An Automated Method for Extracting DEMON Spectral Features by Multi-Subband Fusion

Abstract

This invention proposes an automated method for extracting DEMON spectral features using multi-subband fusion, belonging to the field of DEMON spectral feature extraction for ship radiated noise. First, this invention utilizes a parameter-optimized wavelet packet decomposition algorithm for adaptive decomposition of the signal to be analyzed. The number of decomposition levels is pre-determined by an empirical mode decomposition algorithm, solving the problems of blindness and mismatch in fixed decomposition levels during different signal decomposition processes, and achieving adaptive selection of decomposition parameters. Then, a higher-order spectral analysis method is used to calculate the DEMON spectrum for each order wavelet packet component, further suppressing marine environmental noise and improving the signal-to-noise ratio of the DEMON spectrum harmonic group line spectrum. Finally, wavelet packet components with significant modulation information are selected for DEMON spectrum calculation and fusion, avoiding the need for manual selection of bandpass filter cutoff frequencies during DEMON spectrum calculation.

Description

Technical Field

[0001] This invention relates to the field of underwater acoustic signal processing, specifically to an automated method for extracting DEMON (Detection of Envelope Modulation On Noise) spectral features through multi-subband fusion. Background Technology

[0002] When ships and other underwater acoustic targets are navigating or operating, their mechanical systems and propellers are in motion. The vibrations they generate radiate sound waves outward through the hull, forming radiated ship noise. The components of radiated ship noise mainly include propeller noise, mechanical noise, and hydrodynamic noise, which manifests as a superposition of continuous and line spectra in the frequency domain. The propeller noise varies depending on the target type or navigation state, allowing for the classification and identification of underwater acoustic targets.

[0003] During navigation, underwater acoustic targets generate a series of periodic signals due to the rotation of their propellers. These signals are modulated by cavitation noise from bursting bubbles, resulting in a shift of discrete line spectra from low frequencies to higher frequencies. Modulation information exists across different frequency bands, exhibiting uneven modulation. These discrete line spectrum systems characterize the propeller shaft frequency, blade frequency, and their harmonic frequencies. Since other noise sources during navigation are predominantly located in the low-frequency range, it is difficult to directly extract the periodic signals from the propeller rotation from the low-frequency spectrum. Therefore, demodulating the high-frequency components containing modulation information allows for the separation of the low-frequency line spectra from the high-frequency modulation continuous spectrum. From these demodulated harmonic groups, characteristic parameters such as the propeller shaft frequency, blade frequency, number of blades, and speed can be estimated, ultimately providing crucial information for target identification.

[0004] See Figure 1 The traditional demodulation method for the DEMON spectrum generally involves first designing a bandpass filter, selecting a frequency band in the underwater acoustic target radiated noise signal where there is obvious modulation information, then performing square demodulation or absolute value demodulation, then passing the signal through a low-pass filter, and finally performing an FFT transform to obtain the DEMON spectrum.

[0005] The aforementioned method for DEMON spectrum extraction suffers from significant uncertainty and ambiguity due to the need for manually designing the cutoff frequency of the bandpass filter, with unclear criteria for frequency band selection. When the selected frequency band's modulation characteristics are weak, the demodulated harmonic spectrum components will also be indistinct, severely interfering with or even misleading subsequent propeller characteristic parameter estimation. Furthermore, even with strong modulation characteristics in the selected frequency band, the non-uniformity of propeller signal modulation across the entire frequency band will distort the demodulated harmonic spectrum, affecting subsequent blade frequency and blade number parameter estimation. Moreover, manually selecting the demodulation frequency band cannot meet the requirements for real-time analysis in actual equipment or application in unmanned equipment. Finally, after low-pass filtering, directly calculating the power spectrum and detecting the line spectrum results in low signal-to-noise ratio, hindering harmonic spectrum detection. Therefore, effectively and automatically extracting DEMON spectrum features and estimating related parameters has become an urgent problem to be solved. Summary of the Invention

[0006] To address the problems of traditional DEMON spectral feature extraction methods, such as the need to manually select the bandpass filter cutoff frequency, low signal-to-noise ratio of harmonic group line spectra, and large fluctuations in background noise, this invention provides an automated DEMON spectral feature extraction method based on multi-subband fusion.

