Modulation spectrum adaptive feature purification method based on feature energy entropy reconstruction

Abstract

The modulation spectrum adaptive feature purification method based on feature energy entropy reconstruction belongs to the field of underwater acoustic target recognition. The method is used for solving the problem that the existing modulation spectrum feature extraction method cannot solve the influence of harmonic interference on the final estimation result. The method comprises the following processes: in step one, the cavitation noise in the ship radiated noise is demodulated through a plurality of preset demodulation frequency bands, a plurality of demoded signals are obtained, and the optimal demodulation frequency band is selected from the demodulation frequency bands corresponding to the demoded signals by adopting the minimum entropy criterion; in step two, the demoded signal obtained under the optimal demodulation frequency band is decomposed through the VMD algorithm, and the signals of the intrinsic mode functions of all orders are obtained; in step three, the feature energy entropy of the signals of the intrinsic mode functions of all orders is calculated, the feature energy entropy is reconstructed by using the FEER algorithm, and the reconstructed signal is obtained; in step four, the power spectrum analysis is carried out on the reconstructed signal, and the purified modulation spectrum is obtained. The method is mainly used for purifying the modulation spectrum.

Description

Technical Field

[0001] This invention belongs to the field of underwater acoustic target recognition, specifically involving a feature purification method based on variational mode decomposition and feature energy entropy reconstruction. Background Technology

[0002] The classification of ship radiated noise has been a major focus of attention in recent decades. The complex underwater noise sources, the reverberation effects of the ocean surface and seabed, and the lack of prior information about the target all contribute to the difficulty of solving the classification problem.

[0003] Existing technologies systematically discuss the theoretical basis of cavitation noise and point out that the stability of ship radiated noise has a significant impact on the demodulation effect of cavitation noise. Using wavelet packet decomposition, the DEMON spectra of different subspaces were compared, their differences were identified, and the most stable one was selected. However, wavelet packet decomposition automatically introduces the influence of the frequency band containing higher-order harmonics, and these higher-order harmonic characteristics are not inherent features of the modulation spectrum.

[0004] Currently, modulation spectrum feature extraction techniques include interference suppression gates, greatest common divisor, maximum likelihood estimators, and expert systems. However, these techniques have not addressed the impact of harmonic interference on the final estimation results, and these issues urgently need to be resolved. Summary of the Invention

[0005] The purpose of this invention is to address the problem that existing modulation spectrum feature extraction methods have failed to resolve the impact of harmonic interference on the final estimation results. This invention provides an adaptive feature purification method for modulation spectrum based on feature energy entropy reconstruction.

[0006] An adaptive feature cleanup method for modulation spectrum based on feature energy entropy reconstruction includes the following steps:

[0007] Step 1: Demodulate the cavitation noise in the ship's radiated noise using multiple preset demodulation frequency bands to obtain multiple demodulated signals. Select the optimal demodulation frequency band from the demodulation frequency bands corresponding to each demodulated signal using the minimum entropy criterion.

[0008] Step 2: Decompose the demodulated signal obtained under the optimal demodulation frequency band using the VMD algorithm to obtain the intrinsic mode function signals of each order;

[0009] Step 3: Calculate the characteristic energy entropy of each order of intrinsic mode function signal, and reconstruct all characteristic energy entropies using the FEER algorithm to obtain the reconstructed signal;

[0010] Step 4: Perform power spectrum analysis on the reconstructed signal to obtain the purified modulation spectrum.

[0011] Preferably, in step one, the method for selecting the optimal demodulation frequency band is as follows:

[0012] Step 1: By analyzing the energy probability distribution at each modulation frequency in the modulation spectrum of each demodulated signal, calculate the modulation spectrum information entropy H of the demodulated signal.

[0013] Where, p i This is the energy probability distribution at the i-th modulation frequency in the demodulated signal; i = 1, 2, ..., N;

[0014] Step 12: Select the minimum value from all modulation spectrum information entropy H. The demodulation frequency band corresponding to the minimum modulation spectrum information entropy H is taken as the optimal demodulation frequency band.

