CN106485209A - Non-stationary electromagnetic interference signal mode identification method based on EEMD feature extraction and PNN - Google Patents

Abstract

The invention discloses a non-stationary electromagnetic interference signal pattern recognition method based on EEMD feature extraction and PNN, which belongs to the field of electromagnetic signal identification and analysis research; the specific steps are: first, use a spectrum analyzer to measure electromagnetic interference signals of different types of electromagnetic equipment respectively, and Divided into training samples and test samples; then, perform hierarchical EEMD decomposition on all training samples to obtain the intrinsic mode function IMF of each layer of each training sample; then, calculate the energy distribution and kurtosis of all layers of IMF for each training sample , as the feature vector of the training sample; and the feature vector and the category label of the electromagnetic device are trained to obtain a PNN classifier; finally, the PNN classifier is used to identify the type of spectrum data of different test samples; the advantage is: using EEMD The method has better adaptability to signal extraction features; kurtosis and IMF energy distribution are used as feature vectors to effectively distinguish different types of electromagnetic interference signals.

Description

Pattern Recognition of Non-stationary Electromagnetic Interference Signal Based on EEMD Feature Extraction and PNN method

technical field

The invention belongs to the field of electromagnetic signal identification and analysis research, and relates to a non-stationary electromagnetic interference signal pattern recognition method based on EEMD (global empirical mode decomposition) feature extraction and PNN (probability neural network).

Background technique

With the development of electronic information technology, the electromagnetic interference caused by an increasing number of electronic equipment seriously affects the normal operation of electronic equipment. In order to solve the problem of electromagnetic interference, electromagnetic compatibility design came into being. In the actual environment, the electromagnetic signals emitted by the electromagnetic interference sources in electronic equipment often have high nonlinearity, non-Gaussianity and instability, which makes it very difficult to identify them and brings great difficulties to the electromagnetic compatibility design.

Pattern Recognition (Pattern Recognition) is the process of automatic processing and interpretation of patterns studied by computers using mathematical methods. Here, the actual electromagnetic interference signal and the corresponding environment are collectively referred to as "mode". Pattern recognition is a basic intelligence of human beings. With the advent of computers in the 1940s and the rise of artificial intelligence in the 1950s, it was hoped that computers could replace or extend human pattern recognition. Pattern recognition developed rapidly in the early 1960s and became a new discipline. Its basic process is to extract features from patterns in the objective environment and transform them into mathematical models, and then use computers to complete classification and recognition.

Artificial Neural Network (ANN) is a research hotspot in the field of artificial intelligence since the 1980s. It imitates the connection mode of human brain neuron network, and establishes a simple computer neuron network model. Artificial neural network is a computing model, which is composed of a large number of nodes (or called neurons) connected to each other. Each node represents a specific output function called an activation function. The connection between each two nodes represents a weighted value for the signal passing through the connection, called weight, which is equivalent to the memory of the neural network. The output of the network depends on the way the network is connected, the weight values and the activation function. The network itself is usually an approximate expression of some algorithm, function or logic in nature.

Probabilistic Neural Networks (PNN) was proposed by D.F.Specht in 1990. It is a special kind of feedforward network model based on statistical principles, which is usually used to implement clustering functions. Its main strategy is to use Bayesian decision rules to separate the decision space within the multidimensional input space such that the expected risk of misclassification is minimized.

The main advantages of PNN are the following three points: (1) fast training speed; (2) with enough training samples, the optimal solution under Bayesian criterion can always be guaranteed; (3) only considering the probability characteristics of the sample space, Allows to increase training samples without retraining for a long time.

PNN combines the advantages of radial basis neural network and classic probability density estimation principle. Compared with other feedforward neural networks, it has significantly superior performance in clustering tasks.

Contents of the invention

In view of the fact that the actual electromagnetic interference signal has a large spectrum range and many data points, and it is difficult to directly analyze and calculate the data, the invention proposes a non-stationary electromagnetic interference signal pattern recognition method based on EEMD feature extraction and PNN.