[0007] The technical solution adopted in this invention is as follows:

[0008] An automated method for extracting DEMON spectral features through multi-subband fusion includes the following steps:

[0009] S1, Optimal selection of wavelet packet decomposition level: Based on the empirical mode decomposition algorithm, the original ship radiated noise signal is pre-decomposed to determine the optimal wavelet packet decomposition level, and the original ship radiated noise signal is decomposed into wavelet packets according to the optimal wavelet packet decomposition level.

[0010] S2, Wavelet packet component selection: Calculate the spectrum of the wavelet packet components obtained in step S1, and select wavelet packet components without spectrum contamination.

[0011] S3, Squared Filtering and Higher-Order Spectrum Calculation: The wavelet packet components selected in step S2 are squared, and then the envelope signal is obtained through a low-pass filter and 1 (1 / 2) higher-order spectrum calculation is performed.

[0012] S4, Higher-order DEMON spectrum fusion: The 1 (1 / 2) higher-order spectra of the envelopes of each wavelet packet component calculated in step S3 are fused to obtain the DEMON higher-order spectrum;

[0013] S5, Feature Extraction and Parameter Estimation: Harmonic groups are extracted from the DEMON high-order spectrum obtained in step S4, and the shaft frequency, blade frequency and number of blades of the target ship propeller are estimated based on empirical formulas to complete the DEMON spectrum feature extraction.

[0014] Furthermore, the specific method of step S1 is as follows:

[0015] S11: Based on the empirical mode decomposition algorithm, the original ship radiated noise signal is pre-decomposed until the decomposition reaches the termination condition, and the decomposition level N is recorded.

[0016] S12: Use N as the optimal wavelet packet decomposition level and perform wavelet packet decomposition on the original ship radiated noise signal.

[0017] Furthermore, the specific method of step S2 is as follows:

[0018] S21: Perform FFT calculation on the wavelet packet components obtained from step S1 to obtain the spectrum of each order of wavelet packet components;

[0019] S22: Calculate the -3dB cutoff frequency for the spectrum of wavelet packet components of each order from low to high. When the lower cutoff frequency is higher than 100Hz, all wavelet packet components of lower order are discarded.

[0020] Furthermore, the specific method of step S3 is as follows:

[0021] S31: Calculate the square of each wavelet packet component obtained in step S2;

[0022] S32: Perform low-pass filtering on the signals output in step S31 to obtain the signal quantity containing the envelope modulation information of the target ship's propeller.

[0023] S33: For the signal output in step S32, calculate its 1 (1 / 2) higher-order spectrum, and normalize the energy of the higher-order spectrum of each component. The formula for calculating the higher-order spectrum is:

[0024]

[0025] Where x(t) is a time-domain signal sequence.

[0026] Furthermore, the specific method of step S5 is as follows:

[0027] S51: Design a bidirectional α smoothing filter and smooth the DEMON high-order spectrum obtained in step S4 to remove the trend term;

[0028] S52: Calculation of axis frequencies of DEMON spectrum harmonic groups based on the greatest common divisor method;

[0029] S53: Based on the multiple relationship between the shaft frequency and its harmonic group frequencies, solve for the frequencies and amplitudes corresponding to the 2nd to 8th harmonics in the DEMON higher-order spectrum;

[0030] S54: Based on the harmonic group vector obtained in S53, search for the amplitude values ​​of the 3rd to 7th harmonics. If there is a maximum harmonic amplitude that exceeds the amplitude values ​​of other harmonics by more than 6dB, then directly confirm the propeller blade frequency and the number of blades.

[0031] S55: If the conditions in S54 are not met, then score the blade number and blade frequency estimation parameters of the DEMON spectrum harmonic vector group based on the amplitude relationship characteristics of the DEMON spectrum harmonic group.

[0032] S56: Take the highest score from S55 to determine the target propeller blade frequency and number of blades.

