A method for extracting shaft-leaf frequency from a demon spectrum

Abstract

The application discloses a DEMON spectrum analysis shaft leaf frequency extraction method, belongs to the field of underwater acoustic target recognition, and can be used for passive sonar DEMON spectrum analysis, extraction of target shaft frequency leaf frequency information, and improvement of target recognition capability. In the pretreatment stage, the received data is subjected to spatial domain filtering through beam synthesis to improve the signal-to-noise ratio; the background of the DEMON spectrum is balanced through the sorting truncation average method to more accurately extract the line spectrum; and then the improved greatest common divisor method is used to obtain the weight of the line spectrum difference frequency to accurately extract the shaft frequency and leaf frequency information. In the case that the number of line spectrums is insufficient at an unstable moment, the line spectrum weight is weighted to improve the data utilization rate. The method disclosed by the application effectively improves the accuracy of DEMON spectrum analysis shaft leaf frequency extraction.

Description

Technical Field

[0001] This invention belongs to the field of underwater acoustic target recognition, and specifically relates to a DEMON spectrum analysis method for extracting shaft and blade frequencies. It can be used for passive sonar noise demodulation (Detection of Envelope Modulation on Noise, DEMON) spectrum analysis to accurately extract target shaft and blade frequency information and improve target recognition capabilities. Background Technology

[0002] Underwater ship target identification is of great significance for maritime defense and marine resource and environmental monitoring. Currently, ship target identification mainly relies on the ship's radiated noise, which consists primarily of propeller noise, internal mechanical noise, and hydrodynamic noise. Compared to mechanical and hydrodynamic noise, propeller noise has a higher energy level and multiple stable frequency components. Propeller noise consists of a discrete line spectrum and a high-frequency continuous spectrum. The propeller's rotation causes its shaft frequency and blade frequency to modulate cavitation noise. The shaft frequency equals the propeller rotation frequency, and the blade frequency is numerically equal to the product of the shaft frequency and the number of blades. From the shaft and blade frequency information, important ship identification features such as the target's propeller speed and number of blades can be obtained. However, the target's shaft and blade frequency information cannot be directly obtained from the target's power spectrum. DEMON spectrum analysis is needed to perform spectrum shifting of the continuous spectrum to extract the shaft frequency and its harmonic frequencies, thereby obtaining information such as the ship target's propeller speed and number of blades for target classification and identification.

[0003] Traditional DEMON spectrum extraction methods for axis and leaf frequencies include the greatest common divisor (GCD) method, the remainder threshold method, the harmonic extraction method, and neural network-based methods. The GCD and remainder threshold methods are prone to missing line spectrum values, resulting in low overall extraction accuracy at low signal-to-noise ratios. The harmonic extraction method suffers from the difficulty in selecting the axis frequency assumption, which significantly impacts the algorithm's extraction performance. Artificial intelligence methods heavily rely on the size of the training sample, but obtaining samples from adversarial targets is challenging, affecting the extraction results. Therefore, it is necessary to design and propose a DEMON spectrum analysis method with low line spectrum omission probability, high accuracy in axis and leaf frequency extraction, and independence from training samples. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a method for extracting shaft and blade frequencies using DEMON spectrum analysis. This addresses the issue of low accuracy in extracting shaft and blade frequencies under low signal-to-noise ratio conditions using traditional DEMON spectrum analysis methods.

