入预先训练好的深度学习神经网络模型的多维序列特征提取模块中，获得对时频域和空域特征进行下采样的时频域和空域特征。
在一些实施例中，所述将提取出的不同维度的域特征输入预先训练好的深度学习神经网络模型的特征融合层和全连接层中，得到所述复合干扰信号的分类识别结果，包括：
将得到的注意力权重大于预设的权重阈值的时域和频域特征及所述下采样的时频域和空域特征输入预先训练好的深度学习神经网络模型的特征融合层进行特征融合；
将所述融合后的特征数据输入预先训练好的深度学习神经网络模型的全连接层，得到所述待识别复合干扰信号的分类识别结果。
在一些实施例中，所述预先训练好的深度学习神经网络模型的训练过程包括：
获取预处理后的历史时段内复合干扰信号的多域特征，其中所述多域特征包括：时域、频域、时频域和空域特征；
将所述预处理后的历史时段内复合干扰信号的时域、频域分别依次输入初始的深度学习神经网络模型中，将交叉熵作为模型的损失函数，用自适应矩阵估计Adam优化算法对所述模型进行训练，得到训练好的深度学习神经网络模型。
在一些实施例中，所述一维序列特征提取模块或多维序列特征提取模块，包括：递归神经网络和卷积神经网络。
为达到上述目的，本公开第二方面提出了一种复合干扰信号识别系统，包括：
获取模块，用于将待识别的复合干扰信号进行预处理，并获取所述预处理后的待识别复合干扰信号的多域特征；
提取模块，用于将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型，一维序列特征输入到一维序列特征提取模块，多维序列特征输入到多维序列特征提取模块，然后分别对不同维度的域特征进行提取；
识别模块，用于将提取出的不同维度的域特征输入预先训练好的深度学习神经网络模型的特征融合层和全连接层中，得到所述复合干扰信号的分类识别结果。
本公开实施例的复合干扰信号识别系统，将复合干扰信号进行预处理，提取信号的时域特征、频域特征、时频域特征和空域特征，分别输入预先训练好的深度学习神经网络，输出对复合干扰信号的分类识别结果。本公开提高了复合干扰信号的分类识别精度，且计算过程简单、方便、快捷。
为达上述目的，本公开第三方面提供一种电子设备，包括：
存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，处理器执行所述程序时，实现如上述第一方面中任一项所述的方法。
为达上述目的，本公开第四方面提供一种存储有计算机指令的非瞬时计算机可读存储介质，其中，计算机指令用于使计算机执行如上述第一方面中任一项所述的方法。
为达上述目的，本公开第五方面提供一种计算机程序产品，包括计算机程序，计算机程序在被处理器执行时实现如上述第一方面中任一项所述的方法。
本公开提高了复合干扰信号的分类识别精度，且计算过程简单、方便、快捷。
本公开附加的方面和优点将在下面的描述中部分给出，部分将从下面的描述中变得明显，或通过本公开的实践了解到。
附图说明
本公开上述的和/或附加的方面和优点从下面结合附图对实施例的描述中将变得明显和容易理解，其中：
图1为根据本公开实施例的复合干扰信号识别方法的流程图；
图2为根据本公开实施例1的基于卷积神经网络和长短时记忆网络对卫星导航复合干扰信号识别方法的流程图；
图3为根据本公开实施例2的基于卷积神经网络和长短时记忆网络针对复杂的干扰信号识别方法的流程图；
图4为根据本公开实施例的复合干扰信号识别系统的结构示意图；
图5为根据本公开实施例的电子设备的结构示意图。
具体实施方式
需要说明的是，在不冲突的情况下，本申请中的实施例及实施例中的特征可以相互组合。下面将参考附图并结合实施例来详细说明本公开。
为了使本技术领域的人员更好地理解本公开方案，下面将结合本公开实施例中的附图，对本公开实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本公开一部分的实施例，而不是全部的实施例。基于本公开中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本公开保护的范围。