[0033] Due to the adoption of the above technical solutions, the beneficial effects of this invention are:

[0034] 1. This invention proposes a novel method for extracting the DEMON spectrum of underwater acoustic target radiated noise and estimating the number of propeller blades, combining wavelet packet decomposition algorithm and high-order spectral analysis techniques. Wavelet packet decomposition is a modern time-frequency analysis and processing method capable of effectively processing various non-stationary random signals. Through wavelet packet transform, the acquired signal can be decomposed into multiple two-dimensional parameters (time, position) and frequencies, realizing the feature decomposition of the signal at different frequency bands and times. Wavelet packet decomposition is actually an improvement on wavelet decomposition, decomposing the high-frequency and low-frequency components of the signal. It is more refined and comprehensive than wavelet transform, better reflecting the full frequency characteristics of the signal. The feature vector can adaptively select the frequency band, exhibiting time-frequency localization characteristics. The features correspond to the spectrum, thereby improving the time-frequency resolution of the signal.

[0035] 2. This invention pre-decomposes the original time-domain signal using an empirical mode decomposition algorithm, adaptively selecting the optimal wavelet packet decomposition level for the target signal, thus avoiding the blindness of a fixed decomposition level in different signal decompositions. Furthermore, by calculating the 1 (1 / 2) spectrum of the envelope of each wavelet packet component instead of the traditional spectrum, the signal-to-noise ratio of the DEMON spectral harmonics can be improved, facilitating subsequent propeller parameter estimation. Finally, by fusing the higher-order spectral components of wavelet packets from different frequency bands, the noise floor can be further reduced, improving the signal-to-noise ratio of the DEMON spectral harmonics.

[0036] 3. The underwater acoustic target propeller blade number estimation and scoring method constructed by the present invention integrates the amplitude characteristics of the first 8 harmonics of propellers with different blade numbers, avoiding the problem of blade frequency judgment error caused by the line spectrum amplitude at the blade frequency not being the strongest due to interference factors such as environmental noise. This makes the automatic estimation result of propeller blade number more accurate, and while ensuring a high degree of automation, it also solves the problem of large estimation error.

[0037] In summary, the method of this invention is based on wavelet packet decomposition and 1(1 / 2) spectrum analysis algorithm, and by setting automatic estimation criteria for propeller blade number, it achieves the tasks of DMEON spectrum feature extraction with more stable background noise and higher harmonic signal-to-noise ratio, as well as automatic estimation of propeller shaft frequency, blade frequency and blade number, with higher reliability and automation level. Attached Figure Description

[0038] Figure 1 The background technology describes the general calculation process of the traditional DEMON spectrum;

[0039] Figure 2 This is a schematic diagram of the decomposition of the wavelet packet tree in a specific embodiment of the present invention;

[0040] Figure 3 This is a schematic diagram illustrating the principle of an automated DEMON spectral feature extraction method based on multi-subband fusion in a specific embodiment of the present invention.

[0041] Figure 4 This is a schematic diagram of the automated propeller parameter estimation process in a specific embodiment of the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings.

[0043] An automated method for extracting DEMON spectral features using multi-subband fusion is proposed. First, an optimization algorithm is used to obtain the optimal number of decomposition layers. Then, wavelet packet decomposition is used to adaptively decompose the radiated noise of the underwater acoustic target, selecting the effective components and calculating the square values ​​of each order component. Next, the square values ​​of each effective component are passed through a low-pass filter, and the 1 (1 / 2) spectrum of each order signal is calculated. Gaussian white noise is suppressed by accumulating higher-order spectra, resulting in higher-order spectra with a higher signal-to-noise ratio. These higher-order spectra are then fused to reduce spectral noise floor fluctuations and further enhance the harmonic line spectrum. Finally, the harmonic group line spectrum features are obtained using a line spectrum detection algorithm, and the target blade number parameter is automatically estimated based on empirical formulas. Specifically, the method includes the following steps:

[0044] S1: Optimal wavelet packet decomposition level selection: Based on the empirical mode decomposition algorithm, the original signal is pre-decomposed to determine the optimal wavelet packet decomposition level, and the original signal is decomposed into wavelet packets according to the optimal level to obtain wavelet packet components in different frequency bands.

