A method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback

Through the methods of frequency domain convolution and marginal spectrum feedback, the modal aliasing and information redundancy problems of Hilbert yellow transformation algorithm in shrimp acoustic signal analysis are solved, and the accuracy and convenience of time-frequency feature extraction are improved. sex.

CN114882901BActive Publication Date: 2024-11-15TONGJI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210446792.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-04-26
Publication Date
2024-11-15
Estimated Expiration
2042-04-26

AI Technical Summary

Technical Problem

The existing Hilbert yellow transformation algorithm has modal aliasing and time spectrum information redundancy in shrimp acoustic signal analysis, which increases the difficulty of extracting time frequency features.

Method used

Using the method based on frequency domain convolution and marginal spectrum feedback, the frequency domain convolution vector is constructed to smooth the time spectrum through complementary set empirical modal decomposition and Hilbert spectrum analysis, and the marginal spectrum is used to provide feedback to improve frequency components. Distinguishing degree and reducing information redundancy further emphasize the strong frequency components of the signal through secondary feedback.

Benefits of technology

It effectively avoids modal aliasing, improves the intuitiveness of the time spectrum and the convenience of feature extraction, reduces information redundancy, and improves the accuracy of time-frequency feature extraction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114882901B_ABST
    Figure CN114882901B_ABST
Patent Text Reader

Abstract

The present invention relates to a method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback. The method adopts frequency domain convolution processing and marginal spectrum feedback to perform feature analysis and extraction on the time-frequency spectrum of shrimp sound signals. S1: Complementary set empirical mode decomposition is performed on shrimp sound signals to obtain a number of intrinsic mode function components, and then Hilbert spectrum analysis is performed on these intrinsic mode function components, and the time-frequency spectrum of shrimp sound signals is obtained after normalization processing; S2: Construct a frequency domain convolution vector according to actual conditions, perform frequency domain convolution processing on the normalized time-frequency spectrum, and obtain a time-frequency spectrum after frequency domain smoothing; S3: Calculate and obtain the marginal spectrum, and use its characteristic of reflecting the time-frequency spectrum characteristics to perform primary and secondary feedback on the time-frequency spectrum after frequency domain smoothing to obtain the optimized time-frequency spectrum under different marginal spectrum feedback times; S4: Compare and analyze the optimized time-frequency spectrum under different marginal spectrum feedback times, and obtain the time-frequency features of shrimp sound signals through comprehensive spectrum performance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the field of shrimp sound signal processing, and in particular to a shrimp sound signal time-frequency feature extraction method based on frequency domain convolution and marginal spectrum feedback. Background Art

[0002] Shrimp acoustic signals mainly exist in shrimp's eating and communication behaviors, and contain rich biological information. In actual production life, the analysis of shrimp acoustic signals can provide a reference for the study of shrimp habits and high-density intelligent farming. As a non-stationary signal, shrimp acoustic signals cannot be fully displayed by a single global time domain analysis or global frequency domain analysis such as Fourier transform, so time-frequency dual-domain analysis is needed. In addition, shrimp acoustic signals are mostly short and rapid, so the analysis method needs to ensure sufficient frequency resolution.

[0003] The most common time-frequency analysis methods are short-time Fourier transform, wavelet transform and Hilbert-Huang transform. The window shape of short-time Fourier transform is fixed and is not suitable for the high- and low-frequency variation characteristics of actual signals. Although wavelet transform has the advantage of multi-resolution, it is subject to the uncertainty theorem like Fourier transform and has limited frequency resolution. Compared with the first two transforms, the advantage of Hilbert-Huang transform is that it jumps out of the framework of Fourier thought. As an adaptive transform based on the characteristics of the signal itself, it does not need to consider the selection of basis functions. Its high frequency resolution also makes the frequency resolution of densely distributed areas in the time-frequency spectrum significantly higher than that of short-time Fourier transform and wavelet transform.

