A method for detecting electromagnetic spectrum signals based on correlation between frequency points

Abstract

The present invention belongs to the technical field of signal detection methods, and specifically relates to an electromagnetic spectrum signal detection method based on the correlation between frequency points, comprising the following steps: calculating the correlation coefficient of adjacent frequency points; calculating the normalized maximum value, information entropy, duty cycle, and variance characteristic parameters; forming a characteristic vector from the correlation coefficient of adjacent frequency points and the normalized maximum value, information entropy, duty cycle, and variance characteristic parameters; and detecting the electromagnetic spectrum signal. The present invention has a high accuracy rate in detecting signals and a low false alarm and missed alarm rate. When the signal strength is weak, it can clearly and accurately judge and discover the signal and its related frequency points in a complex environment, and can perform signal identification quickly, accurately, and stably. The method is very suitable for the identification and detection of electromagnetic signals close to background noise. The present invention can realize convenient, fast, and real-time signal detection, and can realize fast and real-time scanning spectrum signal detection by setting the target frequency band.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal detection methods, and in particular relates to an electromagnetic spectrum signal detection method based on correlation between frequency points. Background Art

[0002] Signal detection is the process of extracting received signals containing interference noise. In the field of electromagnetic signal detection, detection is typically based on feature engineering, utilizing signal characteristics or extracting hidden layer features of the signal through feature engineering, and using the obtained features to determine whether there is a signal in the electromagnetic space. Currently, algorithms such as feature parameters and deep learning are frequently used in electromagnetic signal detection. However, in today's dense, complex and changing signal environments, existing signal detection methods also have the following shortcomings and limitations:

[0003] 1. The electromagnetic signal is identified by using signal characteristic parameters with certain physical meanings, such as the peak value, minimum value, mean value, and variance of the signal waveform. The detection effect is poor, the accuracy needs to be improved, and the stability is poor.

[0004] 2. It is effective in identifying specific abnormal signals or can identify abnormal states that change significantly in the electromagnetic space, but it cannot achieve good results when the abnormal signals are weaker.

[0005] 3. It is sensitive to the adjustment of characteristic signal parameters and the selection of functions. It is difficult to identify all characteristic parameters with classification capabilities, resulting in reduced signal detection accuracy.

[0006] 4. In pursuit of high precision, the existing detection algorithms have disadvantages such as complex operation process, excessive computational complexity, and high algorithm complexity.

[0007] 5. Most detection algorithms rely on adjusting thresholds to complete signal recognition, which makes it difficult to achieve adaptive recognition based on the signal environment and lacks flexibility and universality.

[0008] To solve this problem, a method of electromagnetic spectrum signal detection based on the correlation between frequency points is proposed. Summary of the Invention

[0009] In view of the technical problem that the above-mentioned related algorithms such as characteristic parameters and deep learning have been frequently used in the detection of electromagnetic signals, and the existing signal detection methods also have shortcomings and limitations in the current dense, complex and changeable signal environment, the present invention provides an electromagnetic spectrum signal detection method based on the correlation between frequency points. Based on frequency domain data analysis and utilizing the correlation coefficients of adjacent frequency points, an electromagnetic space multi-feature abnormal signal detection method is proposed, which can detect electromagnetic signals by setting the target frequency band.

[0010] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0011] A method for detecting electromagnetic spectrum signals based on correlation between frequency points comprises the following steps:

[0012] S1. Calculate the correlation coefficient of adjacent frequency points;

[0013] S2. Calculate normalized maximum value, information entropy, duty cycle, and variance characteristic parameters;

[0014] S3, forming a feature vector by combining the correlation coefficient of adjacent frequency points with the normalized maximum value, information entropy, duty cycle, and variance feature parameters;

[0015] S4. Detect electromagnetic spectrum signals.

[0016] The method for calculating the correlation coefficient of adjacent frequency points in S1 is:

[0017] Assume that there are n time slots of equal length continuous spectrum data sequence, and each data sequence has 2*m+1 frequency points, that is, there is the following data sequence matrix:

[0018]

[0019] Let S = {X -m ,......,X -3 ,X -2 ,X -1 ,X0,X1,X2,......,X m}, then the frequency sequence adjacent to X0 is

[0020] X -1 ,X -2 ,X -3 ,......,X -m and X1, X2, ..., X m , then the correlation coefficient R between X0 and its adjacent frequency point on the left is calculated as follows:

[0021]

[0022] The formula for calculating the correlation coefficient R between X0 and its adjacent frequency point on the right is as follows:

[0023]

[0024] where x 0，j ∈X0, k=1,2,3,......,m; j=1,2,3,......,n

[0025] Associate the frequency point X0 with the adjacent frequency points, X -1 It is adjacent to the left of X0 once, X-2 It is adjacent to the left twice, and so on. -m is m times adjacent; X1 is 1 time adjacent to the right of X0, X2 is 2 times adjacent to the right, and so on. m is m times adjacent;

[0026] According to the number of adjacent frequency points of X0, there are m adjacent frequency points on both sides of X0, and the correlation coefficient formula (S-2) and (S-3) are used to calculate the correlation coefficient L of the adjacent frequency points on the left and right sides of the frequency point X0. k ,R k ;

[0027] The correlation coefficient of the adjacent frequency point on the left side of X0 is L k =R(X0,X -k ) (S-4)

[0028] The correlation coefficient of the adjacent frequency point to the right of X0 is R k =R(X0,X k ) (S-5)

[0029] Where k = 1, 2, 3, ..., m.

[0030] The method for calculating the normalized maximum value of S2 is:

[0031] Normalized maximum value A0:

[0032]

[0033] Among them, M0 is the maximum value of frequency point X0, M0=max{x 0,,j |x 0,,j ∈X0}, j=1,2,......,n,; B0 is the maximum value of the background noise at frequency point X0; F is the set threshold.

[0034] The method for calculating information entropy in S2 is:

[0035] Information entropy P0:

[0036]

[0037] Where Q is the quantization level of the spectrum data X0; p h is the probability of the spectrum data X0 appearing at the hth level.

[0038] The method of the duty cycle in S2 is:

[0039] Duty cycle C0:

[0040]

[0041] Among them, T0 is the given number of sampling times; V0 is the number of sampling values greater than the maximum background noise B0 among T0 samplings.

[0042] The method of variance in S2 is as follows:

[0043] Variance D0:

[0044]

[0045] Among them, x 0，j ∈X0; j = 1, 2, 3,......, n, is the mean value of the spectral data of frequency point X0, X max is the maximum value of X0, X min is the minimum value of X0, and X max >X min .

[0046] The method of forming a feature vector by combining the correlation coefficients of adjacent frequency points with the normalized maximum value, information entropy, duty cycle, and variance characteristic parameters in S3 is as follows:

[0047] Set the adjacent frequency points of X0 to be taken K times, calculate the correlation coefficients of the 1st to Kth adjacent frequency points of frequency point X0, and the correlation coefficients are L1, L2,......, L K , R1, R2,......, R K , These correlation coefficients and multiple characteristic parameters such as information entropy P0, duty cycle C0, variance D0, and normalized maximum value A0 form a set of feature vectors, that is, it is considered that there are 2*K + 4 eigenvalues for the signal of each group of data, and the feature vector U is represented as follows:

[0048] U = (P0, C0, D0, A0, L1, L2,......, L K , R1, R2,....., R K ) (S-10).

[0049] The method of detecting electromagnetic spectrum signals in S4 is as follows:

[0050] If A0 = 1, it is considered that there is a signal; if A0 = 0, it is considered that there is no signal; if 0 < A0 < 1, it participates in the calculation of the feature vector as an eigenvalue.

[0051] The beneficial effects of the present invention compared with the prior art are:

[0052] The signals detected by the present invention have a high accuracy rate and a low false alarm and missed alarm rate. When the signal strength is weak, the signal and its related frequency points can be clearly and accurately judged and discovered in a complex environment, and the signal can be identified quickly, accurately and stably. The method is very suitable for the identification and detection of electromagnetic signals close to background noise. The present invention can realize convenient, fast and real-time signal detection, and can realize fast and real-time scanning spectrum signal detection by setting the target frequency band. In addition, the model training of the present invention is simple and easy to use, and it can be realized by only calculating the eigenvalue. The computational complexity of the present invention is low and easy to understand. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and those skilled in the art can, without inventive effort, derive other implementation drawings based on the provided drawings.

[0054] The structures, proportions, sizes, etc. illustrated in this specification are intended solely to complement the contents disclosed herein and to facilitate understanding and reading by persons skilled in the art. They are not intended to limit the conditions under which the present invention may be implemented and therefore have no substantive technical significance. Any structural modifications, changes in proportions, or adjustments in sizes, without affecting the efficacy and objectives of the present invention, shall remain within the scope of the technical contents disclosed herein.