[0015] Preferably, 0≤p i ≤1.

[0016] Preferably, in step three,

[0017] When the demodulated signal includes one fundamental frequency modulation component, the characteristic energy entropy FEE of each eigenmode function signal is... k The expression is:

[0018] Among them, FEE k The k-th order intrinsic mode function signal u k Characteristic energy entropy; k = 1, 2, ..., M; The k-th order eigenmode function signal u k The energy at the m-th modulation frequency in the modulation spectrum; m = 1, 2, ..., h; E k The sum of the energy at all modulation frequencies of the k-th order intrinsic mode function signal;

[0019] When the demodulated signal includes n fundamental frequency modulation components, the characteristic energy entropy FEE of each intrinsic mode function signal is... k The expression is:

[0020]

[0021] Among them, FEE k The k-th order intrinsic mode function signal u k Characteristic energy entropy; k = 1, 2, ..., M; The k-th order eigenmode function signal u k In the modulation spectrum of f t f is the energy at the m-th modulation frequency of the fundamental frequency; t Let t be the t-th fundamental frequency modulation component; m = 1, 2, ..., h; t = 1, 2, ..., n; E k Let be the sum of the energy at all modulation frequencies of the k-th order intrinsic mode function signal.

[0022] Preferably, in step three, the FEER algorithm is used to reconstruct the energy entropy of all features to obtain the reconstructed signal. The implementation methods include:

[0023] Step 3. Calculate the k-th order eigenmode function signal u. k Feature Energy Entropy FEE k The proportion of the entire mode β k ;

[0024]

[0025] Where k = 1, 2, ..., M;

[0026] Step 3.2: Calculate the k-th order intrinsic mode function signal u. k Normalized reconstructed weights

[0027]

[0028] Where β is the reconstructed weight coefficient vector, and β = [β1, β2, ..., β] M ];

[0029] Step 3: Calculate the reconstructed signal u pur ;in,

[0030] Preferably, in step four, a Fourier transform is performed on the reconstructed signal to achieve power spectrum analysis of the reconstructed signal.

[0031] The benefits of this invention are:

[0032] (1) More accurate demodulation frequency band:

[0033] This invention utilizes the minimum entropy criterion to select the optimal demodulation frequency band from the demodulation frequency bands corresponding to each demodulated signal. The demodulation frequency band corresponding to the minimum spectral information entropy has a better spectral structure and is more suitable for modulation spectrum analysis, providing a reliable foundation for further obtaining the accuracy of the reconstructed signal.

[0034] (2) Capable of feature independence:

[0035] VMD algorithm decomposition makes the mixed features of complex signals independent, which makes feature processing and feature selection more convenient.

[0036] (3) The modulation spectrum structure is simpler and clearer:

[0037] In the modulation spectrum structure obtained by combining the VMD algorithm and the FEER algorithm, higher-order line spectra are suppressed, and the inherent modulation characteristics of the signal are clearer overall, improving the utilization rate of the purified modulation spectrum features. Specifically, this invention fully explores the feature content of each eigenmode function signal obtained by decomposition through the VMD algorithm, and constructs a linear space based on feature energy entropy based on the PEER algorithm, providing a theoretical basis for the reconstruction process. The independent features include inherent features and higher-order harmonic features, where the energy of inherent features is always present, while higher-order harmonic features are random features excited by blade resonance. Therefore, the content of inherent modulation features in the eigenmode function space is much higher than the content of random features. Thus, by reconstructing the linear space, the spatial proportion of inherent features can be increased while the spatial proportion of random features can be reduced. Therefore, this invention retains the main modulation quantity while suppressing the influence of higher-order harmonics on the modulation spectrum structure. As a result, the modulation spectrum structure processed by this invention is simpler and clearer. Attached Figure Description

[0038] Figure 1 This is a schematic diagram illustrating the principle of the modulation spectrum adaptive feature purification method based on feature energy entropy reconstruction described in this invention;

[0039] Figure 2 This is a schematic diagram illustrating the principle of obtaining the reconstructed signal.