[0005] The technical problem to be solved by this invention is achieved by the following technical solution:

[0006] A method for extracting leaf frequency from DEMON spectrum analysis includes the following steps:

[0007] (1) Perform frequency domain beamforming on the array received data to obtain multiple sets of demodulated DEMON spectrum data in the target direction;

[0008] (2) Background equalization was performed on each group of demodulated DEMON spectrum data using the sorted truncation averaging method, and line spectra were extracted. The set of line spectrum frequency points for each group of data was denoted as P = {f1, f2, ..., f K};

[0009] (3) Sort the extracted line spectra according to intensity, take the I line spectra with the highest intensity, sort them in ascending order of frequency, and denote the set of frequency points of the resulting I line spectra as Q={f1,f2,...,f I}; where I is the set value;

[0010] (4) Calculate the difference frequency between the frequency points in the line spectrum frequency point set Q to obtain the difference frequency point set, and calculate the weight corresponding to each difference frequency point;

[0011] (5) Update the difference frequency weight, take the difference frequency corresponding to the maximum weight as the axis frequency of the data, obtain the axis frequency value of each group of data, and put it into the axis frequency set;

[0012] (6) The shaft frequency that appears most frequently in the shaft frequency set is taken as the shaft frequency extraction result f. Z ;

[0013] (7) Extract the blade frequency based on the shaft frequency extraction result to obtain the blade frequency extraction result f. Y .

[0014] Furthermore, step (2) specifically includes the following steps:

[0015] (201) For each set of demodulated DEMON spectrum data X(N)={x(1),x(2),...,x(N)}, each element x(n) in X(N), n=1,2,...,N, where N is the number of demodulated DEMON spectrum data in each set, the data is expanded with a window length M, resulting in 2M+1 data:

[0016] x(nM),...,x(n-1),x(n),x(n+1),...,x(n+M)

[0017] Arrange the data in ascending order as follows:

[0018] y(1),y(2),...,y(2M+1)

[0019] If the median of the sequence is y(M+1), then the cutoff mean of the sequence is:

[0020]

[0021] The sequence is truncated and the mean is corrected. After removing the points, we get:

[0022]

[0023] Among them, the rejection threshold parameter α is a constant and M is a set value;

[0024] After processing each element x(n) in X(N), n = 1, 2, ..., N, a new set of data Z(N) after background equalization is obtained;

[0025] (202) Set the shaft frequency and blade frequency detection and extraction interval [f min ,f max Extract the line spectrum within the detection interval from Z(N), where the frequency interval of the extracted line spectrum is greater than or equal to f. min The frequency resolution of Z(N) is f. re =f s / N d Extracting interval N from spectral points span =floor(f min / f re Extracting line spectra from Z(N) yields the line spectrum frequency set P = {f1, f2, ..., f...} K}; where N d The number of points for DFT when performing frequency domain beamforming on X(N), f s is the sampling rate, and floor is the floor value for rounding down.

[0026] Furthermore, step (3) specifically includes:

[0027] For each set of extracted line spectra P, the line spectra are sorted by intensity. Under stable signal and receiving environment conditions, the I line spectra with the highest intensity are selected and sorted by frequency from smallest to largest. The resulting set of I line spectra frequencies is denoted as Q = {f1, f2, ..., f...} I The line spectral intensities corresponding to each frequency in the frequency set are S = {s1, s2, ..., s}. I}; Each frequency f in the line spectrum frequency set i The corresponding normalized weight is w i =s i / max(S);

[0028] In situations where the signal and receiving environment are unstable, only I can be extracted. s Root line spectrum, I s <I, at this point, the weights corresponding to the line spectrum data at that moment are adjusted proportionally to w. i =s i I / max(S)I s .

[0029] Furthermore, step (4) includes the following specific steps:

[0030] (401) Calculate the difference frequencies of each set of line spectra in Q in turn:

[0031] F ij =|f i -f j |,i=1,2,...,I-1;j=2,3,...,I;i<j

[0032] W ij =w i +w j ,i=1,2,...,I-1;j=2,3,...,I;i<j

[0033] Among them, F ij For the obtained difference frequency, W ij The weights are the difference frequencies;

[0034] (402) For the obtained frequency difference set F, the elements whose frequency difference is less than the set value are unified into one frequency value;

[0035] Let the corresponding difference frequency weight be W. a and W b The difference frequency F a and F b , satisfying |F a -F b If |<ε, then the difference frequency F a and F b Unified as F ab F ab The calculation method is as follows:

[0036]

[0037] Unified difference frequency value F ab The corresponding weight is W ab =W a +W b After unification is completed, for elements with the same frequency value in the difference frequency set F, only one is retained, and the corresponding weights are added together; where the adjustment difference frequency unification parameter ε is a constant.

