Marine environment noise feature extraction method based on adaptive transfer scatter entropy

Abstract

The invention discloses a marine environment noise feature extraction method based on adaptive transfer scatter entropy, and relates to the technical field of marine environment noise processing, comprising the following steps: step 1, calculating skewness of a marine environment noise sequence, and performing adaptive mapping according to the skewness; step 2, using phase space reconstruction to construct the time sequence after mapping processing into a dispersion mode; 3, calculating the transition probability between the non-adjacent distribution modes, and counting the occurrence frequency of each transition probability; 4, calculating a self-adaptive transfer dispersion entropy value according to the information entropy definition, and taking the self-adaptive transfer dispersion entropy value as the feature of the marine environment noise sequence; and 5, inputting the features into an extreme learning machine classifier for classification verification. According to the marine environment noise feature extraction method based on the self-adaptive transfer spread entropy, the nonlinear characteristics of the marine environment noise sequence are described more comprehensively, the extracted features have higher stability and separability, and therefore classification and recognition of different marine environment noise signals are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of marine environmental noise processing, and in particular to a method for extracting features of marine environmental noise based on adaptive transfer dispersion entropy. Background Art

[0002] Marine environmental noise is an important research object in the field of underwater acoustic signal processing, covering various acoustic phenomena caused by natural factors and human activities, including wind and wave noise, biological noise, rainfall noise, and background noise generated by ship activities. Marine environmental noise contains rich environmental information, and the accurate extraction of its features is of great significance for applications such as marine environmental monitoring, target detection, and acoustic scene analysis. However, marine environmental noise usually has non-stationary, non-Gaussian, and non-linear characteristics, and the signal spectrum is complex and variable. Traditional linear signal processing methods are difficult to comprehensively characterize its dynamic features. In recent years, non-linear dynamics analysis methods have gradually become important technical means in this field, including various non-linear dynamics indicators such as fractal dimension, Lempe l-Ziv complexity, Lyapunov exponent, and information entropy. These indicators can more effectively describe the internal characteristics of complex noise signals. Among them, dispersion entropy has received extensive attention due to its high computational efficiency, strong robustness to noise interference, and strong feature separation ability. As a non-linear dynamics indicator that can quantify the complexity and uncertainty of signals, dispersion entropy provides an effective tool for the accurate feature extraction of marine environmental noise, which helps to promote the development of marine acoustics research and applications.

[0003] Dispersion entropy is an efficient tool for measuring the complexity of a finite time series, which quantifies the complexity of the time series by statistically analyzing the dispersion patterns of the time series. Specifically, dispersion entropy can reflect the uncertainty of the time series. The stronger the uncertainty, the larger the value of the dispersion entropy; conversely, if the time series has strong regularity or periodicity, the value of the dispersion entropy is smaller. Compared with other entropies, dispersion entropy shows better performance in terms of computational efficiency and noise resistance. In the research of marine environmental noise, dispersion entropy is widely used to evaluate the complexity of noise signals. It can not only characterize the characteristic differences of different types of noise signals but also has good class separation ability. However, when the types of marine environmental noise signals further increase, the ability of dispersion entropy to characterize some noise signal categories may weaken. Therefore, it is necessary to combine other methods or improvement strategies to further improve its ability to extract features of marine environmental noise. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for extracting features of marine environmental noise based on adaptive transfer dispersion entropy, which realizes the classification and recognition of marine environmental noise under different sea conditions and improves the recognition accuracy of the traditional feature extraction method based on dispersion entropy for marine environmental noise.

[0005] To achieve the above object, the present invention provides a method for extracting ocean environmental noise features based on adaptive transfer dispersion entropy, including the following steps:

[0006] Step 1: For the ocean environmental noise sequence, each type of sequence is divided into M samples, and the length of each sample is n. For any sample sequence X = {x i , i = 1, 2, …, N}, calculate the skewness of the ocean environmental noise sequence and perform adaptive mapping according to the skewness;

[0007] Step 2: Use phase space reconstruction to construct the processed time series into a dispersion pattern;

[0008] Step 3: Calculate the transition probability between non-adjacent dispersion patterns to obtain a transition probability matrix, and count the frequency of each transition probability;

[0009] Step 4: Calculate the adaptive transfer dispersion entropy value according to the definition of information entropy as the feature of the ocean environmental noise sequence;

[0010] Step 5: Input the feature into an extreme learning machine classifier for classification verification.