[0004] However, the original Hilbert-Huang transform algorithm still has problems such as mode mixing and time-frequency spectrum information redundancy. Intermittent phenomena are often caused by abnormal signal events. In the empirical mode decomposition of the Hilbert-Huang transform, the distribution of the extreme points of the signal will be changed, so that the local envelope of the abnormal event appears in the fitting envelope, and the obtained signal intrinsic mode function components have "mode mixing", resulting in erroneous intrinsic mode function components; the high frequency resolution of the Hilbert-Huang transform causes the information contained in the time-frequency spectrum to be too precise, so that the spectrum contains excessive information, resulting in "information redundancy" in the time-frequency spectrum, making it difficult to directly extract the signal time-frequency features from the time-frequency spectrum. In order to better extract the time-frequency features of shrimp sound signals, it is necessary to solve the problems of mode aliasing and time-frequency spectrum information redundancy in the original Hilbert-Huang transform algorithm. Summary of the invention

[0005] The purpose of the present invention is to solve the problems of mode aliasing and time-frequency spectrum information redundancy existing in the Hilbert-Huang transform algorithm and to provide a method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback. The method avoids mode aliasing by introducing complementary set empirical mode decomposition, constructs frequency domain convolution vectors and uses marginal spectra to give a primary feedback to the time-frequency spectrum to improve the distinction between different frequency components and reduce the information redundancy of the time-frequency spectrum. Then, the strong frequency components of the signal time-frequency spectrum are emphasized by secondary feedback of the marginal spectrum to the time-frequency spectrum, so as to effectively improve the intuitiveness of the time-frequency spectrum and the convenience of time-frequency feature extraction.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback, the method adopts frequency domain convolution processing and marginal spectrum feedback to perform feature analysis and extraction on the time-frequency spectrum of shrimp sound signals, the method comprising the following steps:

[0008] S1: Complementary set empirical mode decomposition is performed on the shrimp sound signal to obtain several intrinsic mode function components, and then Hilbert spectrum analysis is performed on these intrinsic mode function components. After normalization, the time-frequency spectrum of the shrimp sound signal is obtained;

[0009] S2: Construct a frequency domain convolution vector according to the actual situation, perform frequency domain convolution processing on the normalized time-frequency spectrum, so that the time-frequency spectrum is smooth in the frequency domain, and obtain a time-frequency spectrum after frequency domain smoothing;

[0010] S3: Calculate the marginal spectrum, and use its characteristics of reflecting the time-frequency spectrum to perform primary and secondary feedback on the time-frequency spectrum after frequency domain smoothing to obtain the optimized time-frequency spectrum under different marginal spectrum feedback times;

[0011] S4: Compare and analyze the optimized time-frequency spectrograms under different marginal spectrum feedback times, and obtain the time-frequency characteristics of shrimp sound signals through comprehensive spectrum performance.

[0012] Furthermore, the method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback is characterized in that the preliminary processing of the method for extracting time-frequency features of shrimp sound signals in step S1 includes complementary set empirical mode decomposition and Hilbert spectrum analysis; optionally, the time-frequency spectrum obtained by the Hilbert spectrum analysis is normalized.

[0013] Furthermore, the method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback is characterized in that the step S1 performs complementary set empirical mode decomposition on the shrimp sound signal to obtain the intrinsic mode function component and then performs Hilbert spectrum analysis on it to obtain a normalized time-frequency spectrum. The complementary set empirical mode decomposition and Hilbert spectrum analysis include the following steps:

[0014] S11: Let the original signal be x(i), and add a pair of positive and negative complementary white noise sequences to x(t)

[0015] x b1 (t) = x(t) + (-1) b σn(t)

[0016] S12: By empirical mode decomposition, x b1 (t) is decomposed into n intrinsic mode function components and the remainder R(t)

[0017]

[0018] S13: Repeat steps S11 and S12 m times and use different white noise sequences. The j-th decomposition result can be expressed as:

[0019]

[0020] S14: sum and average the corresponding intrinsic mode function components in each decomposition result to obtain the final intrinsic mode function:

[0021]

[0022] In the formula, b = 1, 2, n(t) is a white noise sequence with an expectation of 0 and a variance of 1, n is the number of IMFs, m is the number of integrations, σ is the standard deviation, and the value of σ is generally 0.1 to 0.2 times the standard deviation of the signal. At this point, the complementary set empirical mode decomposition is completed and the Hilbert spectrum analysis step is entered.

