阅读说明：本技术 基于深度卷积神经网络集成的雷达干扰信号识别方法 (Radar interference signal identification method based on deep convolutional neural network integration ) 是由 张浩宇 陈雨时 于雷 位寅生 叶春茂 李迎春 于 2019-11-08 设计创作，主要内容包括：基于深度卷积神经网络集成的雷达干扰信号识别方法,属于雷达信号识别领域,本发明为解决采用现有深度学习模型识别雷达信号存在过拟合、模型泛化能力差,导致识别系统识别准确率低、鲁棒性弱的问题。本发明方法包括以下步骤：步骤一、将雷达干扰信号时域数据集划分为训练集、验证集以及测试集三部分；步骤二、对训练集X做有放回的随机采样T次,获得T个相互独立的采样训练集；步骤三、采用一维CNN卷积神经网络作为特征提取器、采用支持向量机作为分类器来构造个体学习器,根据步骤二的T个采样训练集来训练T个个体学习器以构造同质集成,构建模型；步骤四、将待测雷达干扰信号输入至步骤三的模型中进行识别。(The invention discloses a radar interference signal identification method based on deep convolutional neural network integration, belongs to the field of radar signal identification, and aims to solve the problems of low identification accuracy and weak robustness of an identification system due to overfitting and poor model generalization capability of radar signals identified by adopting an existing deep learning model. The method comprises the following steps: dividing a radar interference signal time domain data set into a training set, a verification set and a test set; step two, performing replaced random sampling on the training set X for T times to obtain T mutually independent sampling training sets; step three, adopting a one-dimensional CNN convolutional neural network as a feature extractor and adopting a support vector machine as a classifier to construct an individual learner, and training T individual learners according to the T sampling training sets in the step two to construct homogeneous integration and construct a model; and step four, inputting the radar interference signal to be detected into the model in the step three for identification.)

1. The radar interference signal identification method based on deep convolutional neural network integration is characterized by comprising the following steps of:

dividing a radar interference signal time domain data set into a training set, a verification set and a test set;

recording a training set as X and recording the number of training samples as m;

step two, performing replaced random sampling on the training set X for T times to obtain T mutually independent sampling training sets X₁,X₂,...,X_T(ii) a The number of the sampling training samples in each sampling training set is m';

step three, adopting a one-dimensional CNN convolutional neural network as a feature extractor and adopting a support vector machine as a classifier to construct an individual learner, and training T individual learners according to the T sampling training sets in the step two to construct homogeneous integration and construct a model;

and step four, inputting the radar interference signal to be detected into the model in the step three for identification.

2. The method for identifying the radar interference signal based on the deep convolutional neural network integration as claimed in claim 1, wherein the process of dividing the radar interference signal time domain data set in the first step is as follows:

step one, marking original radar data to form a radar interference signal time domain data set: storing each sample in a vector, the first 50% of the vector being labeled as real data of the sample and the last 50% being labeled as imaginary data of the sample;

and step two, randomly dividing a radar interference signal time domain data set into three mutually disjoint sets which are respectively a training set, a verification set and a test set.

3. The method for identifying the radar interference signal based on the deep convolutional neural network integration as claimed in claim 2, wherein the types of the original radar data are 12, and each type of data is divided into a training set, a verification set and a test set according to a ratio of 3:1: 1.

4. The radar interference signal identification method based on deep convolutional neural network integration according to claim 1, wherein the identification process of the step four is as follows:

the radar interference signal to be detected is parallelly input to T trained one-dimensional CNN convolutional neural networks for feature extraction, each feature is correspondingly input to a trained support vector machine for recognition, T individual learners output T results in total, and the result with the largest number of votes is used as the final recognition result of the model according to a relative majority voting method.

5. The method for identifying the radar interference signal based on the deep convolutional neural network integration as claimed in claim 1, wherein the number m' of the training samples sampled in the sampling training set is equal to the number m of the training samples sampled in the training set.

Technical Field

The invention belongs to the field of radar signal identification, and relates to a technology for identifying radar signals by using a convolutional neural network.