[0011] Preferably, the formula for skewness in Step 1 is:

[0012]

[0013] In the formula, σ and μ are the standard deviation and mean of the sequence X respectively, and x i is the i-th sample point of the sequence X;

[0014] Using adaptive mapping processing, map the sequence X = {x i , i = 1, 2, …, N} to Y = {y i , i = 1, 2, …, N}, and y i can be expressed as:

[0015]

[0016] In the formula, y i is the i-th sample point of the sequence Y, and t represents all possible value points in the normal distribution.

[0017] Preferably, the sequence obtained in Step 1 uses adaptive mapping instead of the single mapping method of dispersion entropy to improve the accuracy of entropy estimation.

[0018] Preferably, Step 2 includes the following steps:

[0019] S1. Convert the sequence Y into a positive integer sequence with an interval of [1, c] through the rounding function It is expressed as follows:

[0020]

[0021] In the formula, round represents the rounding function, c is the number of categories, is the i-th sample point of the sequence Z c ;

[0022] S2. Reconstruct the phase space of to construct the embedding vector as shown in the following formula:

[0023]

[0024] In the formula, m and d are the embedding dimension and the delay time respectively, and d usually takes 1;

[0025] S3. Each embedding vector corresponds to a spreading pattern where v 0 represents v 1 represents v m-1 represents

[0026] Preferably, in step three, the transition probability is calculated as follows:

[0027]

[0028] In the formula, k is the step size parameter, satisfying 1 < k < L; L = N - (m - 1)d - m, and number{*} represents the number of times * appears; represents denoting as the spreading pattern π α , represents denoting as the spreading pattern π β ; represents the transition probability between the spreading patterns π α and π β with an interval of k steps.

[0029] Preferably, the transition probability matrix P in step three can be expressed as:

[0030]

[0031] Preferably, the calculation formula of the adaptive transition spreading entropy in step four is:

[0032]

[0033] Therefore, the present invention adopts the above-mentioned method for extracting ocean environmental noise features based on adaptive transfer dispersion entropy. Starting from dispersion entropy, this method uses an adaptive mapping to replace the single mapping method of dispersion entropy, which can improve the accuracy of entropy values. By using the transition probability between non-adjacent dispersion patterns to replace the dispersion pattern of dispersion entropy, the introduced state transition information can improve the accuracy of feature extraction.

[0034] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0035] Figure 1 is a flowchart of the method for extracting ocean environmental noise features based on adaptive transfer dispersion entropy of the present invention;

[0036] Figure 2 is a time-domain waveform diagram of four types of measured ocean environmental noise signals of the present invention. (a) is noise 1, (b) is noise 2, (c) is noise 3, and (d) is noise 4;

[0037] Figure 3 is a distribution diagram of the adaptive transfer dispersion entropy features of four types of measured ocean environmental noise of the present invention. Detailed Embodiments

[0038] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0039] Unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meaning understood by those of ordinary skill in the field to which the present invention belongs. The "first", "second" and similar terms used in the present invention do not denote any order, quantity or importance, but are only used to distinguish different components. The terms such as "comprising" or "including" mean that the elements or objects appearing before this term cover the elements or objects listed after this term and their equivalents, without excluding other elements or objects. The terms such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left" and "right" are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0040] Embodiment

[0041] Please refer to Figures 1-3 , the present invention provides a method for extracting ocean environmental noise features based on adaptive transfer dispersion entropy, including the following steps:

[0042] Step 1. For several given types of ocean environmental noise sequences, each type of sequence is divided into M samples, and the length of each sample is n. For any sample sequence X = {x i , i =

[0043] 1, 2, …, N}, calculate the skewness of this sequence. The formula is as follows:

[0044]

[0045] where σ and μ are the standard deviation and mean of sequence X respectively, and x i is the i-th sample point of sequence X.