[0045] S2: Wavelet packet component selection. The spectrum of the wavelet packet components obtained in S1 based on the optimal decomposition level is calculated, and wavelet packet components without spectrum contamination are selected.

[0046] S3: Square filtering and higher-order spectrum calculation. The wavelet packet components selected in S2 are squared and then passed through a low-pass filter to obtain the envelope signal, and 1 (1 / 2) higher-order spectrum calculation is performed.

[0047] S4: Higher-order DEMON spectrum fusion, which fuses the 1 (1 / 2) higher-order spectra of the envelopes of each wavelet packet component calculated in S3;

[0048] S5: Feature extraction and parameter estimation. Harmonic groups are extracted from the DEMON high-order spectrum obtained in S4, and the shaft frequency, blade frequency and number of blades of the target propeller are estimated based on empirical formulas.

[0049] Step S1 includes the following steps:

[0050] S11: Based on the empirical mode decomposition algorithm, pre-decompose the original time domain signal until the decomposition reaches the termination condition, and record the decomposition level N;

[0051] S12: Determine N as the optimal wavelet packet decomposition level, and perform wavelet packet decomposition on the original signal to obtain wavelet packet components in different frequency bands.

[0052] Step S2 includes the following steps:

[0053] S21: Perform FFT calculation on the wavelet packet components obtained from the optimal decomposition level N in S1;

[0054] S22: Calculate the -3dB cutoff frequency for the spectrum of each wavelet packet component obtained in step S21 from low to high. When the lower cutoff frequency is higher than 100Hz, all wavelet packet components below that order are discarded.

[0055] Step S3 includes the following steps:

[0056] S31: Calculate the square of each effective component obtained in step S2;

[0057] S32: Perform low-pass filtering on the signals output in step S31 to obtain the signal quantity containing the target propeller envelope modulation information;

[0058] S33: For the signal output in step S32, calculate its 1 (1 / 2) higher-order spectrum, and normalize the energy of the higher-order spectrum of each component. The formula for calculating the higher-order spectrum is:

[0059]

[0060] Where x(t) is a time-domain signal sequence.

[0061] Step S4 includes the following steps:

[0062] S41: The 1 (1 / 2) spectra of each component obtained in step S3 are summed according to the corresponding frequency points and the average value is taken as the DEMON higher-order spectrum of the target radiated noise signal.

[0063] Step S5 includes the following steps:

[0064] S51: Design a bidirectional α smoothing filter and smooth the DEMON high-order spectrum obtained in step S4 to remove the trend term;

[0065] S52: Calculation of axis frequencies of DEMON spectrum harmonic groups based on the greatest common divisor method;

[0066] S53: Based on the multiple relationship between the shaft frequency and its harmonic group frequencies, solve for the frequencies and amplitudes corresponding to the 2nd to 8th harmonics in the DEMON higher-order spectrum;

[0067] S54: Based on the amplitude relationship characteristics of the harmonic group in the DEMON spectrum, evaluate the frequency and amplitude combination vector of the harmonic group obtained in S53, and calculate the target number of propeller blades.

[0068] Here is a more specific example:

[0069] An automated method for extracting DEMON spectrum features by multi-subband fusion is proposed. The method first performs wavelet packet decomposition on the original time-domain signal, then performs square detection and low-pass filtering on the wavelet packet components of each order to obtain the signal envelope components of different frequency bands of each order, then calculates the 1 (1 / 2) spectrum of each order component, and adds and fuses them according to the corresponding frequency points to obtain the high-order DEMON spectrum.

[0070] Reference Figure 3 The specific steps of this method are as follows:

[0071] S1, Optimal selection of wavelet packet decomposition level:

[0072] Wavelet packets have three parameters: position, scale (as with general wavelet decomposition), and frequency. For a given orthogonal wavelet function, a set of wavelet packet bases can be generated, each providing a specific signal encoding method. It completely preserves the full energy of the signal and can fully reconstruct all features of the original signal. Wavelet packets can be used to decompose signals in many different ways, but for a given signal, wavelet function, number of decomposition levels, and entropy criterion, there is an optimal wavelet packet decomposition method. See [link to relevant documentation]. Figure 2 The specific method for this step is as follows:

[0073] S11: Decompose the original time-domain signal based on the Empirical Mode Decomposition (EMD) algorithm until the decomposition is completed, and determine the optimal number of decomposition layers N;

[0074] S12: Perform wavelet packet decomposition on the original time-domain signal according to the optimal decomposition level N determined in S11 to obtain wavelet packet components of different frequency bands.