[0055] Figure 1 It is a flow chart of the steps of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only part of the embodiments of this application, not all the embodiments. These descriptions are only to further illustrate the features and advantages of the present invention, rather than to limit the claims of the present invention. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0057] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following embodiments are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0058] In this embodiment, if Figure 1 As shown, the following steps are included:

[0059] 1. Problem Description

[0060] In a natural environment, collect a data set of equal length sequences within a certain frequency band over a certain period of time, ensuring that a signal exists in each data set within 2 seconds. There are 2*m+1 frequency points in the data set. Collect n data sets and use a detection algorithm to determine whether a signal exists at each frequency point in each data set.

[0061] Assume that the spectrum data from the 1st time slot to the nth time slot can be expressed as a random variable sequence, that is, there is the following data sequence matrix:

[0062] Among them, X -m ......X m Recorded as frequency point data sequence; x -m,1 , x -m,2 ......, x -m,n For X -m Spectrum data sequence; and so on. m,1 , x m,2 ,......,x m,n For X m The problem is that we need to determine whether there is a signal at the frequency point in each set of spectrum data sequences.

[0063] 2. Algorithm Principle

[0064] The implementation algorithm principle is as follows.

[0065] First, the linear correlation between frequency points is used to calculate the correlation coefficient of adjacent frequency points. Secondly, characteristic parameters such as normalized maximum value (A), information entropy (P), duty cycle (C), and variance (D) are calculated, and the correlation coefficient and these characteristic parameters are used together as eigenvalues ​​to form a eigenvector. Finally, the eigenvector and related algorithms are used to implement the detection method, optimize the parameters, form a training model, and make judgments based on the model to detect k signal frequency points for each set of data.

[0066] The specific contents are as follows:

[0067] 1. In data analysis and processing, the correlation coefficient R of adjacent frequency points is first calculated based on the characteristic of linear correlation of frequency points. Adjacent frequency points refer to any frequency point in the spectrum data sequence that is adjacent to the frequency points on its left or right. In this case, the frequency point and the series of frequency points on its left or right are also considered to be correlated. Therefore, any frequency point and the series of frequency points on its left or right are called adjacent frequency points.

[0068] The correlation coefficient R ranges from -1 to 1. A larger |R| value indicates a higher degree of linear correlation between frequency points. The closer |R| is to 0, the lower the degree of linear correlation between frequency points. The closer |R| is to 1, the stronger the linear correlation between frequency points. |R| = 1 indicates complete correlation, and |R| = 0 indicates no correlation at all.

[0069] Let S = {X -m ,......,X -3 ,X -2 ,X -1 ,X0,X1,X2,......,X m}, then the frequency sequence adjacent to X0 is

[0070] X -1 ,X -2 ,X -3 ,......,X -m and X1, X2, ..., X m , then the correlation coefficient R between X0 and its adjacent frequency point on the left is calculated as follows:

[0071]

[0072] The formula for calculating the correlation coefficient R between X0 and its adjacent frequency point on the right is as follows:

[0073]

[0074] where x 0，j ∈X0, k=1,2,3,...,m; j=1,2,3,...,n.

[0075] Associate the frequency point X0 with the adjacent frequency points, X -1 It is adjacent to the left of X0 once, X -2 It is adjacent to the left twice, and so on. -m is m times adjacent; X1 is 1 time adjacent to the right of X0, X2 is 2 times adjacent to the right, and so on. m is m times adjacent.

[0076] According to the number of adjacent frequency points of X0, there are m adjacent frequency points on both sides of X0, and the correlation coefficient formulas (2) and (3) are used to calculate the correlation coefficient L of the adjacent frequency points on the left and right sides of the frequency point X0. k ,R k .

[0077] The correlation coefficient of the adjacent frequency point on the left side of X0 is L k =R(X0,X -k ) (4)

[0078] The correlation coefficient of the adjacent frequency points on the right side of X0 is R k = R(X0, X k ) (5)

[0079] where k = 1, 2, 3,......, m.

[0080] 2. Taking the frequency point X0 as an example, the formulas for calculating its normalized maximum value, information entropy, duty cycle, and variance are as follows:

[0081] 1) Normalized maximum value (A0): A0 is obtained by normalizing the maximum value of the frequency point X0. The principle is as follows:

[0082] Set the threshold F. After normalization, if it is greater than the background noise F, it is considered 1 and there is a signal; if it is less than the background noise F, it is considered 0 and there is no signal; for values in the range (0, 1), a feature vector needs to be further used to determine whether there is a signal. The formula is as follows:

[0083]

[0084] where M0 is the maximum value of the frequency point X0, M0 = max{x 0,,j |x 0,,j ∈X0}, j = 1, 2,......, n; B0 is the maximum value of the background noise of the frequency point X0; F is the set threshold.