下面参照附图描述根据本公开实施例提出的卫星导航复合干扰信号识别方法及系统，首先将参照附图描述根据本公开实施例提出的卫星导航复合干扰信号识别方法。
图1是本公开实施例的卫星导航复合干扰信号识别方法的流程图。
如图1所示，该卫星导航复合干扰信号识别方法包括以下步骤：
步骤1：将待识别的复合干扰信号进行预处理，并获取所述预处理后的待识别复合干扰信号的多域特征；
步骤2：将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型，一维序列特征输入到一维序列特征提取模块，多维序列特征输入到多维序列特征提取模块，然后分别对不同维度的域特征进行提取；
步骤3：将提取出的不同维度的域特征输入预先训练好的深度学习神经网络模型的特征融合层和全连接层中，得到所述复合干扰信号的分类识别结果。
需要说明的是，所述预先训练好的深度学习神经网络模型包括：LSTM层、Attention层、 一维序列特征提取模块及多维序列特征提取模块、特征融合层和全连接层；
所述预先训练好的深度学习神经网络模型是基于获取历史时段内复合干扰信号的多域特征对初始深度学习神经网络进行训练得到的。
本公开实施例的卫星导航复合干扰信号识别方法，将复合干扰信号进行预处理，提取信号的时域特征、频域特征、时频域特征和空域特征，分别输入预先训练好的深度学习神经网络，输出对复合干扰信号的分类识别结果。本公开提高了卫星导航复合干扰信号的分类识别精度，且计算过程简单、方便、快捷。
图2是本公开实施例1的基于卷积神经网络和长短时记忆网络对卫星导航复合干扰信号识别方法的流程图。
如图2所示，该基于卷积神经网络和长短时记忆网络对卫星导航复合干扰信号识别方法包括：
(1)将要送入网络输入层的信号时域序列进行预处理，以保证与训练网络所使用的数据约束条件相同。
也就是说，为了消除信号强度对信号识别的影响，首先将复基带信号进行功率归一化处理，又由于干扰信号是复信号，在预处理阶段将信号进行绝对值处理后作为网络模型的输入。
(2)对预处理后的信号分别进行时域、频域、时频域、空域的特征提取，并将参数分别输入神经网络。
具体地，参数分别输入神经网络，包括：
由于信号的时域特征和频域特征由于是1维特征向量，要先通过LSTM层，而时频域特征和空域特征由于是2维特征向量，无需通过LSTM层，可直接通过相应的卷积神经网络。
(3)构建LSTM网络层，负责接收样本数据并从样本数据中学习干扰信号的特征信息。
具体地，LSTM是一种特殊的循环神经网络，解决了传统RNN的梯度消失的问题，通过遗忘门可以减弱历史信息中的噪声干扰，通过输入门可以选择历史信息中重要的部分，遗忘门和输入门决定了当前神经网络层的状态信息，输出门决定了哪些信息将作为输出。而单纯的卷积神经网络并不能有效的分析时序信号的特征和历史信息。
也就是说，在卷积神经网络前嵌入LSTM网络，LSTM会将学习到的重要特征保存为长期记忆，并根据学习的进行选择性的保留、更新或遗忘已保存的长期记忆，而在多次迭代中权重始终很小的特征便会被视为短期记忆而被遗忘，这种机制使得重要的特征随着迭代次数的增加而传递，使得网络在处理长时间依赖序列的分类问题上性能表现良好。因此，加入LSTM网络能将干扰信号识别的准确率得到提升。
(4)构建注意力机制Attention层，注意力机制层的输入就是LSTM的输出，负责对LSTM层中学习到的特征去分配相应的注意力权重。
具体地，Attention机制借助了人脑识别事物的特点，专注于事物的显著性特征，而忽略其他不重要的细节，将有限的精力和资源放在重要的位置以提高工作效率。
也就是说，利用Attention机制为经过LSTM提取的特征向量分配注意力权重，将那些较为显著的特征集合进行高度总结后再输入卷积神经网络进而产生分类结果。Attention机制的引入可以使模型在训练时始终关注那些较为显著的特征，不仅可以降低训练时的参数从而提升训练效率，还能提升模型的分类准确率。