Specific steps are as follows:

Step 1. Use the spectrum analyzer to measure the actual electromagnetic equipment respectively, obtain the electromagnetic interference signals of different types of equipment, and divide the spectrum data into training samples and test samples;

Step 2. For all training samples, perform hierarchical EEMD decomposition respectively to obtain the intrinsic mode function IMF of each layer of each training sample;

Specific steps are as follows:

Step 201, for a certain training sample X in the spectrum data, initialize the number of EEMD, and construct Gaussian white noise with different amplitudes;

The length of the training sample X is n, X=[x ₁ ,x ₂ ,...,x _j ,...x _n ]; the number of EEMD is M; Gaussian white noise with different amplitudes is M groups, the mth group Amplitude Gaussian white noise N _m =[n ₁ ^(m) ,n ₂ ^(m) ,...n _j ^(m) ,...n _n ^(m) ]; m=1,2,... M;

Step 202, sequentially adding Gaussian white noise with a given amplitude to the training sample X to obtain M interference signals;

The mth signal interfered by Gaussian white noise is: X _m =XN _m ;

M interference signal sets are: [X ₁ ,X ₂ ,...X _m ,...X _M ];

Step 203, using the EMD method, decompose each interference signal into I-layer intrinsic mode functions IMF to obtain M×I intrinsic mode functions.

The set of M×I intrinsic mode functions is expressed as follows:

c _i,m represent the intrinsic mode function of the i-th layer decomposed by the m-th interference signal X _m ; i=1,2,3,...I; m=1,2,3,...M;

Step 204, for the intrinsic mode function IMF of each layer, respectively calculate the average value of the M intrinsic mode functions of the layer as the final IMF value of the layer, and obtain the intrinsic mode function IMF of each layer of the training sample X;

The intrinsic mode function IMF _i of the i-th layer of the training sample X is the average value of all IMFs in this layer, and the formula is as follows:

Step 205. Repeat the above steps for all training samples to obtain the intrinsic mode function IMF of each layer of each training sample.

Step 3, for each training sample, calculate the energy distribution and kurtosis of all layers of IMF respectively, as the feature vector of the training sample;

First, calculate the energy distribution and entropy of all layers of IMF;

The energy distribution of all layer IMFs is characterized by Entropy of energy distribution in layer I

P _i is the energy of the i-th layer IMF: is an element in the element set of the i-th layer IMF of the training sample X, and the element set of the i-th layer IMF is expressed as: P is the total energy of the training sample X:

Then, calculate the kurtosis of each layer IMF in all layers IMFs of the training sample.

The i-th layer IMF kurtosis K(IMF _i ) is defined as:

is the average value of all points in the i-th layer IMF element set IMF _i : n is the length of the i-th layer IMF signal, which is the same as the length of the training sample;

σ is the corresponding standard deviation:

The final feature vector of the training sample X is:

Step 4, input the feature vectors of all training samples and the category labels of the electromagnetic equipment into the untrained PNN for training to obtain a PNN classifier;

The PNN classifier is used to identify unknown EMI signals.

Step 5, for each test sample, input the PNN classifier to identify the type of spectrum data;

The advantages of the present invention are:

1), based on EEMD feature extraction and PNN non-stationary electromagnetic interference signal pattern recognition method, in the process of signal feature extraction, using EEMD method to decompose the signal has better self-adaptation than wavelet decomposition, Fourier analysis, etc. sex.

2), based on EEMD feature extraction and PNN non-stationary electromagnetic interference signal pattern recognition method, using kurtosis and IMF energy distribution as feature vectors can reflect data uncertainty and distribution instability, etc. These statistical characteristics can effectively distinguish Different types of EMI signals.

3), based on the EEMD feature extraction and PNN non-stationary electromagnetic interference signal pattern recognition method, the establishment of PNN for electromagnetic interference signal pattern recognition has achieved a high degree of automation and integration; and in practical applications, with the continuous expansion of training samples, The recognition accuracy of PNN will continue to increase.

Description of drawings

Fig. 1 is the flow chart of the non-stationary electromagnetic interference signal pattern recognition method based on EEMD feature extraction and PNN of the present invention;

Fig. 2 is a flow chart of the method for obtaining the intrinsic mode function IMF by performing hierarchical EEMD decomposition on training samples according to the present invention.

detailed description

The present invention will be further described in detail below in conjunction with the accompanying drawings.

The present invention provides a pattern recognition system for non-stationary electromagnetic interference signals in the actual environment, specifically a pattern recognition system for non-stationary electromagnetic interference signals based on EEMD feature extraction and PNN. First, it is necessary to use EEMD technology to extract the mathematical features of the signal To represent the original signal, then the probabilistic neural network will be introduced to complete the pattern recognition work on the feature vector.