[0038] Furthermore, step (5) includes the following specific steps:

[0039] (501) For each set of difference frequencies F, use the line frequency difference ratio β = f i / F ij To update the difference frequency weight value, if β satisfies:

[0040]

[0041] Then the difference frequency F ij The corresponding weight W ij Add 1; where round means rounding to the nearest whole number.

[0042] (502) After updating the weights corresponding to each group of difference frequencies, take the difference frequency corresponding to the maximum weight in each group of weights as the axis frequency of the data and put it into the axis frequency set.

[0043] Furthermore, step (7) specifically includes:

[0044] The blade frequency is numerically equal to the shaft frequency multiplied by the number of blades. For the obtained shaft frequency result f... Z At frequency f i ∈[3f Z ,7f Z The frequency with the highest line spectrum intensity within the range is taken as the leaf frequency result f. Y .

[0045] Compared with the prior art, the present invention has the following advantages:

[0046] Compared with traditional methods, the present invention has a higher accuracy in extracting the blade frequency.

[0047] This invention improves the signal-to-noise ratio of the DEMON spectrum by performing spatial filtering on the array-received data through beamforming.

[0048] In the DEMON spectrum preprocessing process, this invention performs background equalization on the spectral data using the sorted truncation averaging method, eliminating the interference of strong noise on weak line spectra, making the modulated line spectra more obvious and easier to extract.

[0049] This invention addresses the situation where the number of data line elements is less than I at certain times by weighting the line element weights to improve data utilization and obtain higher time gain. Attached Figure Description

[0050] Figure 1 This is a flowchart of the blade frequency extraction process of the present invention;

[0051] Figure 2 This is the DEMON spectral history diagram obtained after beamforming during data testing;

[0052] Figure 3 This is the history diagram obtained by equalizing the background of the beamforming DEMON spectrum history diagram during data testing using this invention.

[0053] Figure 4 This is a comparison chart of the shaft frequency and blade frequency extraction results of the present invention and the traditional method in data testing. Detailed Implementation

[0054] The technical solution and effects of the present invention will be further described in detail below with reference to the accompanying drawings.

[0055] The present invention provides a method for extracting leaf frequency from DEMON spectrum analysis, the specific process of which is as follows: Figure 1 As shown, the implementation steps are as follows:

[0056] Step 1: Perform frequency domain beamforming on the array received data to obtain multiple sets of demodulated DEMON spectrum data in the target direction.

[0057] Step 2: Perform background equalization on each group of DEMON spectra using the sorted truncated averaging method, and extract the line spectra. The set of line spectrum frequency points for each group of data is denoted as P = {f1, f2, ..., f...} K}

[0058] (201) For each set of DEMON spectrum data X(N)={x(1),x(2),...,x(N)}, assume that the nth element x(n) of the set of data is processed, n=1,2,...,N, where N is the number of DEMON spectrum data in each set of demodulated data. In order to eliminate the edge effect of data, the data is expanded with a window length M, and after expansion, 2M+1 data are obtained:

[0059] x(nM),...,x(n-1),x(n),x(n+1),...,x(n+M)

[0060] Arrange the data in ascending order as follows:

[0061] y(1),y(2),...,y(2M+1)

[0062] The median of the sequence is y(M+1), and the cutoff mean of the sequence is defined as:

[0063]

[0064] The sequence is truncated and the mean is corrected. After removing the points, we get:

[0065]

[0066] Among them, the rejection threshold parameter α is a constant, usually between 1.0 and 1.1, and M is a set value;

[0067] By performing the above processing on each element x(n) in X(N), n = 1, 2, ..., N, we can obtain a new set of data Z(N) after background equalization.