(5)构建多层卷积神经网络模型，包括多个卷积层，每个卷积层后都有归一化层和池化层，激活函数都设定为ReLU函数，时域特征和频域特征在经过LSTM层和Attention后通过卷积神经网络构建2维特征向量，时频域特征和空域特征直接通过相应的卷积神经网络进行下采样。
(6)使用通道注意力机制的方法将信号的4种特征进行融合，并使用卷积层对通道进行压缩。
需要说明的是，将特征进行融合能够实现特征互补的效果，能有效地提高信号识别的准确率，而将通道进行压缩能够更加有效的计算通道注意力，对空间信息的聚合通常使用平均池化或最大池化的方法。
(7)最后的输出层是Softmax全连接层，能将分类结果进行有效的输出。
具体地，将分类结果进行有效输出，包括：
利用训练集对构建的卷积神经网络进行优化训练直到损失函数的误差小于设定值，输出对复合干扰信号的分类识别结果。使用80％的样本数据集去训练构建的复杂的深度学习神经网络，不断的调整网络的参数，使用20％的样本数据集去测试神经网络的识别准确率，继续调整网络参数；训练模型时使用的是交叉熵损失函数，优化器采用AdamOptimizer。
需要说明的是，卷积神经网络经过多次的卷积、池化等操作来前向传播，并通过误差反向传播的方式来训练网络，卷积神经网络对于有规则的结构性数据，有很强的特征提取能力，但是对时间序列变化的特征提取能力偏弱，因此在信号的时域特征和频域特征进入卷积神经网络前需要经过LSTM网络和Attention层，先形成对干扰信号特征的有效提取，最终与下采样后的时频域特征和空域特征进行特征融合，再通过卷积层及Softmax全连接层实现分类，进而形成干扰信号识别的整体模型。
需要说明的是，使用ReLU激活函数来加入非线性因素，具备适度的稀疏性，加速网络的收敛，并且减少了参数的相互依存关系，避免模型的过拟合问题，从而提高模型的泛化能力
需要说明的是，全连接层将得到的特征向量经过Softmax分类器归一化产生分类概率并将其输出。
图3是本公开实施例2的基于卷积神经网络和长短时记忆网络针对复杂的干扰信号识别方法的流程图。
该基于卷积神经网络和长短时记忆网络针对复杂的干扰信号识别方法，包括：
生成训练和测试所需数据集，生成单个干扰信号和不同干扰信号叠加的信号。单个干扰信号有如下6种：单音干扰、多音干扰、扫频干扰、脉冲干扰、宽频带干扰和部分带干扰；两类干扰信号叠加的组合情况有
Figure PCTCN2022084693-appb-000001
种；三类干扰信号叠加的组合情况有
Figure PCTCN2022084693-appb-000002
种；四类干扰信号叠加的情况有
Figure PCTCN2022084693-appb-000003
种，所以构建的深度学习神经网络要识别6+15+20+15＝56种类别的复杂的干扰信号。
设置接收端收到的信号长度是1x1024，即接收端的一位信号采样点数为1024，取干噪比(JNR)为-5dB～14dB的干扰信号作为接收信号，JNR间隔为1dB，取值情况有20种，对每个混合信号的对应的每个JNR的值都生成100个样本，则每种混合信号都有2000个样本，将数据集随机打乱，其中80％作为训练集，所以训练集样本的数据大小为56*2000*0.8＝89600，20％作为测试集，测试集样本的数据大小为56*2000*0.2＝22400。
如图3所示，对数据集进行4个维度的特征提取并输入相应的神经网络中，其中一维的时域特征和频域特征先通过LSTM层，网络结构中LSTM层负责接收样本数据并学习样本的特征信息，神经元个数为128个，采用L2正则化，L2范数设定为0.001，同时设定LSTM网络为双向网络；
Attention层负责对从LSTM层学到的特征集合去分配注意力权重，采用sigmoid函数，评分函数为乘积矩阵；