[0075] S2, Wavelet packet component selection:

[0076] After obtaining the optimal wavelet packet decomposition level of the original time-domain signal, the low-frequency wavelet packet components need to be discarded because the inherent line spectra of numerous underwater acoustic targets in the low-frequency band can interfere with the final DEMON spectrum calculation results. This yields a purer envelope signal containing propeller modulation information. The specific method for this step is as follows:

[0077] S21: Perform an L-point FFT calculation on the wavelet packet components obtained from the optimal decomposition level N in S1;

[0078] S22: Calculate the -3dB cutoff frequency for the spectrum of each wavelet packet component obtained in step S21 from low to high. When the lower cutoff frequency of the i-th order component is higher than 100Hz, all wavelet packet components below that order are discarded.

[0079] S3, squared filtering and higher-order spectrum calculation; the specific method for this step is as follows:

[0080] S31: If the Kth order effective component is selected in step S2, then the squares of each component are calculated.

[0081] S32: Set the cutoff frequency of the FIR digital low-pass filter to 100Hz and the filter order to 2048. Perform low-pass filtering on the signal obtained in S31 to obtain the envelope signal b(n) of each wavelet packet component.

[0082] S33: For the envelope signal b(n) output in step S32, calculate its 1 (1 / 2) higher-order spectrum, and normalize the energy of the higher-order spectrum of each component. The formula for calculating the higher-order spectrum is:

[0083]

[0084] Where x(t) is a time-domain signal sequence, and the formula for the energy normalization method is:

[0085]

[0086] S4, High-order DEMON spectrum fusion; the specific method for this step is as follows:

[0087] S41: For each order 1 (1 / 2) spectrum obtained in step S3, sum them according to the corresponding frequency points and take the average value to obtain the DEMON higher-order spectrum of the target radiated noise signal. The calculation formula is as follows:

[0088]

[0089] Where i represents the i-th frequency point, N represents the wavelet packet decomposition level of the time-domain signal, j represents the first j-th order wavelet packet components to be discarded, and D k The 1(1 / 2) spectrum represents the envelope of the k-th effective wavelet packet component.

[0090] S5, Feature Extraction and Parameter Estimation:

[0091] After processing steps S1 to S4, a DEMON spectrum with stable background noise and high signal-to-noise ratio can be obtained. In order to further estimate the physical parameters of the underwater acoustic target, such as the propeller shaft frequency, blade frequency, and number of blades, from the DEMON spectrum, and thus identify the attributes of the underwater acoustic target and assist in military decision-making, the harmonic line spectrum of the DEMON spectrum needs to be detected, extracted, and analyzed.

[0092] The velocity spectrum of propeller blades typically has the largest amplitude at the blade frequency. Therefore, in an ideal DEMON spectrum, the frequency of the harmonic line spectrum with the largest amplitude is the propeller blade frequency. However, interference from factors such as environmental noise can cause the amplitude at the blade frequency in the DEMON spectrum to not be the largest, leading to incorrect estimation of propeller parameters. Based on the amplitude characteristics of the DEMON spectrum harmonic group (see Table 1), this method designs a propeller blade number estimation criterion based on the characteristics of the DEMON spectrum harmonic group.

[0093] Table 1. Characteristics of the amplitude relationship of DEMON spectrum harmonic groups

[0094]

[0095] Where P(i) represents the amplitude of the i-th harmonic.

[0096] The specific method is as follows:

[0097] S51: Design a smoothing filter with parameters α = 0.5 and n = 5, where α is the filter weight coefficient and n is the prediction point length. Apply a smoothing filter to the DEMON spectrum obtained in step S4 to obtain a smoothed DEMON spectrum after removing the background term.