[0068] (202) Set the shaft frequency and blade frequency detection and extraction interval [f min ,f maxSince the modulation spectrum mainly exists in the low-frequency range and is usually 1-100Hz, f is generally set to... min =1,f max =100. Extract the line spectrum within the detection interval from Z(N), since the axis frequency is greater than or equal to f. min Therefore, when extracting the line spectrum, the frequency interval of the extracted line spectrum should be greater than or equal to f. min For doing N d The DEMON spectrum Z(N) obtained by point DFT has a frequency resolution of f. re =f s / N d Extracting interval N from spectral points span =floor(f min / f re Extracting line spectra from Z(N) yields the line spectrum frequency set P = {f1, f2, ..., f...} K}; where f s is the sampling rate, and floor is the floor value for rounding down.

[0069] Step 3: Sort the extracted line spectra by intensity, take the I line spectra with the highest intensity, and sort them by frequency from smallest to largest. Denote the set of frequency points of the resulting I line spectra as Q = {f1, f2, ..., f...} I}

[0070] For each set of extracted line spectra P, the line spectra are sorted by intensity. Under stable signal and receiving environment conditions, the I line spectra with the highest intensity are selected and sorted by frequency from smallest to largest. The resulting set of I line spectra frequencies is denoted as Q = {f1, f2, ..., f...} I The line spectral intensities corresponding to each frequency in the frequency set are S = {s1, s2, ..., s}. I}

[0071] Each frequency f in the line spectrum frequency set i The corresponding normalized weight is w i =s i / max(S).

[0072] For the line spectrum set Q, due to the instability of the signal and the receiving environment, only I can be extracted from the DEMON spectrum data obtained at certain times. s Root line spectrum, I s <I. At this point, the weights corresponding to the line spectrum data at that moment are adjusted proportionally to w. i =s i I / max(S)I s .

[0073] Step 4: Calculate the difference frequencies between the frequency points in the line spectrum frequency point set Q to obtain the difference frequency point set, and calculate the weight corresponding to each difference frequency point.

[0074] (401) For each set of line spectra Q, calculate the difference frequency between the frequency points in each set in turn. The weight of the difference frequency is the sum of the weights of the frequencies of the two line spectra that are differed:

[0075] F ij =|f i -f j |,i=1,2,...,I-1;j=2,3,...,I;i<j

[0076] W ij =w i +w j ,i=1,2,...,I-1;j=2,3,...,I;i<j

[0077] Among them, F ij For the difference frequency, W ij For the difference frequency F ij The corresponding weights.

[0078] (402) For the obtained difference frequency point set F, unify the elements with similar frequency values ​​into one frequency value, and ensure that the same difference frequency appears only once in the set.

[0079] Assume the corresponding difference frequency weight is W a and W b The difference frequency F a and F b , satisfying |F a -F b |<ε, where the adjustment difference frequency uniformity parameter ε is a constant, typically taken as 0.5, corresponding to a minimum difference frequency of 1Hz. Then the difference frequency F a and F b Unified as F ab F ab The calculation method is as follows:

[0080]

[0081] Unified difference frequency value F ab The corresponding weight is W ab =W a +W b After unifying the elements with similar frequency values, for elements with the same frequency value in the difference frequency set F, only one is retained, and their corresponding weights are added together.

[0082] Step 5: Update the difference frequency weights, take the difference frequency corresponding to the maximum weight as the axis frequency of the data, obtain the axis frequency value of each group of data, and put it into the axis frequency set.

[0083] (501) For each set of difference frequencies F, use the line frequency difference ratio β = fi / F ij To update the difference frequency weight value, if β satisfies:

[0084]

[0085] Then the difference frequency F ij The corresponding weight W ij Add 1. Here, "round" indicates rounding.

[0086] (502) After updating the weights corresponding to each group of difference frequencies, take the difference frequency corresponding to the maximum weight in each group of weights as the axis frequency of the data and put it into the axis frequency set.