[0085] If A0 = 1, it is considered that there is a signal; if A0 = 0, it is considered that there is no signal; if 0 < A0 < 1, it is necessary to calculate through the feature vector to determine again whether a certain frequency point has a signal.

[0086] 2) Information entropy (P0): It represents the description of the information distribution and reflects the distribution of the uncertainty of random variables in the system.

[0087] The formula is as follows:

[0088]

[0089] where Q is the quantization level of the spectrum data X0; p h is the probability that the spectrum data appears at the h-th level of X0.

[0090] 3. Duty cycle (C0): It represents the peak value of a certain service within a sampling period. The formula is as follows:

[0091]

[0092] where T0 is the given number of samplings; V0 is the number of sampling values greater than the maximum value B0 of the background noise among T0 samplings.

[0093] 4. Variance (D0): Indicates the degree of fluctuation in the frequency intensity of each part of the signal, that is, the stability of the signal.

[0094] The formula is as follows:

[0095]

[0096] Among them, x 0，j ∈X0;j=1,2,3,......,n, is the mean of the frequency spectrum data at frequency point X0, X max is the maximum value of X0, X min is the minimum value of X0, and X max >X min .

[0097] Secondly, set the frequency point X0 to be adjacent to the Kth frequency point, and calculate the correlation coefficient of the 1st to Kth adjacent frequency points of X0. The correlation coefficient is L1, L2, ..., L K ,R1,R2,......,R K These correlation coefficients, together with multiple characteristic parameters such as information entropy P0, duty cycle C0, variance D0, and normalized maximum value A0, form a set of eigenvectors. That is, it is assumed that the signal of each set of data has 2*K+4 eigenvalues. The eigenvector U is represented as follows:

[0098] U=(P0,C0,D0,A0,L1,L2,......,L K ,R1,R2,......,R K ) (10)

[0099] 3. Algorithm Example

[0100] For ease of calculation, set the acquisition time for one data point to 50ms, and collect 20 data points continuously as a group. Data acquisition is performed in the 10-3600 (MB) frequency band. Within the 192MHz to 195MHz frequency band, continuous sampling is performed for 20s. The frequency points (MHz) are: 192.383096, 192.627234, 192.871375, 193.115516, 193.359657, etc. 5 frequency points are taken as an example, and 20 groups of spectrum data are listed to calculate the frequency correlation coefficient, normalized maximum value, information entropy, duty cycle, variance, etc. The 20 groups of data are given as follows:

[0101]

[0102] 1. Calculate the correlation coefficient of the adjacent frequency points of frequency point 192.871375. Let X0 be the frequency spectrum data sequence of 192.871375; -1 The spectrum data sequence is 192.383096, X-2 The spectral data sequence is 192.627234, X1 is the spectral data sequence of 193.115516, and X2 is the spectral data sequence of 193.359657

[0103] Associate X0 with adjacent frequency points, X -1 Is the first adjacent on the left side of X0, X -2 Is the second adjacent on the left side; X1 is the first adjacent on the right side of X0, and X2 is the second adjacent on the right side. Calculate the fourth adjacent of X0 according to the number of adjacent frequency points of X0, as shown below.

[0104] L1 = R(X0, X -1 ) ≈ 0.8706

[0105] L2 = R(X0, X -1 ) ≈ 0.9459

[0106] R1 = R(X0, X1) ≈ 0.95275

[0107] R2 = R(X0, X2) ≈ 0.9099

[0108] 2. Calculate the normalized maximum value, information entropy, duty cycle, and variance of the frequency point X0

[0109] 1) Calculate the normalized maximum value A0: From 20 groups of data, M0 = -76.07057 is obtained. Let B0 = -75 and F = 5.

[0110] From formula (6), A0 ≈ 0.6 is obtained. Since 0 < A0 < 1, it is necessary to calculate through the eigenvector to determine whether there is a signal at a certain frequency point again.

[0111] 2) Calculate the information entropy P0: Let Q = 8. From formula (7), P0 = 0.8798 is obtained.

[0112] 3) Calculate the duty cycle C0: From formula (8), C0 = 0.375 is obtained.