一维特征向量时域特征和频域特征经过LSTM和注意力机制处理后，通过的卷积神经网络结构如下：第一层的卷积核数量是32个，卷积核大小是1*10；将一维向量resize成二维张量后输入第二层，第二层的卷积核数量是32个，卷积核大小是3*3；第三层的卷积核数量是64个，卷积核大小是5*5；
二维特征向量时频域特征和空域特征经过的卷积神经网络结构如下：第一层的卷积核数量是32个，卷积核大小是3*3；第二层的卷积核数量是64个，卷积核大小是5*5。
将通过卷积神经网络的4种特征向量通过通道注意力机制的方法进行特征融合，然后使用卷积核对通道进行压缩，能够更加有效的计算通道注意力。
最后输入到Softmax全连接层，卷积核数量是56，表示识别的干扰信号类别，实现对复合干扰信号的分类。
具体地，神经网络训练过程，包括：
初始化模型中的超参数，学习率设定为0.0001；将训练集中的89600个样本用大小为256的batch进行切分；随机选取一个batch送入神经网络进行训练；LSTM层对输入网络的batch 中的数据进行特征提取；再由Attention层为提取到的特征设置注意力权重；进而输入卷积神经网络进行卷积和池化；最后通过Softmax层输出识别结果；使用训练集去训练构建复杂的神经网络，并不断的调整网络的参数，使用测试集去测试神经网络的识别准确率，并继续调整网络参数。
为了实现上述实施例，如图4所示，本实施例中还提供了一种卫星导航复合干扰信号识别系统10，该系统10包括：获取模块100，提取模块200，识别模块300。
获取模块100，用于将待识别的复合干扰信号进行预处理，并获取所述预处理后的待识别复合干扰信号的多域特征；
提取模块200，用于将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型，一维序列特征输入到一维序列特征提取模块，多维序列特征输入到多维序列特征提取模块，然后分别对不同维度的域特征进行提取；
识别模块300，用于将提取出的不同维度的域特征输入预先训练好的深度学习神经网络模型的特征融合层和全连接层中，得到所述复合干扰信号的分类识别结果。
所述预先训练好的深度学习神经网络模型包括：LSTM层、Attention层、一维序列特征提取模块与多维序列特征提取模块、特征融合层和全连接层。
所述预先训练好的深度学习神经网络模型是基于获取历史时段内复合干扰信号的多域特征对初始深度学习神经网络进行训练得到的。
本公开实施例的卫星导航复合干扰信号识别系统，将复合干扰信号进行预处理，提取信号的时域特征、频域特征、时频域特征和空域特征，分别输入预先训练好的深度学习神经网络，输出对复合干扰信号的分类识别结果。本公开提高了卫星导航复合干扰信号的分类识别精度，且计算过程简单、方便、快捷。
在本公开实施例中，所述将待识别的复合干扰信号进行预处理，包括：
将待识别的复合干扰信号进行绝对值处理、归一化处理、滤波去噪和信号下变频处理。
在本公开实施例中，所述多域包括：时域、频域、时频域和空域特征。
在本公开实施例中，所述将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型前还包括：
将获取的所述待识别复合干扰信号的时域和频域特征分别输入所述预先训练好的深度学习神经网络模型的LSTM层中，分别学习得到所述待识别复合干扰信号的时域和频域特征信息；
将所述学习得到的所述待识别复合干扰信号的时域和频域特征信息分别输入预先训练好的深度学习神经网络模型的Attention层中，对学习得到所述信号的时域和频域特征信息分配对应的注意力权重，得到分配注意力权重后的时域和频域一维序列特征；
其中，所述预先训练好的深度学习神经网络模型包括：LSTM层、Attention层、一维序列特征提取模块与多维序列特征提取模块、特征融合层和全连接层；