[0087] Step 6: The shaft frequency that appears most frequently in the shaft frequency set is taken as the shaft frequency extraction result f. Z .

[0088] Step 7: Extract the blade frequency based on the shaft frequency result to obtain the blade frequency extraction result f. Y .

[0089] The blade frequency is numerically equal to the shaft frequency multiplied by the number of blades; ship propellers typically have 3 to 7 blades. The obtained shaft frequency result f... Z At frequency f i ∈[3f Z ,7f Z The frequency with the highest line spectrum intensity within the range is taken as the leaf frequency result f. Y .

[0090] The effectiveness of this invention can be illustrated by the following test results:

[0091] 1. Data testing conditions

[0092] Select a specific test to collect data. The target propeller is a 5-bladed propeller, corresponding to a shaft frequency of 3.05 Hz and a blade frequency of 15.25 Hz.

[0093] 2. Data Test Results

[0094] like Figure 2 The image shows the DEMON spectrum obtained by beamforming the acquired data in the target direction. The spectrum obtained after sorting, truncating, averaging, and equalizing the DEMON spectrum is shown below. Figure 3 As shown, the weak line spectrum of the DEMON spectrum is significantly enhanced after background equalization, facilitating the accurate extraction of the line spectrum frequencies for subsequent processing. The proposed DEMON spectrum analysis method for extracting blade frequencies is compared with traditional methods such as the greatest common divisor method and harmonic detection method. The blade frequency results obtained by each method are shown below. Figure 4As shown in the figure. Experimental results show that the accuracy of shaft frequency and blade frequency extraction by the method proposed in this invention is significantly improved compared with the existing greatest common divisor method and harmonic detection method.

[0095] In summary, the present invention proposes a method for extracting axis and leaf frequencies from DEMON spectrum analysis, achieving accurate extraction of axis and leaf frequencies from the DEMON spectrum. Spatial filtering of the array-received data is performed using beamforming to improve the signal-to-noise ratio of the DEMON spectrum. During DEMON spectrum preprocessing, background equalization is performed on the spectral data using a sorted truncation averaging method to eliminate interference from strong noise on weak line spectra, making the modulated line spectra more prominent. For cases where the number of data line elements is less than I at certain times, the line spectrum weights are weighted to improve data utilization and achieve higher temporal gain.

Claims

1. A method for extracting leaf frequency from DEMON spectrum analysis, characterized in that, Includes the following steps: (1) Perform frequency domain beamforming on the array received data to obtain multiple sets of demodulated DEMON spectrum data in the target direction; (2) Background equalization was performed on each group of demodulated DEMON spectrum data using the sorted truncation averaging method, and line spectra were extracted. The set of line spectrum frequency points for each group of data was denoted as P = {f1, f2, ..., f K }; (3) Sort the extracted line spectra according to intensity, take the I line spectra with the highest intensity, sort them in ascending order of frequency, and denote the set of frequency points of the resulting I line spectra as Q={f1,f2,...,f I }; where I is the set value; (4) Calculate the difference frequency between the frequency points in the line spectrum frequency point set Q to obtain the difference frequency point set, and calculate the weight corresponding to each difference frequency point; (5) Update the difference frequency weight, take the difference frequency corresponding to the maximum weight as the axis frequency of the data, obtain the axis frequency value of each group of data, and put it into the axis frequency set; (6) The shaft frequency that appears most frequently in the shaft frequency set is taken as the shaft frequency extraction result f. Z ; (7) Extract the blade frequency based on the shaft frequency extraction result to obtain the blade frequency extraction result f. Y .