Specifically, it is divided into two main processes: training process and recognition process. In the training process, the electromagnetic interference signal training samples are converted into corresponding feature vectors through EEMD, and input into the PNN network for training, so that it has the ability to identify non-stationary electromagnetic interference signals; in the identification process, unknown electromagnetic interference signals Through the same feature extraction technology, it is converted into a feature vector, and then input into the mature PNN network obtained in the training process for recognition.

As shown in Figure 1, the specific steps are as follows:

Step 1. Use the spectrum analyzer to measure the actual electromagnetic equipment respectively, obtain the electromagnetic interference signals of different types of equipment through testing, and divide the corresponding spectrum data into training samples and test samples;

Step 2. For all training samples, perform hierarchical EEMD decomposition respectively to obtain the intrinsic mode function IMF of each layer of each training sample;

Use EEMD to decompose all electromagnetic interference signal training samples to obtain the corresponding IMF, and the number of intrinsic model functions of all training samples is the same;

As shown in Figure 2, the specific steps are as follows:

Step 201, for a certain training sample X in the spectrum data, initialize the number of EEMD, and construct Gaussian white noise with different amplitudes;

The training sample X=[x ₁ ,x ₂ ,...,x _j ,...x _n ], the length is n; the number of EEMD is M, and Gaussian white noise with different amplitudes is M groups, expressed as N= [N ₁ ,N ₂ ,...N _m ,...N _M ];

N _m ＝[n ₁ ^(m) ,n ₂ ^(m) ,...n _j ^(m) ,...n _n ^(m) ] is Gaussian white noise with the amplitude of the mth group; m=1,2 ,...M

Step 202, sequentially adding Gaussian white noise with a given amplitude to the training sample X to obtain M interference signals;

The mth signal interfered by Gaussian white noise is:

All sets of M interference signals are: [X ₁ ,X ₂ ,...X _m ,...X _M ]

Step 203, using the EMD method to decompose each interference signal into I-layer intrinsic mode functions IMFs to obtain M×I IMFs.

The set of M×I intrinsic mode functions is expressed as follows:

Each column in the set represents the I-layer IMF decomposed by each interfering signal, and each row represents the IMF decomposed by all interfering signals of the same layer; c _i,m represents the i-th layer intrinsic mode function decomposed by the m-th interfering signal X _m ;i=1,2,3,...I; m=1,2,3,...M;

Step 204, for the intrinsic mode function IMF of each layer, respectively calculate the average value of the M IMFs of the layer as the final IMF value of the layer, and obtain the intrinsic mode function IMF of each layer of the training sample X;

The intrinsic mode function IMF _i of the i-th layer of the training sample X is the average value of all IMFs in this layer, and the formula is as follows:

Step 205. Repeat the above steps for all training samples to obtain the intrinsic mode function IMF of each layer of each training sample.

Step 3, for each training sample, calculate the energy distribution and kurtosis of all layers of IMF respectively, as the feature vector of the training sample;

First, calculate the energy distribution and entropy of all layers of IMF;

The energy distribution of all layer IMFs is characterized by Consider the energy distribution as a probability distribution, and calculate the entropy of the energy distribution of the I layer

The total energy of the training sample X is P _i is the energy of the i-th layer IMF: is an element in the element set of the i-th layer IMF of the training sample X, and the element set of the i-th layer IMF is expressed as: P ₁ +P ₂ +...P _i +...+P _I =P.

Then, calculate the kurtosis of each layer IMF in all I-layer IMFs of the training sample.

The i-th layer IMF kurtosis K(IMF _i ) is defined as:

is the average value of all points in the i-th layer IMF element set IMF _i :

n is the length of the i-th layer IMF signal, which is the same as the length of the training sample;

σ is the corresponding standard deviation:

Finally, the final feature vector of the training sample X is:

Step 4, input the feature vectors of all training samples and the category labels of the electromagnetic equipment into the untrained PNN for training to obtain a PNN classifier;

PNN classifiers can be used to identify unknown EMI signals.