所述预先训练好的深度学习神经网络模型是基于获取历史时段内复合干扰信号的多域特征对初始深度学习神经网络进行训练得到的。
在一些实施例中，所述提取模块200，包括：
第一提取单元，用于将所述分配注意力权重后的时域和频域一维序列特征分别输入预先建立的深度学习神经网络模型的一维序列特征提取模块中，提取出注意力权重大于预设的权重阈值的时域和频域特征；
第二提取单元，用于将获取的所述待识别复合干扰信号的时频域和空域多维序列特征分别输入预先训练好的深度学习神经网络模型的多维序列特征提取模块中，获得对时频域和空域特征进行下采样的时频域和空域特征。
在本公开实施例中，所述识别模块300，包括：
融合单元，用于将得到的注意力权重大于预设的权重阈值的时域和频域特征及所述下采样的时频域和空域特征输入预先训练好的深度学习神经网络模型的特征融合层进行特征融合；
识别单元，用于将所述融合后的特征数据输入预先训练好的深度学习神经网络模型的全连接层，得到所述待识别复合干扰信号的分类识别结果。
在本公开实施例中，所述预先训练好的深度学习神经网络模型的训练过程包括：
获取预处理后的历史时段内复合干扰信号的多域特征，其中所述多域特征包括：时域、频域、时频域和空域特征；
将所述预处理后的历史时段内复合干扰信号的时域、频域分别依次输入初始的深度学习神经网络模型中，将交叉熵作为模型的损失函数，用自适应矩阵估计Adam优化算法对所述模型进行训练，得到训练好的深度学习神经网络模型。
需要说明的是，所述一维序列特征提取模块或多维序列特征提取模块，包括：递归神经网络和卷积神经网络。
需要说明的是，前述对卫星导航复合干扰信号识别方法实施例的解释说明也适用于该实施例的卫星导航复合干扰信号识别系统，此处不再赘述。
图5为本公开实施例提供的一种电子设备1000的结构示意图。
如图5所示，该电子设备1000包括：处理器1001，存储器1002和通信接口1003。通信接口1003用于实现和外部设备的数据交换，处理器1001可以调用存储器1002中存储的计算机程序，以实现：
将待识别的复合干扰信号进行预处理，并获取所述预处理后的待识别复合干扰信号的多 域特征；
将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型，一维序列特征输入到一维序列特征提取模块，多维序列特征输入到多维序列特征提取模块，然后分别对不同维度的域特征进行提取；
将提取出的不同维度的域特征输入预先训练好的深度学习神经网络模型的特征融合层和全连接层中，得到所述复合干扰信号的分类识别结果。
应该理解的是，本申请实施例中所描述的电子设备1000可执行前文图1到图3所对应实施例中对卫星导航复合干扰信号识别方法的描述，也可执行前文图4所对应实施例中该卫星导航复合干扰信号识别系统10的描述，在此不再赘述。另外，对采用相同方法的有益效果描述，也不再进行赘述。
此外，术语“第一”、“第二”仅用于描述目的，而不能理解为指示或暗示相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。在本公开的描述中，“多个”的含义是至少两个，例如两个，三个等，除非另有明确具体的限定。
在本说明书的描述中，参考术语“一个实施例”、“一些实施例”、“示例”、“具体示例”、或“一些示例”等的描述意指结合该实施例或示例描述的具体特征、结构、材料或者特点包含于本公开的至少一个实施例或示例中。在本说明书中，对上述术语的示意性表述不必须针对的是相同的实施例或示例。而且，描述的具体特征、结构、材料或者特点可以在任一个或多个实施例或示例中以合适的方式结合。此外，在不相互矛盾的情况下，本领域的技术人员可以将本说明书中描述的不同实施例或示例以及不同实施例或示例的特征进行结合和组合。
尽管上面已经示出和描述了本公开的实施例，可以理解的是，上述实施例是示例性的，不能理解为对本公开的限制，本领域的普通技术人员在本公开的范围内可以对上述实施例进行变化、修改、替换和变型。