[0023] S16: The generated intrinsic mode function components are analyzed by Hilbert spectrum to obtain the Hilbert amplitude spectrum, which is expressed as

[0024]

[0025] Where w is the angular frequency, y i (t) is IMF i The Hilbert transform value of (t), P is the Cauchy principal value,

[0026] Furthermore, the method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback is characterized in that, in step S2, a frequency domain convolution vector is constructed according to actual needs, and the vector is used to perform frequency domain convolution processing on the time-frequency spectrum to smooth the time-frequency spectrum; optionally, different frequency domain convolution vectors are constructed according to actual needs for the time-frequency feature extraction of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback, and different time-frequency spectrum smoothing performances are obtained accordingly; the frequency domain convolution process comprises the following steps:

[0027] S21: The Hilbert amplitude spectrum obtained in step S16 can be rewritten in discrete form, and its expression is:

[0028]

[0029] Where s ij , i∈[1,k], j∈[1,l] represent the amplitude values ​​of the corresponding points in the time-frequency domain; given and The value of constructs the frequency domain convolution vector K, K is expressed as:

[0030]

[0031] S22: According to the given Pad S with zeros to get the matrix S0, the expression of S0 is:

[0032]

[0033] S23: Convolution matrix S0 using frequency domain convolution vector K. The center point of K is from s 11 Slide to s one by one lk . Set s′ pq , p∈[1,k],q∈[1,l] represents the initial position in S0 The convolution value of the point at row and column q is:

[0034]

[0035] Then the new value of the point after convolution is:

[0036]

[0037] The time-frequency spectrum after frequency domain convolution can be expressed as:

[0038]

[0039] Furthermore, the method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback is characterized in that the marginal spectrum is calculated in step S3, and the characteristic of reflecting the frequency characteristics of the time-frequency spectrum is utilized to perform primary feedback and secondary feedback on the time-frequency spectrum graph after frequency domain smoothing to obtain optimized time-frequency spectrum graphs under different marginal spectrum feedback times, and the specific flow of the feedback process is as follows:

[0040] S31: Calculate the marginal spectrum of the Hilbert time-frequency spectrum in step S16, and the expression of the marginal spectrum is:

[0041]

[0042] Where w is the signal angular frequency, f is the signal frequency, and T is the signal time span; the marginal spectrum can be regarded as the spectrum of the signal, reflecting the intensity of different frequency components in the signal. Therefore, using the value of the marginal spectrum to perform marginal spectrum feedback processing on the signal can increase the contrast between high-intensity and low-intensity frequency components in the time spectrum, and can more clearly distinguish the instantaneous frequency points of different intensities reflected in the spectrum.

[0043] S32: In order to avoid the loss of strong and weak energy discrimination caused by excessive feedback amplitude, the average marginal spectrum is used as the feedback factor. The time-frequency spectrum expression after one-time feedback optimization is:

[0044]

[0045] Its discrete expression is:

[0046]

[0047] Where l is S p The number of columns of S, k is p the number of rows, YesS p The Hilbert marginal spectrum of ;

[0048] S33: In order to highlight the strong frequency components, the marginal frequency is used to perform secondary feedback processing on the time-frequency spectrum, and the optimized time-frequency spectrum expression is obtained as follows:

[0049]

[0050] Its discrete expression is

[0051]

[0052] Where l is S z The number of columns of S, k is z The number of rows, A z YesS z The Hilbert marginal spectrum of

[0053]

[0054] Furthermore, the method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback is characterized in that the step S4 compares and analyzes the optimized time-frequency spectrograms under different marginal spectrum feedback times, and the time-frequency features of shrimp sound signals are obtained by integrating the spectrum performance. The marginal spectrum primary feedback improves the distinction between different frequency components in the time-frequency spectrum, and the intuitiveness of the result is improved; the marginal spectrum secondary feedback is processed again on the basis of the primary feedback result, the key frequency components of the time-frequency spectrum are highlighted, and other components are filtered out. The time-frequency features of shrimp sound signals can be extracted by combining and analyzing the time-frequency spectrograms of the primary feedback and the secondary feedback.