[0040] Figure 3 The diagram shows a comparison of modulation spectra obtained by the traditional DEMON spectrum analysis algorithm and the adaptive feature purification method for modulation spectra based on feature energy entropy reconstruction of the present invention; wherein, (a) the modulation spectrum obtained by the traditional DEMON spectrum analysis algorithm; and (b) the modulation spectrum obtained by the adaptive feature purification method for modulation spectra based on feature energy entropy reconstruction of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0043] Example 1:

[0044] The following is combined with Figure 1 This embodiment 1 describes an adaptive feature cleanup method for modulation spectrum based on feature energy entropy reconstruction. The method includes the following steps:

[0045] Step 1: Demodulate the cavitation noise in the ship's radiated noise using multiple preset demodulation frequency bands to obtain multiple demodulated signals. Select the optimal demodulation frequency band from the demodulation frequency bands corresponding to each demodulated signal using the minimum entropy criterion.

[0046] Step 2: Decompose the demodulated signal obtained under the optimal demodulation frequency band using the VMD algorithm to obtain the intrinsic mode function signals of each order;

[0047] Step 3: Calculate the characteristic energy entropy of each order of intrinsic mode function signal, and reconstruct all characteristic energy entropies using the FEER algorithm to obtain the reconstructed signal;

[0048] Step 4: Perform power spectrum analysis on the reconstructed signal to obtain the purified modulation spectrum.

[0049] In this embodiment, the ship's radiated noise includes two parts: cavitation noise and environmental noise. This invention is mainly used to purify and extract the modulation spectrum in the cavitation noise, extract the purified modulation spectrum from the cavitation noise, and improve the utilization rate of the purified modulation spectrum features. The purified modulation spectrum contains more prominent ship propeller modulation features, including important information such as the number of blades of the ship's propeller and its speed. By analyzing the cavitation noise, the ship's propeller feature information can be reflected.

[0050] The reconstructed space is divided into an inherent modulation space, a harmonic space, and a noise space. In the signal reconstruction process, this invention only considers the inherent modulation space of the reconstructed signal (the inherent modulation characteristic space in the linear space). The reconstruction result of this invention suppresses the harmonic and noise spaces, thereby suppressing harmonic interference in cavitation noise. A more accurate modulation spectrum is obtained from the reconstructed signal, which, as an observation signal, more accurately reflects the ship's motion information.

[0051] The PEER algorithm fully exploits the feature content of each eigenmode function signal obtained through VMD decomposition, constructing a linear space based on feature energy entropy, providing a theoretical basis for the reconstruction process. The independent features include intrinsic features and higher-order harmonic features. The energy of intrinsic features is always present, while higher-order harmonic features are random features excited by blade resonance. Therefore, the content of intrinsic modulation features in the eigenmode function space is much higher than that of random features. Thus, reconstructing the linear space can increase the spatial proportion of intrinsic features while reducing the spatial proportion of random features. This invention retains the main modulation quantity while suppressing the influence of higher-order harmonics on the modulation spectrum structure. The results show that the modulation spectrum structure processed by this invention is simpler and clearer.

[0052] In step two, the demodulated signal obtained under the optimal demodulation frequency band is decomposed using the VMD algorithm to obtain the eigenmode function signals of each order; at the same time, the FEER algorithm is used to reconstruct all the characteristic energy entropies to obtain the reconstructed signal. The FEER algorithm reconstructs the complete observation signal using the least squares method.

[0053] The optimal demodulation frequency band selection process of this invention is based on the principle of minimum entropy; the smaller the entropy value, the more stable the system; the larger the entropy value, the worse the system stability. By analyzing the probability distribution of energy (i.e., frequency coefficients) at each modulation frequency in the modulation spectrum, the spectral information entropy of each sub-band is calculated, thereby obtaining a better demodulation frequency band.

[0054] Specifically, in step one, the method for selecting the optimal demodulation frequency band is as follows:

[0055] Step 1: By analyzing the energy probability distribution at each modulation frequency in the modulation spectrum of each demodulated signal, calculate the modulation spectrum information entropy H of the demodulated signal.