Step 5, for each test sample, input the PNN classifier to identify the type of spectrum data;

Specific steps are as follows:

Step 501, performing hierarchical EEMD decomposition on a certain test sample to obtain the intrinsic mode function IMF of each layer;

Obtain test samples: Obtain electromagnetic interference signal test samples by testing electromagnetic equipment; and repeat step 2 for hierarchical EEMD decomposition to obtain the intrinsic mode function IMF;

Step 502, respectively calculate the energy distribution and kurtosis of each layer of IMF, as the feature vector of the test sample;

EEMD feature extraction: Repeat step 3 for the test sample to obtain the feature vector of the test sample.

Step 503: Input the feature vector of the test sample into the PNN classifier for matching to obtain the type of the electromagnetic interference signal.

PNN identification: Input the feature vector of the obtained test sample into the PNN classifier for identification, and output the result of its category.

Example:

Step 1: Obtain electromagnetic interference signal spectrum data through testing;

The spectrum analyzer is used to measure 4 kinds of equipment in the actual electromagnetic environment, and 8 electromagnetic interference signal spectrum data are obtained for each type of equipment, of which 7 of each type of data are used as training samples, and 1 is used as a test sample. The device names of the spectrum data of the measured electromagnetic interference signal are shown in Table 1.

Table 1

Step 2: Perform 6-layer EEMD decomposition on all 28 training samples to obtain 28 groups of intrinsic mode functions, each group containing 6 intrinsic mode functions.

Step 3: Calculate the energy distribution and kurtosis of each group of intrinsic mode functions separately.

Select a signal for each type of training sample, and calculate the energy distribution of each group of intrinsic mode functions IMF, the results are shown in Table 2;

Table 2

Calculate the kurtosis of each group of intrinsic mode functions IMF, and the results are shown in Table 3;

table 3

Source of training samples K ₁ K ₂ K ₃ K ₄ K ₅ K ₆ notebook 1.9523 2.3561 2.7802 2.8945 2.9801 1.3512 hair dryer 1.8646 2.6346 2.9132 3.1645 3.6543 2.0465 unstarted car 1.9123 2.7194 3.2452 3.6543 3.9812 2.1632 starting car 1.7529 2.1452 2.3567 2.6236 2.7236 1.2251

Step 4: Input all the feature vectors and corresponding category indicators (labels) obtained above into PNN for training. The labels of various EMI signal training samples are shown in Table 4.

Table 4

Source of training samples Label notebook 1 hair dryer 2 unstarted car 3 starting car 4

Step 5: Calculate the eigenvectors of the 4 groups of test samples of the electromagnetic interference signal, and input them into the PNN for identification.

The output of the classification and recognition results of the test samples is shown in Table 5.

table 5

Source of test samples classification result notebook 1 hair dryer 2 unstarted car 3 starting car 4

Through the analysis of the results in Table 5, it can be seen that the trained PNN classifier can correctly classify the test samples into their categories, so that the pattern recognition task can be effectively completed.

Therefore, the pattern recognition system that the present invention proposes, by reading the training sample data, carries out EEMD feature extraction to the training sample, then inputs the characteristic vector obtained into PNN to train, obtains the PNN classification that can be used to identify the new electromagnetic interference signal For a new unknown type of electromagnetic interference signal, the same feature extraction is performed on it and input into the PNN classifier for identification, and its type can be known. Combining the knowledge of feature extraction, probability statistics and artificial intelligence, it is of great significance to be able to effectively classify and identify the spectrum data of electromagnetic interference signals.

Claims (3)