[0055] The present invention discloses a method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback, which is characterized by comprising complementary set empirical mode decomposition, Hilbert spectrum analysis, frequency domain convolution processing of time-frequency spectrum, primary frequency domain marginal spectrum feedback and secondary frequency domain marginal spectrum feedback of time-frequency spectrum, and the shrimp sound signal adopts the feature detection and extraction method program to execute the steps of the above method.

[0056] Compared with the prior art, the present invention has the following advantages:

[0057] This method introduces complementary set empirical mode decomposition to decompose the signal, effectively avoiding the modal aliasing problem that often exists in empirical mode decomposition of shrimp sound signals. The use of frequency domain convolution vectors and the primary feedback of marginal spectrum to time-frequency spectrum improve the distinction between different frequency components and reduce the spectral information redundancy caused by the high frequency resolution of original Hilbert spectrum analysis. The secondary feedback of marginal spectrum to time-frequency spectrum emphasizes the strong frequency components of the signal time-frequency spectrum, which is reflected in the extraction of time-frequency features of shrimp sound signals. This method can effectively improve the intuitiveness of time-frequency spectrum and the convenience of time-frequency feature extraction. The combination of primary and secondary feedback results also improves the accuracy of feature extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is the time domain waveform of the effective samples of shrimp sound signal;

[0059] Figure 2 A signal time-frequency spectrum diagram of a Hilbert-Huang transform algorithm using complementary set empirical mode decomposition for an embodiment;

[0060] Figure 3 The signal time spectrum of the Hilbert-Huang transform algorithm with frequency domain convolution processing added for the embodiment - the frequency domain convolution vector is [1, 1, 1, 1, 1] T ;

[0061] Figure 4 The signal time spectrum of the Hilbert-Huang transform algorithm with frequency domain convolution processing added for the embodiment - the frequency domain convolution vector is [1, 2, 4, 2, 1] T ;

[0062] Figure 5 The signal time spectrum of the Hilbert-Huang transform algorithm using a single feedback process for the embodiment - the frequency domain convolution vector is [1, 1, 1, 1, 1] T ;

[0063] Figure 6 The signal time spectrum of the Hilbert-Huang transform algorithm using quadratic feedback processing for the embodiment - the frequency domain convolution vector is [1, 1, 1, 1, 1] T ;

[0064] Figure 7 The flowchart of the method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback of the present invention is as follows; in the figure, t is time and f is frequency. DETAILED DESCRIPTION

[0065] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0066] Example 1

[0067] like Figure 7 As shown, a method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback is provided. The method uses frequency domain convolution processing and marginal spectrum feedback to perform feature analysis and extraction on the time-frequency spectrum of shrimp sound signals.

[0068] Shrimp sound signals are non-stationary signals, and conventional single-domain transforms such as Fourier transform cannot fully display the signal characteristics, so time-frequency dual-domain analysis is required. In this embodiment, the sound signal of white shrimp is taken as an example, and a series of data is obtained by collecting through a hydrophone, and then the data is pre-processed by sound identification, short-time Fourier analysis, and interception to obtain sample sound signal segments, which are provided to step S1.