[0056] Where, p i This is the energy probability distribution at the i-th modulation frequency in the demodulated signal; i = 1, 2, ..., N; 0 ≤ p i ≤1;

[0057] Step 12: Select the minimum value from all modulation spectrum information entropy H. The demodulation frequency band corresponding to the minimum modulation spectrum information entropy H is taken as the optimal demodulation frequency band.

[0058] In this preferred method, the smaller the entropy value, the more stable the system; the larger the entropy value, the worse the system stability. By analyzing the energy probability components at each modulation frequency, the spectral information entropy of each sub-band is calculated, thus obtaining a better demodulation frequency band. For the modulation spectrum, spectral stability is related to its spectral structure: a modulation spectrum with a chaotic spectral structure has a larger spectral information entropy; a modulation spectrum with only inherent modulation characteristics has a smaller spectral information entropy.

[0059] Furthermore, in step three, when the demodulated signal includes one fundamental frequency modulation component, the characteristic energy entropy FEE of each eigenmode function signal is... k The expression is:

[0060] Among them, FEE k The k-th order intrinsic mode function signal u k Characteristic energy entropy; k = 1, 2, ..., M; The k-th order eigenmode function signal u k The energy at the m-th modulation frequency in the modulation spectrum; m = 1, 2, ..., h; E k The sum of the energy at all modulation frequencies of the k-th order intrinsic mode function signal;

[0061] When the demodulated signal includes n fundamental frequency modulation components, the characteristic energy entropy FEE of each intrinsic mode function signal is... k The expression is:

[0062]

[0063] Among them, FEE k The k-th order intrinsic mode function signal u k Characteristic energy entropy; k = 1, 2, ..., M; The k-th order eigenmode function signal u k In the modulation spectrum of f t f is the energy at the m-th modulation frequency of the fundamental frequency; t Let t be the t-th fundamental frequency modulation component; m = 1, 2, ..., h; t = 1, 2, ..., n; E k Let be the sum of the energy at all modulation frequencies of the k-th order intrinsic mode function signal.

[0064] Furthermore, see Figure 2 In step three, the FEER algorithm is used to reconstruct the energy entropy of all features, and the reconstructed signal is obtained through the following methods:

[0065] Step 3. Calculate the k-th order eigenmode function signal u. k Feature Energy Entropy FEE k The proportion of the entire mode β k ;

[0066]

[0067] Where k = 1, 2, ..., M;

[0068] Step 3.2: Calculate the k-th order intrinsic mode function signal u. k Normalized reconstructed weights

[0069]

[0070] Where β is the reconstructed weight coefficient vector, and β = [β1, β2, ..., β] M ];

[0071] Step 3: Calculate the reconstructed signal u pur ;in,

[0072] In this preferred embodiment, the present invention first calculates the reconstruction weight coefficients corresponding to each order of intrinsic mode function signals using the FEER algorithm during the final reconstruction process. Specifically, the calculation step involves calculating the ratio of the corresponding frequency point in the power spectrum of each order of intrinsic mode function signal to the analysis frequency band, i.e., FEE. k Then, through The reconstructed weight coefficient vector β is obtained. The reconstructed space is divided into an intrinsic modulation space, a harmonic space, and a noise space. The final reconstruction result is the suppression of the latter two, i.e., suppression of the harmonic and noise spaces.

[0073] Furthermore, in step four, a Fourier transform is performed on the reconstructed signal to achieve power spectrum analysis of the reconstructed signal.

[0074] Verification experiment:

[0075] Figure 3 The modulation spectrum obtained by the traditional DEMON spectrum analysis algorithm and the modulation spectrum adaptive feature purification method based on feature energy entropy reconstruction of the present invention are compared to verify the technical effect of the present invention. Figure 3 The modulation spectrum results obtained by the traditional DEMON spectral analysis algorithm and the modulation spectrum results of the present invention are presented respectively. Figure 3 (a) yields three principal characteristic quantities and three pseudo-characteristic quantities. The principal characteristic quantities include the shaft frequency and its 3rd and 4th modulation components, located at 2.1Hz, 6.2Hz, and 8.4Hz respectively, but the 2nd modulation component is missing. The pseudo-characteristic quantities include the 8th, 12th, and 20th harmonic components, at 16.6Hz, 25Hz, and 41.6Hz respectively. In contrast, Figure 3 (b) After processing using the method of this invention, the third modulation component is still clearly visible, but its energy percentage is significantly lower than that of the traditional DEMON spectral analysis algorithm. The method of this invention significantly outperforms the classical algorithm in suppressing spurious features; the 12th and 20th harmonic components are almost completely suppressed, and the final calculated spectral entropy is 7.36, lower than the 0.46 of the traditional DEMON spectral analysis algorithm. Therefore, the method of this invention has a stronger advantage in the modulation spectrum analysis of ship radiated noise and can suppress harmonic interference.

[0076] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.

Claims

1. A modulation spectrum adaptive feature purification method based on feature energy entropy reconstruction, characterized in that, The method comprises the following steps: Step one, demodulate the cavitation noise in the ship radiated noise by a plurality of preset demodulation frequency bands to obtain a plurality of demodulated signals, and select the optimal demodulation frequency band from the demodulation frequency bands corresponding to the demodulated signals by using the minimum entropy criterion; The implementation of selecting the optimal demodulation frequency band is as follows: Step one, by analyzing the energy probability distribution at each modulation frequency in the modulation spectrum of each demodulated signal, the modulation spectrum information entropy of the demodulated signal is calculated ​​ wherein is the probability distribution of energy at the mth modulation frequency in the demodulated signal; is the probability distribution of energy at the mth modulation frequency in the demodulated signal; is the probability distribution of energy at the mth modulation frequency in the demodulated signal; Step one two, select the minimum value from all modulation spectrum information entropy The minimum modulation spectrum information entropy Corresponding demodulation frequency band as the optimal demodulation frequency band; Step two, decompose the demodulated signal obtained under the optimal demodulation frequency band by the VMD algorithm to obtain a signal of each order intrinsic mode function; Step three, calculate the characteristic energy entropy of the signal of each order intrinsic mode function, and reconstruct all the characteristic energy entropies by using the FEER algorithm to obtain a reconstructed signal; Step four, perform power spectrum analysis on the reconstructed signal to obtain a purified modulation spectrum.

2. The modulation spectrum adaptive feature purifying method based on feature energy entropy reconstruction according to claim 1, characterized in that, 。 3. The modulation spectrum adaptive feature purifying method based on feature energy entropy reconstruction according to claim 1, characterized in that, In step three, When the demodulated signal includes 1 base frequency modulation component, the characteristic energy entropy of each order intrinsic mode function signal is The expression is: wherein, is the characteristic energy entropy of the th order intrinsic mode function signal ; ; denotes the energy at the th order intrinsic mode function signal ; th modulation frequency in the modulation spectrum of the ; th order intrinsic mode function signal ; When the demodulated signal includes a base frequency modulation component, the expression of the characteristic energy entropy of each eigenmode function signal is: ​ in, For the first eigenmode function signal The characteristic energy entropy; ; Indicates the first eigenmode function signal In the modulation spectrum The first of the fundamental frequency Energy at each modulation frequency; For the first One fundamental frequency modulation component; ; ; For the first The sum of the energy at all modulation frequencies of the eigenmode function signal.

4. The modulation spectrum adaptive feature purifying method based on feature energy entropy reconstruction according to claim 1, characterized in that, In step three, the implementation of reconstructing the signal by using the FEER algorithm to all the characteristic energy entropies comprises: Step three i. Calculate the characteristic energy entropy of the first order intrinsic modal function signal ​​​ wherein ; Step three two, compute the first order modal function signal normalized reconstruction weight ; wherein is a reconstructed weight coefficient vector, and ; Step three three, calculating the reconstructed signal ; wherein .

5. The modulation spectrum adaptive feature purifying method based on feature energy entropy reconstruction according to claim 1, characterized in that, In step four, Fourier transform is performed on the reconstructed signal to realize power spectrum analysis on the reconstructed signal.

Images