Claims (12)

  1. 一种复合干扰信号识别方法，其特征在于，所述方法包括：
    将待识别的复合干扰信号进行预处理，并获取所述预处理后的待识别复合干扰信号的多域特征；
    将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型，一维序列特征输入到一维序列特征提取模块，多维序列特征输入到多维序列特征提取模块，然后分别对不同维度的域特征进行提取；
    将提取出的不同维度的域特征输入预先训练好的深度学习神经网络模型的特征融合层和全连接层中，得到所述复合干扰信号的分类识别结果。
  2. 如权利要求1所述的方法，其特征在于，所述将待识别的复合干扰信号进行预处理，包括：
    将待识别的复合干扰信号进行绝对值处理、归一化处理、滤波去噪和信号下变频处理。
  3. 如权利要求1所述的方法，其特征在于，所述多域包括：时域、频域、时频域和空域特征。
  4. 如权利要求3所述的方法，其特征在于，所述将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型前还包括：
    将获取的所述待识别复合干扰信号的时域和频域特征分别输入所述预先训练好的深度学习神经网络模型的LSTM层中，分别学习得到所述待识别复合干扰信号的时域和频域特征信息；
    将所述学习得到的所述待识别复合干扰信号的时域和频域特征信息分别输入预先训练好的深度学习神经网络模型的Attention层中，对学习得到所述信号的时域和频域特征信息分配对应的注意力权重，得到分配注意力权重后的时域和频域一维序列特征；
    其中，所述预先训练好的深度学习神经网络模型包括：LSTM层、Attention层、一维序列特征提取模块与多维序列特征提取模块、特征融合层和全连接层；
    所述预先训练好的深度学习神经网络模型是基于获取历史时段内复合干扰信号的多域特征对初始深度学习神经网络进行训练得到的。
  5. 如权利要求4所述的方法，其特征在于，所述将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型，一维序列特征输入到一维序列特征提取模块，多维序列特征输入到多维序列特征提取模块，然后分别对不同维度的域特征进行提取，包括：
    将所述分配注意力权重后的时域和频域一维序列特征分别输入预先建立的深度学习神经网络模型的一维序列特征提取模块中，提取出注意力权重大于预设的权重阈值的时域和频域特征；
    将获取的所述待识别复合干扰信号的时频域和空域多维序列特征分别输入预先训练好的深度学习神经网络模型的多维序列特征提取模块中，获得对时频域和空域特征进 行下采样的时频域和空域特征。
  6. 如权利要求5所述的方法，其特征在于，所述将提取出的不同维度的域特征输入预先训练好的深度学习神经网络模型的特征融合层和全连接层中，得到所述复合干扰信号的分类识别结果，包括：
    将得到的注意力权重大于预设的权重阈值的时域和频域特征及所述下采样的时频域和空域特征输入预先训练好的深度学习神经网络模型的特征融合层进行特征融合；
    将所述融合后的特征数据输入预先训练好的深度学习神经网络模型的全连接层，得到所述待识别复合干扰信号的分类识别结果。
  7. 如权利要求1所述的方法，其特征在于，所述预先训练好的深度学习神经网络模型的训练过程包括：
    获取预处理后的历史时段内复合干扰信号的多域特征，其中所述多域特征包括：时域、频域、时频域和空域特征；
    将所述预处理后的历史时段内复合干扰信号的时域、频域分别依次输入初始的深度学习神经网络模型中，将交叉熵作为模型的损失函数，用自适应矩阵估计Adam优化算法对所述模型进行训练，得到训练好的深度学习神经网络模型。
  8. 如权利要求1所述的方法，其特征在于，所述一维序列特征提取模块或多维序列特征提取模块，包括：递归神经网络和卷积神经网络。
  9. 一种复合干扰信号识别系统，其特征在于，所述系统包括：
    获取模块，用于将待识别的复合干扰信号进行预处理，并获取所述预处理后的待识别复合干扰信号的多域特征；
    提取模块，用于将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型，一维序列特征输入到一维序列特征提取模块，多维序列特征输入到多维序列特征提取模块，然后分别对不同维度的域特征进行提取；
    识别模块，用于将提取出的不同维度的域特征输入预先训练好的深度学习神经网络模型的特征融合层和全连接层中，得到所述复合干扰信号的分类识别结果。
  10. 一种电子设备，其特征在于，包括：存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时，实现以下方法：
    将待识别的复合干扰信号进行预处理，并获取所述预处理后的待识别复合干扰信号的多域特征；
    将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型，一维序列特征输入到一维序列特征提取模块，多维序列特征输入到多维序列特征提取模块，然后分别对不同维度的域特征进行提取；
    将提取出的不同维度的域特征输入预先训练好的深度学习神经网络模型的特征融合层和全连接层中，得到所述复合干扰信号的分类识别结果。
  11. 一种存储有计算机指令的非瞬时计算机可读存储介质，其中，所述计算机指令用于使所述计算机执行以下方法：
    将待识别的复合干扰信号进行预处理，并获取所述预处理后的待识别复合干扰信号的多域特征；
    将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型，一维序列特征输入到一维序列特征提取模块，多维序列特征输入到多维序列特征提取模块，然后分别对不同维度的域特征进行提取；
    将提取出的不同维度的域特征输入预先训练好的深度学习神经网络模型的特征融合层和全连接层中，得到所述复合干扰信号的分类识别结果。
  12. 一种计算机程序产品，包括计算机程序，所述计算机程序在被处理器执行时实现以下方法：
    将待识别的复合干扰信号进行预处理，并获取所述预处理后的待识别复合干扰信号的多域特征；
    将获取的多域特征分别基于其对应的维度分别输入预先训练好的深度学习神经网络模型，一维序列特征输入到一维序列特征提取模块，多维序列特征输入到多维序列特征提取模块，然后分别对不同维度的域特征进行提取；
    将提取出的不同维度的域特征输入预先训练好的深度学习神经网络模型的特征融合层和全连接层中，得到所述复合干扰信号的分类识别结果。
PCT/CN2022/084693 2021-11-26 2022-03-31 复合干扰信号识别方法和系统 WO2023092923A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111423556.8 2021-11-26
CN202111423556.8A CN114201988A (zh) 2021-11-26 2021-11-26 卫星导航复合干扰信号识别方法和系统