[0098] S52: Solving the propeller shaft frequency f1 in the smoothed DEMON spectrum in S51 based on the greatest common divisor method;

[0099] S53: Based on the shaft frequency f1 and the harmonic group multiples, solve for the harmonic frequencies and amplitudes of the 2nd to 8th order DEMON spectrum. Specifically, due to the FFT digital operation, the calculated shaft frequency f1 differs from the actual shaft frequency. There is always a frequency offset Δf, which causes the harmonic error to have a frequency offset nΔf, where n is the harmonic order. Therefore, the frequency f of the i-th harmonic is calculated. iThe search for extreme points in the vicinity is set to [f]. i -k×df,f i -k×df], where df is the frequency difference between two adjacent frequency points in the DEMON spectrum, k∈Z, to obtain the harmonic group vector v=[(f1,P1),(f2,P2),…,(f8,P8)] of the smoothed DEMON spectrum, where (f i ,P i ) represents the frequency and amplitude combination of the i-th harmonic;

[0100] Next, adopt the following... Figure 4 The propeller blade number is estimated in the manner shown:

[0101] S54: Using the harmonic group vectors of the first 8 smooth DEMON spectra obtained in S53 as input, firstly search for the maximum value of the 3rd to 7th harmonic vectors. If the maximum amplitude of the i-th harmonic is more than 6dB higher than the amplitude of other harmonics, then take the target propeller blade frequency as the i-th harmonic frequency and the number of propeller blades as i.

[0102] S55: If the 3rd to 7th order harmonic vectors do not have harmonic amplitudes exceeding 6dB above other harmonic amplitudes, then the target propeller blade number is scored according to the harmonic amplitude relationship characteristic table in Table 1. The scoring rules for the target propeller blade number are as follows:

[0103]

[0104] Where i represents the i-th harmonic amplitude characteristic relationship, m represents the number of inequalities that hold true in the i-th harmonic amplitude characteristic relationship, and n represents the total number of inequalities in the i-th harmonic amplitude characteristic relationship.

[0105] S56: Based on the scores of this group of harmonic vectors in different blade numbers in S55, the i-th group with the highest score is selected to determine the target propeller blade frequency f. i And the number of propeller blades i.

[0106] In summary, this invention addresses the problem of automated extraction of DEMON spectral features and propeller blade number estimation for unknown target radiated noise in unmanned platforms. First, it employs a parameter-optimized wavelet packet decomposition algorithm for adaptive decomposition of the signal under analysis. The number of decomposition levels is pre-determined by the Empirical Mode Decomposition (EMD) algorithm, resolving the blindness and mismatch issues associated with fixed decomposition levels in different signal decomposition processes, and achieving adaptive selection of decomposition parameters. Then, a higher-order spectral analysis method is used to calculate the DEMON spectrum for each wavelet packet component, further suppressing marine environmental noise and improving the signal-to-noise ratio (SNR) of the DEMON spectrum harmonic line spectrum. Finally, wavelet packet components with significant modulation information are selected for DEMON spectrum calculation and fusion, avoiding the need for manual selection of bandpass filter cutoff frequencies during DEMON spectrum calculation. The fused DEMON spectrum of this invention has a higher SNR than the harmonic line spectrum extracted by traditional methods, facilitating subsequent harmonic group extraction and propeller blade number estimation.

[0107] The above description is merely a specific embodiment of the present invention. Any feature disclosed in this specification may be replaced by other equivalent or similar features unless otherwise specified. All disclosed features, or steps in all methods or processes, may be combined in any way except for mutually exclusive features and / or steps.