Background

With the continuous improvement of the technological level, electronic warfare has become an important combat means in the existing war, the anti-interference capability of the radar also becomes the key for the success or failure of the war in the increasingly complex electromagnetic environment of the battlefield, and the efficient identification and classification of the radar interference signals are the basis and the key of the anti-interference technology of the radar. The key step of the radar interference signal in the identification process is the extraction of characteristic parameters, but with the high-speed development of modern military technology, the form of the radar interference signal is more and more complex, if the artificial characteristics are extracted continuously depending on artificial experience, a large amount of human and material resources are consumed, the characteristic extraction is time-consuming, and the characteristics extracted manually are easily influenced by noise and easily have the phenomenon of characteristic redundancy. Therefore, a new feature extraction method needs to be researched to design a radar interference signal identification system with high identification accuracy and strong robustness. The deep learning is a type of learning, and can automatically extract a feature that is effective in data, and can avoid the trouble of manually extracting a feature.

In deep learning, a cnn (volumetric Neural networks) convolutional Neural network is usually adopted to perform feature extraction on the radar interference signal, so as to improve the accuracy of the radar interference signal identification system. In deep learning, overfitting is a common phenomenon, namely, a model has a good effect on a training set, and has a common or poor effect in a test set, and at the moment, the generalization capability of the model is poor, so that the existing deep learning model is adopted to identify radar signals, so that the identification accuracy of an identification system is low, and the robustness is weak.

Disclosure of Invention

The invention aims to solve the problems of low identification accuracy and weak robustness of an identification system due to overfitting and poor model generalization capability of an existing deep learning model for identifying radar signals, and provides a radar interference signal identification method based on deep convolutional neural network integration.

The invention discloses a radar interference signal identification method based on deep convolutional neural network integration, which comprises the following steps of:

dividing a radar interference signal time domain data set into a training set, a verification set and a test set;

recording a training set as X and recording the number of training samples as m;

step two, performing replaced random sampling on the training set X for T times to obtain T mutually independent sampling training sets X₁,X₂,...,X_T(ii) a The number of the sampling training samples in each sampling training set is m';

step three, adopting a one-dimensional CNN convolutional neural network as a feature extractor and adopting a support vector machine as a classifier to construct an individual learner, and training T individual learners according to the T sampling training sets in the step two to construct homogeneous integration and construct a model;

and step four, inputting the radar interference signal to be detected into the model in the step three for identification.

Preferably, the process of dividing the radar interference signal time domain data set in the first step is as follows:

step one, marking original radar data to form a radar interference signal time domain data set: storing each sample in a vector, the first 50% of the vector being labeled as real data of the sample and the last 50% being labeled as imaginary data of the sample;

and step two, randomly dividing a radar interference signal time domain data set into three mutually disjoint sets which are respectively a training set, a verification set and a test set.

Preferably, the types of the original radar data are 12, and each type of data is divided into a training set, a verification set and a test set according to the ratio of 3:1: 1.

Preferably, the identification process of step four is:

the radar interference signal to be detected is parallelly input to T trained one-dimensional CNN convolutional neural networks for feature extraction, each feature is correspondingly input to a trained support vector machine for recognition, T individual learners output T results in total, and the result with the largest number of votes is used as the final recognition result of the model according to a relative majority voting method.

Preferably, the number m' of training samples sampled in the training set is equal to the number m of training samples in the training set.

The invention has the beneficial effects that: when the identification system model is constructed, T sampling training sets obtained by a Bootstrap (self-help method) sampling method on the basis of a conventional training set have certain difference, the sampling training sets are mutually overlapped, and the identification precision and robustness of the model are further improved by utilizing the difference and combining homogeneous ensemble learning.

Drawings

Fig. 1 is a flowchart of a radar interference signal identification method based on deep convolutional neural network integration according to the present invention.

Detailed Description

The following detailed description of the embodiments of the present invention will be provided with reference to the drawings and examples, so that how to apply the technical means to solve the technical problems and achieve the technical effects can be fully understood and implemented. It should be noted that, as long as there is no conflict, the embodiments and the features of the embodiments of the present invention may be combined with each other, and the technical solutions formed are within the scope of the present invention.