Publications (1)

Publication Number Publication Date
WO2023092923A1 true WO2023092923A1 (zh) 2023-06-01

Family

ID=80649157

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/084693 WO2023092923A1 (zh) 2021-11-26 2022-03-31 复合干扰信号识别方法和系统

Country Status (2)

Country Link
CN (1) CN114201988A (zh)
WO (1) WO2023092923A1 (zh)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116502071A (zh) * 2023-06-26 2023-07-28 武汉能钠智能装备技术股份有限公司 一种重点信号检测系统及方法
CN116559794A (zh) * 2023-07-12 2023-08-08 西安电子科技大学 一种双多域复数神经网络的雷达抗干扰智能决策方法
CN117639986A (zh) * 2023-11-29 2024-03-01 东营市无线电监测站 一种应用于市域监测的基于深度学习的调频广播信号频谱监测方法及系统
CN117743946A (zh) * 2024-02-19 2024-03-22 山东大学 基于融合特征和组卷积ViT网络的信号类型识别方法及系统
CN117848713A (zh) * 2024-01-05 2024-04-09 湖州槐坎南方水泥有限公司 布袋除尘器脉冲阀运行状态监测系统及方法
CN117850653A (zh) * 2024-03-04 2024-04-09 山东京运维科技有限公司 触摸显示屏的控制方法及系统
CN117970012A (zh) * 2024-04-01 2024-05-03 南京晟迅信息科技有限公司 电磁抗干扰试验信号监测系统
CN117992931A (zh) * 2024-04-03 2024-05-07 国网冀北电力有限公司 一种直流信号特征库与参比信号模型建立系统及方法
WO2024099006A1 (zh) * 2023-07-12 2024-05-16 广西科学院 一种基于小波包分析的北斗海上多径干扰消除方法及装置
CN117992931B (zh) * 2024-04-03 2024-06-11 国网冀北电力有限公司 一种直流信号特征库与参比信号模型建立系统及方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114201988A (zh) * 2021-11-26 2022-03-18 北京理工大学 卫星导航复合干扰信号识别方法和系统
CN115952407B (zh) * 2023-01-04 2024-01-30 广东工业大学 一种考虑卫星时序性和空域交互性的多径信号识别方法
CN117675112B (zh) * 2024-02-01 2024-05-03 阳光凯讯(北京)科技股份有限公司 基于机器学习的通信信号处理方法、系统、设备及介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140145872A1 (en) * 2012-11-26 2014-05-29 Fujitsu Limited Detecting and ranging apparatus and method for identifying interference signal
CN110826630A (zh) * 2019-11-08 2020-02-21 哈尔滨工业大学 基于深度卷积神经网络的雷达干扰信号特征级融合识别方法
CN113114400A (zh) * 2021-04-14 2021-07-13 中南大学 基于时序注意力机制和lstm模型的信号频谱空洞感知方法
CN114201988A (zh) * 2021-11-26 2022-03-18 北京理工大学 卫星导航复合干扰信号识别方法和系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140145872A1 (en) * 2012-11-26 2014-05-29 Fujitsu Limited Detecting and ranging apparatus and method for identifying interference signal
CN110826630A (zh) * 2019-11-08 2020-02-21 哈尔滨工业大学 基于深度卷积神经网络的雷达干扰信号特征级融合识别方法
CN113114400A (zh) * 2021-04-14 2021-07-13 中南大学 基于时序注意力机制和lstm模型的信号频谱空洞感知方法
CN114201988A (zh) * 2021-11-26 2022-03-18 北京理工大学 卫星导航复合干扰信号识别方法和系统

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
WANG PENGYU;CHENG YUFAN;XU HAO;SHANG GAOYANG: "Jamming Classification Using Convolutional Neural Network-Based Joint Multi-domain Feature Extraction", JOURNAL OF SIGNAL PROCESSING, vol. 38, no. 5, 25 October 2021 (2021-10-25), pages 915 - 925, XP093068765, ISSN: 1003-0530, DOI: 10.16798/j.issn.1003-0530.2022.05.003 *