Claims

1. An automated method for extracting DEMON spectral features through multi-subband fusion, characterized in that, Includes the following steps: S1, Optimal selection of wavelet packet decomposition level: Based on the empirical mode decomposition algorithm, the original ship radiated noise signal is pre-decomposed to determine the optimal wavelet packet decomposition level, and the original ship radiated noise signal is decomposed into wavelet packets according to the optimal wavelet packet decomposition level to obtain wavelet packet components in different frequency bands. S2, Wavelet packet component selection: Calculate the spectrum of the wavelet packet components obtained in step S1, and select wavelet packet components without spectrum contamination. S3, Squared Filtering and Higher-Order Spectrum Calculation: The wavelet packet components selected in step S2 are squared, and then the envelope signal is obtained through a low-pass filter and 1 (1 / 2) higher-order spectrum calculation is performed. S4, Higher-order DEMON spectrum fusion: The 1 (1 / 2) higher-order spectra of the envelopes of each wavelet packet component calculated in step S3 are fused to obtain the DEMON higher-order spectrum; the specific method is as follows: The 1st (1 / 2) spectra obtained in step S3 are summed at their corresponding frequency points and the average is taken as the DEMON higher-order spectrum of the target radiated noise signal. The calculation formula is as follows: in, Indicates the first One frequency point, The wavelet packet decomposition level represents the number of wavelet packet decomposition layers of the time-domain signal. The former represents abandonment wavelet packet components, Representing the The 1(1 / 2) spectrum of the envelope of the effective wavelet packet component; S5, Feature Extraction and Parameter Estimation: Harmonic groups are extracted from the DEMON high-order spectrum obtained in step S4, and the shaft frequency, blade frequency and number of blades of the target ship propeller are estimated based on empirical formulas to complete the DEMON spectrum feature extraction.

2. The automated extraction method for DEMON spectral features by multi-subband fusion according to claim 1, characterized in that, The specific method of step S1 is as follows: S11: Based on the empirical mode decomposition algorithm, the original ship radiated noise signal is pre-decomposed until the decomposition reaches the termination condition, and the decomposition level N is recorded. S12: N is taken as the optimal wavelet packet decomposition level, and the original ship radiated noise signal is decomposed into wavelet packets to obtain wavelet packet components in different frequency bands.

3. The automated extraction method for DEMON spectral features by multi-subband fusion according to claim 2, characterized in that, The specific method of step S2 is as follows: S21: Perform FFT calculation on the wavelet packet components obtained from step S1 to obtain the spectrum of each order of wavelet packet components; S22: Calculate the -3dB cutoff frequency for the spectrum of wavelet packet components of each order from low to high. When the lower cutoff frequency is higher than 100Hz, all wavelet packet components of lower order are discarded.

4. The ship noise recognition method based on paired coding networks and contrastive learning according to claim 3, characterized in that, The specific method of step S3 is as follows: S31: Calculate the square of each wavelet packet component obtained in step S2; S32: Perform low-pass filtering on the signals output in step S31 to obtain the signal quantity containing the envelope modulation information of the target ship's propeller. S33: For the signal output in step S32, calculate its 1 (1 / 2) higher-order spectrum, and normalize the energy of the higher-order spectrum of each component. The formula for calculating the higher-order spectrum is: in, It is a time-domain signal sequence.

5. The ship noise recognition method based on paired coding networks and contrastive learning according to claim 1, characterized in that, The specific method of step S5 is as follows: S51: Designed for bidirectional operation A smoothing filter is applied to the DEMON higher-order spectrum obtained in step S4 to remove the trend term; S52: Calculation of axis frequencies of DEMON spectrum harmonic groups based on the greatest common divisor method; S53: Based on the multiple relationship between the shaft frequency and its harmonic group frequency, solve for the frequencies and amplitudes corresponding to the 2nd to 8th harmonics in the DEMON higher-order spectrum, and form the harmonic group vector of the DEMON spectrum from these frequencies and amplitudes. S54: Based on the harmonic group vector obtained in S53, search for the amplitude values ​​of the 3rd to 7th harmonics. If there is a maximum harmonic amplitude that exceeds the amplitude values ​​of other harmonics by more than 6dB, then directly confirm the propeller blade frequency and the number of blades. S55: If the conditions in S54 are not met, then score the blade number and blade frequency estimation parameters of the DEMON spectrum harmonic vector group based on the amplitude relationship characteristics of the DEMON spectrum harmonic group. S56: Take the highest score from S55 to determine the target propeller blade frequency and number of blades.

Images