[0069] The method comprises the following steps:

[0070] S1: Complementary set empirical mode decomposition is performed on the shrimp sound signal to obtain several intrinsic mode function components, and then Hilbert spectrum analysis is performed on these intrinsic mode function components. After normalization, the time-frequency spectrum of the shrimp sound signal is obtained;

[0071] S2: Construct a frequency domain convolution vector according to the actual situation, perform frequency domain convolution processing on the normalized time-frequency spectrum, so that the time-frequency spectrum is smooth in the frequency domain, and obtain a time-frequency spectrum after frequency domain smoothing;

[0072] S3: Calculate the marginal spectrum, and use its characteristics of reflecting the time-frequency spectrum to perform primary and secondary feedback on the time-frequency spectrum after frequency domain smoothing to obtain the optimized time-frequency spectrum under different marginal spectrum feedback times;

[0073] S4: Compare and analyze the optimized time-frequency spectrograms under different marginal spectrum feedback times, and obtain the time-frequency characteristics of shrimp sound signals through comprehensive spectrum performance.

[0074] The following is a detailed description of each part.

[0075] 1. Signal decomposition and Hilbert spectrum analysis

[0076] S11: A series of data is collected by hydrophone, and then the data is pre-processed by sound identification, short-time Fourier analysis and interception to obtain effective signal fragments such as Figure 1 The effective sample time domain waveform of shrimp sound signal is shown in the figure, which is used as the sample signal for extracting the time-frequency features of shrimp sound signal.

[0077] S12: Perform complementary set empirical mode decomposition on the signal x(t) to obtain the intrinsic mode function:

[0078]

[0079] Among them, the expression of the j-th decomposition is:

[0080]

[0081] Where b = 1, 2, n(t) is a white noise sequence with an expected value of 0 and a variance of 1, n is the number of IMFs, m is the number of integrations, and σ is the standard deviation, which is generally 0.1 to 0.2 times the standard deviation of the signal.

[0082] S13: The obtained intrinsic mode function components are analyzed by Hilbert spectrum to obtain the Hilbert amplitude spectrum, which is expressed as:

[0083]

[0084] Where w is the angular frequency of the signal, y i (t) is IMF i The Hilbert transform value of (t) is expressed as P is the Cauchy principal value, That is, the Hilbert-Huang transform algorithm based on complementary set empirical mode decomposition is used to extract the time-frequency features of the signal, and the time-frequency spectrum is obtained as follows: Figure 2 The signal spectrum is shown in the Hilbert-Huang transform algorithm using complementary ensemble empirical mode decomposition.

[0085] 2. Frequency domain convolution processing of signal time spectrum

[0086] Figure 2 The time-frequency spectrum shown is not intuitive enough to directly extract effective time-frequency features of shrimp sound signals. The mean frequency domain convolution vector K = [1, 1, 1, 1] is constructed. T The time-frequency spectrum is convolved in the frequency domain to obtain the time-frequency spectrum as shown in Figure 3 The signal spectrum of the Hilbert-Huang transform algorithm with frequency domain convolution is shown in the figure. Reconstruct the median frequency domain convolution vector K = [1, 2, 4, 2, 1] T The time-frequency spectrum is convolved in the frequency domain to obtain the time-frequency spectrum as shown in Figure 4The spectrum of the signal using the Hilbert-Huang transform algorithm with added frequency domain convolution is shown below.

[0087] 3. Frequency domain marginal spectrum feedback of signal time spectrum

[0088] S31: Observation Figure 3 and Figure 4 , Figure 3 The information representation is more comprehensive. Figure 4 The information contrast is higher. To ensure the full display of information, select Figure 3 Perform subsequent frequency domain marginal spectrum feedback processing. Figure 3 The marginal spectrum of the time spectrum is obtained, and the expression of the marginal spectrum is:

[0089]

[0090] Where f is the signal frequency, T is the signal time span; the marginal spectrum can be regarded as the spectrum of the signal, reflecting the intensity of different frequency components in the signal. Therefore, the marginal spectrum value can be used to perform marginal spectrum feedback processing on the signal to increase the contrast between high-intensity and low-intensity frequency components in the time spectrum, so as to better distinguish the instantaneous frequency points of different intensities reflected in the spectrum.