[0113] 4) Calculate the variance D0: Calculated from the data X max = -76.07057, X min = -94.98058. From formula (9), D0 ≈ 0.07776

[0114] 3. The eigenvector formed by the information entropy, duty cycle, variance, maximum value, and correlation coefficient of X0 is as follows:

[0115] U = (0.8798, 0.375, 0.07776, 0.6, 0.8706, 0.9459, 0.95275, 0.9099)

[0116] 4. Use feature vectors and related algorithms to perform signal detection and recognition.

[0117] 1. First, use a part of the data set to generate a training set, which is divided into two processes: data collection and data processing.

[0118] During the data collection experiment, a data point was collected every 50ms, with 40 consecutive data points collected as a group. Data was collected in the 10-3600 (MB) frequency band, with signals present at frequencies of 190MB and 145MB, ensuring that these two frequencies remained present for 2 seconds. During data processing, the signal eigenvalues ​​of each data set were calculated to form a eigenvector, which served as the training set.

[0119] 2. Use the single-class SVM algorithm to train the training set to form a single-class SVM model. The optimized model parameters formed using Python are as follows:

[0120] svm.setType(cv.ml.SVM_ONE_CLASS) / / Type is single classification SVM

[0121] svm.setKernel(cv.ml.SVM_RBF) / / Select RBF kernel function

[0122] svm.setDegree(3) / / polynomial kernel time variable order

[0123] svm.setGamma(21.78) / / After selecting the RBF function as the kernel, the RBF function comes with the parameter gamma

[0124] svm.setC(1) / / Set the penalty coefficient C value

[0125] svm.setP(5e-3) / / Set parameter γ

[0126] svm.setNu(0.1) / / Set parameters∈

[0127] criteria=(cv.TERM_CRITERIA_EPS,1000,1e-7)

[0128] svm.setTermCriteria(criteria) / / Set training termination conditions

[0129] 3. Model detection. Use model detection to identify whether there is a signal at any frequency point in the spectrum data. The pseudo code for Python implementation is as follows:

[0130] SVM.load(model) / / Load the classification model

[0131] ret = svmpredict(test_data[i],model); / / Get the predicted value of each group of data in the test set

[0132] if(ret == 1) / / get result label as 1

[0133] {

[0134] msgBox("There is a signal");

[0135] }

[0136] else

[0137] {

[0138] msgBox("No signal");

[0139] }

[0140] CR=ClassResult(test_label,test_data,model) / / Calculate the evaluation index of the classification model V. Implementation results

[0141] The generated training data set, test data set and final detection results are as follows:

[0142] During model training, sample collection was performed in a microwave anechoic chamber. A data point was collected every 50ms, with 40 consecutive data points collected as a set. The collected spectrum data served as the training data set. During collection, a continuous or intermittent signal with an intensity between -20dB and -10dB was required at frequencies 190MB and 145MB. The signal was ensured to exist within 2 seconds at each frequency point. The spectrum was continuously scanned from 10MB to 3600MB with a step size of 244,141Hz for 30 minutes. A total of 139,974 positive sample signals were selected for the training set. The specific frequency bands used for sample collection are shown in the following table.

[0143]

[0144] When performing model testing, sample collection is carried out in a microwave darkroom. It is set to collect one data piece every 50ms, and 40 data pieces are collected continuously as a group of data. The collected spectrum data is used as the training data set. During collection, it is necessary to provide a continuous signal or intermittent signal with an intensity within the range of -20Db to -10Db at the frequency points 100MB and 190MB, and ensure that the signal frequency point exists within 2 seconds. The spectrum of the 10MB to 3600MB frequency band is continuously scanned with a step size of 244141Hz. The collection is continuous for 30 minutes. The specific sample data set selected is 3270427 groups, of which 9000 groups are positive samples and 3261427 groups are negative samples. The specific selected frequency band sample results are as follows:

[0145]

[0146] Example

[0147] The generated training data set, test data set and final detection results are as follows:

[0148] During model training, sample collection was performed in a microwave anechoic chamber. A data point was collected every 50ms, with 40 consecutive data points collected as a set. The collected spectrum data served as the training data set. During collection, a continuous or intermittent signal with an intensity between -20dB and -10dB was required at frequencies 190MB and 145MB. The signal was ensured to exist within 2 seconds at each frequency point. The spectrum was continuously scanned from 10MB to 3600MB with a step size of 244,141Hz for 30 minutes. A total of 139,974 positive sample signals were selected for the training set. The specific frequency bands used for sample collection are shown in the following table.