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116502071A (zh) * 2023-06-26 2023-07-28 武汉能钠智能装备技术股份有限公司 一种重点信号检测系统及方法
CN116502071B (zh) * 2023-06-26 2023-11-17 武汉能钠智能装备技术股份有限公司 一种重点信号检测系统及方法
CN116559794A (zh) * 2023-07-12 2023-08-08 西安电子科技大学 一种双多域复数神经网络的雷达抗干扰智能决策方法
CN116559794B (zh) * 2023-07-12 2023-09-29 西安电子科技大学 一种双多域复数神经网络的雷达抗干扰智能决策方法
WO2024099006A1 (zh) * 2023-07-12 2024-05-16 广西科学院 一种基于小波包分析的北斗海上多径干扰消除方法及装置
CN117639986A (zh) * 2023-11-29 2024-03-01 东营市无线电监测站 一种应用于市域监测的基于深度学习的调频广播信号频谱监测方法及系统
CN117848713A (zh) * 2024-01-05 2024-04-09 湖州槐坎南方水泥有限公司 布袋除尘器脉冲阀运行状态监测系统及方法
CN117743946B (zh) * 2024-02-19 2024-04-30 山东大学 基于融合特征和组卷积ViT网络的信号类型识别方法及系统
CN117743946A (zh) * 2024-02-19 2024-03-22 山东大学 基于融合特征和组卷积ViT网络的信号类型识别方法及系统
CN117850653A (zh) * 2024-03-04 2024-04-09 山东京运维科技有限公司 触摸显示屏的控制方法及系统
CN117970012A (zh) * 2024-04-01 2024-05-03 南京晟迅信息科技有限公司 电磁抗干扰试验信号监测系统
CN117992931A (zh) * 2024-04-03 2024-05-07 国网冀北电力有限公司 一种直流信号特征库与参比信号模型建立系统及方法
CN117992931B (zh) * 2024-04-03 2024-06-11 国网冀北电力有限公司 一种直流信号特征库与参比信号模型建立系统及方法

Also Published As

Publication number Publication date
CN114201988A (zh) 2022-03-18

Similar Documents

Publication Publication Date Title
WO2023092923A1 (zh) 复合干扰信号识别方法和系统
CN107316013B (zh) 基于nsct变换和dcnn的高光谱图像分类方法
CN109994203B (zh) 一种基于eeg信号深度多视角特征学习的癫痫检测方法
CN109389058B (zh) 海杂波与噪声信号分类方法及系统
Liu et al. Retracted: Human emotion recognition based on galvanic skin response signal feature selection and svm
CN108827605B (zh) 一种基于改进稀疏滤波的机械故障特征自动提取方法
CN105740887A (zh) 一种基于pso-svm的脑电信号特征分类方法
Ashokkumar et al. Implementation of deep neural networks for classifying electroencephalogram signal using fractional S‐transform for epileptic seizure detection
CN114359738B (zh) 一种跨场景鲁棒的室内人数无线检测方法及系统
Zhang et al. WiFi-based cross-domain gesture recognition via modified prototypical networks
CN110516728B (zh) 基于去噪卷积神经网络的极化sar地物分类方法
CN109512390B (zh) 基于eeg时域多维度特征及m-wsvm的睡眠分期方法及可穿戴装置
CN112465069B (zh) 一种基于多尺度卷积核cnn的脑电情绪分类方法
CN112861813B (zh) 基于复值卷积神经网络的墙后人体行为识别方法
CN111178331A (zh) 雷达图像识别系统、方法、设备及计算机可读存储介质
CN111954250A (zh) 一种轻量级Wi-Fi行为感知方法和系统
CN111387975A (zh) 基于机器学习的脑电信号识别方法
Ganeshbabu Glaucoma image classification using discrete orthogonal stockwell transform
CN114781463A (zh) 一种跨场景鲁棒的室内摔倒无线检测方法及相关设备
CN104463227B (zh) 基于fqpso和目标分解的极化sar图像分类方法
Vrbancic et al. Automatic detection of heartbeats in heart sound signals using deep convolutional neural networks
CN116482618B (zh) 基于多损失特征自校准网络的雷达有源干扰识别方法
Cui et al. Double-branch local context feature extraction network for hyperspectral image classification
Chen et al. Classification of solar radio spectrum based on VGG16 transfer learning
CN115251909B (zh) 基于时空卷积神经网络的脑电信号评估听力的方法及装置