[0091] S32: In order to avoid the loss of strong and weak energy discrimination caused by excessive feedback amplitude, the average marginal spectrum is used as the feedback factor. The time-frequency spectrum expression after one-time feedback optimization is:

[0092]

[0093] Its discrete expression is:

[0094]

[0095] Where l is S p The number of columns of S, k is p the number of rows, YesS p After a marginal spectrum feedback process, the time-frequency spectrum is obtained as follows Figure 5 The signal spectrum is shown in the diagram using the Hilbert-Huang transform algorithm with added frequency domain convolution and single feedback processing.

[0096] S33: In order to highlight the strong frequency components, the marginal frequency is used to perform secondary feedback processing on the time-frequency spectrum, and the optimized time-frequency spectrum expression is obtained as follows:

[0097]

[0098] Its discrete expression is

[0099]

[0100] Where l is S z The number of columns of S, k is z The number of rows, A z YesS z The Hilbert marginal spectrum of

[0101] After the secondary marginal spectrum feedback processing, the time-frequency spectrum is obtained as follows Figure 6 The signal spectrum using the Hilbert-Huang transform algorithm with added frequency domain convolution and quadratic feedback processing is shown in the figure.

[0102] 4. Time-frequency feature extraction of shrimp sound signals

[0103] Combined reads Figure 5 and Figure 6 It can be seen that when the signal frequency is limited to 0-24kHz, the duration of the acoustic signal of white shrimp is about 5.5ms, and the main frequency is between 12 and 18kHz. The energy of the signal is very strong in the first 0.6ms. During this period, the signal continues to output strong energy for about 0.06ms, and the frequency is about 14.9-15.3kHz. After that, there is a small tail in the signal, which was not analyzed due to equipment limitations.