[0149]

[0150] When performing model testing, sample collection is carried out in a microwave darkroom. It is set to collect one data piece every 50ms, and 40 data pieces are collected continuously as a group of data. The collected spectrum data is used as the training data set. During collection, it is necessary to provide a continuous signal or intermittent signal with an intensity within the range of -20Db to -10Db at the frequency points 100MB and 190MB, and ensure that the signal frequency point exists within 2 seconds. The spectrum of the 10MB to 3600MB frequency band is continuously scanned with a step size of 244141Hz. The collection is continuous for 30 minutes. The specific sample data set selected is 3270427 groups, of which 9000 groups are positive samples and 3261427 groups are negative samples. The specific selected frequency band sample results are as follows:

[0151]

[0152] The above only describes in detail the preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary technicians in this field without departing from the purpose of the present invention, and various changes should be included in the scope of protection of the present invention.

Claims

1. A method for detecting electromagnetic spectrum signals based on correlation between frequency points, characterized by: The following steps are involved: S1. Calculate the correlation coefficient of adjacent frequency points; S2. Calculate normalized maximum value, information entropy, duty cycle, and variance characteristic parameters; The method for calculating the normalized maximum value of S2 is: Normalized maximum value : （S-6） in, Frequency The maximum value of =max{x 0,,j |x 0,,j }, j=1,2,......,n,; Frequency Maximum value of background noise; is the threshold value set; S3, forming a feature vector by combining the correlation coefficient of adjacent frequency points with the normalized maximum value, information entropy, duty cycle, and variance feature parameters; S4. detecting electromagnetic spectrum signals; The method for detecting the electromagnetic spectrum signal in S4 is: like , then it is considered that there is a signal; if =0, it is considered that there is no signal; if 0< <1, it is used as an eigenvalue to participate in the eigenvector calculation.

2. The electromagnetic spectrum signal detection method based on frequency correlation according to claim 1, characterized in that: The method for calculating the correlation coefficient of adjacent frequency points in S1 is: Assume that there are n time slots of equal length continuous spectrum data sequence, and each data sequence has 2*m+1 frequency points, that is, there is the following data sequence matrix: （S-1） make ,but The adjacent frequency sequence is as well as ,but The calculation formula of the correlation coefficient R with the adjacent frequency point on its left is as follows: （S-2） The calculation formula of the correlation coefficient R with the adjacent frequency point on the right is as follows: （S-3） in The frequency Associate with adjacent frequency points, for 1 adjacent on the left, It is 2 times adjacent to the left, and so on. is m times adjacent; for 1 adjacent on the right, It is adjacent to the right twice, and so on is m times adjacent; according to The number of adjacent frequency points, There are m adjacent frequencies on both sides, and the correlation coefficient formulas (S-2) and (S-3) are used to calculate the correlation coefficients of the adjacent frequency points on the left and right sides of the frequency point X0. ; The correlation coefficient of the adjacent frequency points on the left is (S-4) The correlation coefficient of the adjacent frequency points on the right is (S-5) in .

3. The electromagnetic spectrum signal detection method based on frequency correlation according to claim 1, characterized in that: The method for calculating information entropy in S2 is: Information entropy : （S-7） Where Q is the spectrum data the level at which quantification is performed; For spectrum data The probability of appearing at level h.

4. The electromagnetic spectrum signal detection method based on frequency correlation according to claim 1, characterized in that: The method of the duty cycle in S2 is: Duty cycle : （S-8） in, is the given number of sampling times; for Among the samples, there are The sampling value is greater than the maximum background noise .

5. The electromagnetic spectrum signal detection method based on inter-frequency correlation according to claim 1, characterized in that: The method for the variance in S2 is: variance : （S-9） in, , Frequency The mean of the spectral data, for The maximum value of for The minimum value of > .

6. The electromagnetic spectrum signal detection method based on inter-frequency correlation according to claim 1, characterized in that: The method for forming a feature vector by combining the correlation coefficient of adjacent frequency points with the normalized maximum value, information entropy, duty cycle, and variance feature parameters in S3 is: set up The adjacent frequency points are taken K times and the frequency points are calculated. The correlation coefficient of the 1st to Kth adjacent frequency points is , , these correlation coefficients and information entropy , duty cycle ,variance , normalized maximum value Multiple characteristic parameters are formed to form a set of characteristic vectors, that is, it is considered that the signal of each set of data has 2*K+4 eigenvalues, and the characteristic vector U is expressed as follows: U=( , , , ,L1,L2, ...... ,L K ,R1,R2,.....,R K )(S-10)。

Images