1. a non-stationary electromagnetic interference signal pattern recognition method based on EEMD feature extraction and PNN, is characterized in that, concrete steps are as follows: Step 1. Use the spectrum analyzer to measure the actual electromagnetic equipment respectively, obtain the electromagnetic interference signals of different types of equipment, and divide the spectrum data into training samples and test samples; Step 2. For all training samples, perform hierarchical EEMD decomposition respectively to obtain the intrinsic mode function IMF of each layer of each training sample; The specific steps are: Step 201, for a certain training sample X in the spectrum data, initialize the number of EEMD, and construct Gaussian white noise with different amplitudes; The length of the training sample X is n, X=[x ₁ ,x ₂ ,...,x _j ,...x _n ]; the number of EEMD is M; Gaussian white noise with different amplitudes is M groups, the mth group Amplitude Gaussian white noise N _m =[n ₁ ^(m) ,n ₂ ^(m) ,...n _j ^(m) ,...n _n ^(m) ]; m=1,2,... M; Step 202, sequentially adding Gaussian white noise with a given amplitude to the training sample X to obtain M interference signals; The mth signal interfered by Gaussian white noise is: X _m =XN _m ; M interference signal sets are: [X ₁ ,X ₂ ,...X _m ,...X _M ]; Step 203, using the EMD method to decompose each interference signal into I-layer intrinsic mode functions IMF, and obtain M×I intrinsic mode functions in total; The set of M×I intrinsic mode functions is expressed as follows: $\left[\begin{array}{cccccc}{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 1</span>}}& {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}}& <span\; class="notranslate">\; ...</span>& {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}}& <span\; class="notranslate">\; ...</span>& {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}}\\ {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 1</span>}}& {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}}& <span\; class="notranslate">\; ...</span>& {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}}& <span\; class="notranslate">\; ...</span>& {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}}\\ & & & <span\; class="notranslate">\; ...</span>& & \\ {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 1</span>}}& {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}}& <span\; class="notranslate">\; ...</span>& {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}}& <span\; class="notranslate">\; ...</span>& {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}}\\ & & & <span\; class="notranslate">\; ...</span>& & \\ {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 1</span>}}& {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; 2</span>}}& <span\; class="notranslate">\; ...</span>& {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}}& <span\; class="notranslate">\; ...</span>& {\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; I</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}}\end{array}\right]$ c _i,m represent the intrinsic mode function of the i-th layer decomposed by the m-th interference signal X _m ; i=1,2,3,...I; m=1,2,3,...M; Step 204, for the intrinsic mode function IMF of each layer, respectively calculate the average value of the M intrinsic mode functions of the layer as the final IMF value of the layer, and obtain the intrinsic mode function IMF of each layer of the training sample X; The intrinsic mode function IMF _i of the i-th layer of the training sample X is the average value of all IMFs in this layer, and the formula is as follows: ${\mathrm{<span\; class="notranslate">\; IMF</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}}$ Step 205. Repeat the above steps for all training samples to obtain the intrinsic mode function IMF of each layer of each training sample. Step 3, for each training sample, calculate the energy distribution and kurtosis of all layers of IMF respectively, as the feature vector of the training sample; Step 4, input the feature vectors of all training samples and the category labels of the electromagnetic equipment into the untrained PNN for training to obtain a PNN classifier; Step 5. For each test sample, input the PNN classifier to identify the type of the spectrum data. 2. a kind of non-stationary electromagnetic interference signal pattern recognition method based on EEMD feature extraction and PNN as claimed in claim 1, is characterized in that, described step 3 is specifically: First, calculate the energy distribution and entropy of all layers of IMF; The energy distribution of all layer IMFs is characterized by Entropy of energy distribution in layer I P _i is the energy of the i-th layer IMF: is an element in the element set of the i-th layer IMF of the training sample X, and the element set of the i-th layer IMF is expressed as: P is the total energy of the training sample X: Then, calculate the kurtosis of each layer of IMF in all layers of IMF of the training sample; The i-th layer IMF kurtosis K(IMF _i ) is defined as: $\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; IMF</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; no</span>}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}^{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>}}{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 4</span>}}$ is the average value of all points in the i-th layer IMF element set IMF _i : n is the length of the i-th layer IMF signal, which is the same as the length of the training sample; σ is the corresponding standard deviation: The final feature vector of the training sample X is: 3. a kind of non-stationary electromagnetic interference signal pattern recognition method based on EEMD feature extraction and PNN as claimed in claim 1, is characterized in that, described step 5 is specifically: Step 501, performing hierarchical EEMD decomposition for a certain test sample to obtain the intrinsic mode function IMF of each layer; Obtain test samples: Obtain electromagnetic interference signal test samples by testing electromagnetic equipment; and repeat the hierarchical EEMD decomposition to obtain the intrinsic mode function IMF; Step 502, respectively calculate the energy distribution and kurtosis of each layer of IMF, as the feature vector of the test sample; Step 503, input the feature vector of the test sample into the PNN classifier for matching, and obtain the type of the electromagnetic interference signal; PNN identification: Input the feature vector of the obtained test sample into the PNN classifier for identification, and output the result of its category.