2. The method for extracting leaf frequency from DEMON spectrum analysis according to claim 1, characterized in that, Step (2) includes the following specific steps: (201) For each set of demodulated DEMON spectrum data X(N)={x(1),x(2),...,x(N)}, each element x(n) in X(N), n=1,2,...,N, where N is the number of demodulated DEMON spectrum data in each set, the data is expanded with a window length M, resulting in 2M+1 data: x(nM),...,x(n-1),x(n),x(n+1),...,x(n+M) Arrange the data in ascending order as follows: y(1),y(2),...,y(2M+1) If the median of the sequence is y(M+1), then the cutoff mean of the sequence is: The sequence is truncated and the mean is corrected. After removing the points, we get: Among them, the rejection threshold parameter α is a constant, and M is a set value; After processing each element x(n) in X(N), n = 1, 2, ..., N, a new set of data Z(N) after background equalization is obtained; (202) Set the shaft frequency and blade frequency detection and extraction interval [f min ,f max Extract the line spectrum within the detection interval from Z(N), where the frequency interval of the extracted line spectrum is greater than or equal to f. min The frequency resolution of Z(N) is f. re =f s / N d Extracting interval N from spectral points span =floor(f min / f re Extracting line spectra from Z(N) yields the line spectrum frequency set P = {f1, f2, ..., f...} K }; where N d The number of points for DFT when performing frequency domain beamforming on X(N), f s is the sampling rate, and floor is the floor value for rounding down.

3. The method for extracting leaf frequency from DEMON spectrum analysis according to claim 1, characterized in that, Step (3) specifically includes: For each set of extracted line spectra P, the line spectra are sorted by intensity. Under stable signal and receiving environment conditions, the I line spectra with the highest intensity are selected and sorted by frequency from smallest to largest. The resulting set of I line spectra frequencies is denoted as Q = {f1, f2, ..., f...} I The line spectral intensities corresponding to each frequency in the frequency set are S = {s1, s2, ..., s}. I }; Each frequency f in the line spectrum frequency set i The corresponding normalized weight is w i =s i / max(S); In situations where the signal and receiving environment are unstable, only I can be extracted. s Root line spectrum, I s <I, at this point, the weights corresponding to the line spectrum data at that moment are adjusted proportionally to w. i =s i I / max(S)I s .

4. The method for extracting leaf frequency from DEMON spectrum analysis according to claim 3, characterized in that, Step (4) includes the following specific steps: (401) Calculate the difference frequencies of each set of line spectra in Q in turn: F ij =|f i -f j |,i=1,2,...,I-1;j=2,3,...,I;i<j W ij =w i +w j ,i=1,2,...,I-1;j=2,3,...,I;i<j Among them, F ij For the obtained difference frequency, W ij The weights are the difference frequencies; (402) For the obtained frequency difference set F, the elements whose frequency difference is less than the set value are unified into one frequency value; Let the corresponding difference frequency weight be W. a and W b The difference frequency F a and F b , satisfying |F a -F b If |<ε, then the difference frequency F a and F b Unified as F ab F ab The calculation method is as follows: Unified difference frequency value F ab The corresponding weight is W ab =W a +W b After unification is completed, for elements with the same frequency value in the difference frequency set F, only one is retained, and the corresponding weights are added together; where the adjustment difference frequency unification parameter ε is a constant.

5. The method for extracting leaf frequency from DEMON spectrum analysis according to claim 4, characterized in that, Step (5) includes the following specific steps: (501) For each set of difference frequencies F, use the line frequency difference ratio β = f i / F ij To update the difference frequency weight value, if β satisfies: Then the difference frequency F ij The corresponding weight W ij Add 1; where round means rounding to the nearest whole number. (502) After updating the weights corresponding to each group of difference frequencies, take the difference frequency corresponding to the maximum weight in each group of weights as the axis frequency of the data and put it into the axis frequency set.

6. The method for extracting leaf frequency from DEMON spectrum analysis according to claim 1, characterized in that, Step (7) specifically includes: The blade frequency is numerically equal to the shaft frequency multiplied by the number of blades. For the obtained shaft frequency result f... Z At frequency f i ∈[3f Z ,7f Z The frequency with the highest line spectrum intensity within the range is taken as the leaf frequency result f. Y .

Images