[0104] The above describes in detail the preferred specific embodiment method of the present invention. The method introduces complementary set empirical mode decomposition to decompose the signal, effectively solving the mode aliasing problem existing in the empirical mode decomposition of shrimp sound signals. The use of frequency domain convolution vectors and the primary feedback of marginal spectrum to time spectrum improve the discrimination between different frequency components, reduce the spectral information redundancy brought by the high frequency resolution of Hilbert spectrum analysis, and the secondary feedback of marginal spectrum to time spectrum highlights the strong frequency components of the signal time spectrum, which is reflected in the extraction of time-frequency features of shrimp sound signals. This method can effectively improve the intuitiveness of time spectrum and the convenience of time-frequency feature extraction, and the combination of primary feedback and secondary feedback results also improves the accuracy of feature extraction. It should be understood that ordinary technicians in this field can make many modifications and changes based on the concept of the present invention without creative labor. Therefore, all technical solutions that can be obtained by technicians in this technical field through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback. It is characterized in that This method uses frequency domain convolution processing and marginal spectrum feedback to analyze and extract the characteristics of the time-frequency spectrum of shrimp sound signals. The method specifically comprises the following steps: S1: Complementary set empirical mode decomposition is performed on the shrimp sound signal to obtain several intrinsic mode function components, and then Hilbert spectrum analysis is performed on these intrinsic mode function components. After normalization, the time-frequency spectrum of the shrimp sound signal is obtained; S2: Construct a frequency domain convolution vector according to the actual situation, perform frequency domain convolution processing on the normalized time-frequency spectrum, so that the time-frequency spectrum is smooth in the frequency domain, and obtain a time-frequency spectrum after frequency domain smoothing; S3: Calculate the marginal spectrum, and use its characteristic of reflecting the frequency characteristics of the time-frequency spectrum to perform primary and secondary feedback on the time-frequency spectrum after frequency domain smoothing to obtain the optimized time-frequency spectrum under different marginal spectrum feedback times; S4: Compare and analyze the optimized time-frequency spectrograms under different marginal spectrum feedback times, and obtain the time-frequency characteristics of shrimp sound signals by integrating the spectrogram performance; The step S2 constructs a frequency domain convolution vector according to actual needs, and uses the vector to perform frequency domain convolution processing on the time spectrum to smooth the time spectrum; for a shrimp sound signal time-frequency feature extraction based on frequency domain convolution and marginal spectrum feedback, different frequency domain convolution vectors are constructed according to actual needs to obtain different time-frequency spectrum smoothing performances; the frequency domain convolution process includes the following steps: S21: The Hilbert amplitude spectrum obtained in step S16 is rewritten into a discrete form, and its expression is: In the formula, s ij , i∈[1,k], j∈[1,l] represent the amplitude values ​​of the corresponding points in the time spectrum; and The value of is used to construct the frequency domain convolution vector K, K is expressed as: S22: According to the given Pad S with zeros to get the matrix S 0 , S 0 The expression is: S23: Convolution matrix S using frequency domain convolution vector K 0 ; The center point of K is from s 11 Slide to s one by one lk ; Set s′ pq , p∈[1,k],q∈[1,l] represents the initial position in S0 The convolution value of the point at row and column q is: Then the new value of the point after convolution is: The time-frequency spectrum after frequency domain convolution is expressed as: In step S3, the time-frequency spectrum diagram after frequency domain smoothing is fed back once and twice, and the specific process is as follows: S31: Calculate the marginal spectrum of the Hilbert time-frequency spectrum in step S16, and the expression of the marginal spectrum is: Where w is the signal angular frequency, f is the signal frequency, and T is the signal time span; the marginal spectrum is regarded as the spectrum of the signal, which reflects the intensity of different frequency components in the signal; S32: In order to avoid the loss of strong and weak energy discrimination due to excessive feedback amplitude, the average marginal spectrum is used as the feedback factor; the time-frequency spectrum expression after one-time feedback optimization is: Its discrete expression is: Where l is S p The number of columns of S, k is p the number of rows, YesS p The Hilbert marginal spectrum of ; S33: The time-frequency spectrum is subjected to secondary feedback processing using the marginal frequency, and the optimized time-frequency spectrum expression is obtained as follows: Its discrete expression is Where l is S z The number of columns of S, k is z The number of rows, A z YesS z The Hilbert marginal spectrum of 2. The method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback according to claim 1, It is characterized in that In step S1, the shrimp sound signal time-frequency feature extraction method comprises the following preliminary processing: complementary set empirical mode decomposition and Hilbert spectrum analysis; and the time-frequency spectrum obtained by the Hilbert spectrum analysis is normalized.

3. The method for extracting time-frequency features of shrimp sound signals based on frequency domain convolution and marginal spectrum feedback according to claim 2, It is characterized in that In step S1, the shrimp sound signal is subjected to complementary set empirical mode decomposition to obtain intrinsic mode function components, and then subjected to Hilbert spectrum analysis to obtain a normalized time-frequency spectrum; The complementary set empirical mode decomposition and Hilbert spectrum analysis include the following steps: S11: Let the original signal be x(t), and add a pair of positive and negative complementary white noise sequences to x(t) x b1 (t)=x(t)+(-1) b σn(t) S12: By empirical mode decomposition, x b1 (t) is decomposed into n intrinsic mode function components and the remainder R(t) S13: Repeat steps S11 and S12 m times and use different white noise sequences. The j-th decomposition result is expressed as: S14: sum and average the corresponding intrinsic mode function components in each decomposition result to obtain the final intrinsic mode function: Where b = 1, 2, n(t) is a white noise sequence with an expected value of 0 and a variance of 1, n is the number of IMFs, m is the number of integrations, σ is the standard deviation, and the value of σ is generally 0.1 to 0.2 times the standard deviation of the signal; At this point, the complementary set empirical mode decomposition is completed and the Hilbert spectrum analysis step is entered; S16: The generated intrinsic mode function components are analyzed by Hilbert spectrum to obtain the Hilbert amplitude spectrum, which is expressed as Where w is the angular frequency, y i (t) is IMF i The Hilbert transform value of (t), P is the Cauchy principal value,