[0046] Use adaptive mapping processing to map the sequence X = {x i , i = 1, 2, …, N} to Y = {y i , i = 1, 2, …, N}. y i can be expressed as:

[0047]

[0048] where y i is the i-th sample point of sequence Y, and t represents all possible value points in the normal distribution; if |Skewness(X)| > 1, it means the sequence X = {x i , i = 1, 2, …, N} is highly skewed, and logarithmic mapping is used; if |Skewness(X)| ≤ 1, then the normal cumulative distribution function mapping is used.

[0049] Using adaptive mapping instead of the single mapping method of scatter entropy improves the accuracy of entropy estimation.

[0050] Step 2. Convert the sequence Y into a positive integer sequence with an interval of [1, c] through the rounding function It is expressed as follows:

[0051]

[0052] where round represents the rounding function, c is the number of categories, is the i-th sample point of sequence Z c .

[0053] Perform phase space reconstruction on to construct the embedding vector as shown in the following formula:

[0054]

[0055] where \(m\) and \(d\) are the embedding dimension and the time delay respectively; the embedding dimension \(m\) can determine the number of elements in each row in the phase space, and the time delay \(d\) is used to determine the composition of the elements in each row, and \(d\) is usually taken as 1.

[0056] Each embedding vector corresponds to a spreading pattern where \(v\) 0 denotes v 1 denotes v m-1 denotes

[0057] Step 3: Calculate the transition probability between non - adjacent spreading patterns As shown in the following formula

[0058]

[0059] where \(k\) is the step - length parameter, satisfying \(1\lt k\lt L\); \(L = N-(m - 1)d - m\), and number{\(*\)} represents the number of times * appears; denotes taking as the spreading pattern \(\pi\) α , denotes taking as the spreading pattern \(\pi\) β ; denotes the transition probability between the spreading patterns \(\pi\) α and \(\pi\) β with a step interval of \(k\).

[0060] The transition probability matrix \(P\) can be expressed as:

[0061]

[0062] Using the transition probability between non - adjacent spreading patterns to replace the spreading patterns of the spreading entropy, the introduced state - transition information can improve the accuracy of feature extraction.

[0063] Step 4: According to the definition of information entropy, the calculation formula of the adaptive transition spreading entropy is:

[0064]

[0065] The adaptive transition spreading entropy values of each sample are formed into a feature matrix.

[0066] Step 5: Classification verification. Randomly select a part of the feature vectors from the feature matrix in Step 4 as the training set, and the remaining feature vectors as the test set. Input the training set and the test set into the extreme learning machine classifier for classification.

[0067] Four types of measured ocean background noise signals are selected for feature extraction experiments. The flow chart is as shown in Figure 1 and the specific implementation steps are as follows:

[0068] Step 1: Use the existing published ocean environmental noise data to verify the effectiveness of the proposed method. For one type of ocean environmental noise signal, 200 samples are selected, and each sample contains 4096 sample points as sequence X. The time-domain waveforms of the four types of ocean environmental noise after normalization are as shown in Figure 2 . Calculate the skewness of sequence X, and obtain the mapped sequence Y according to the adaptive mapping.

[0069] Step 2: Obtain sequence Z according to the integer function c , and set the parameter c = 6. Perform phase space reconstruction on sequence Z c to obtain a phase space containing 4094 subsequences Set the parameters m = 3, d = 1. Each subsequence in the phase space represents a scattering pattern.

[0070] Step 3: Calculate the transition probability between non-adjacent scattering patterns Set the parameter k = 3 to obtain the transition probability matrix P.

[0071] Step 4: Calculate the adaptive transition scattering entropy value of sequence X according to the definition of Shannon entropy and output it as the feature of the ocean background noise signal;

[0072]

[0073] Calculate the adaptive transition scattering entropy value of each sample to obtain a 200×4 feature matrix.

[0074] Step 5: Classification verification. From the feature matrix obtained in Step 4, randomly select 50% of the feature vectors as the training set, and the remaining feature vectors as the test set. Input the training set and the test set into the extreme learning machine classifier for classification. The average recognition rate of the four types of ocean background noise is 94.0%. The distribution diagrams of the adaptive transition scattering entropy features of the four types of ocean background noise are as shown in Figure 3 , and the recognition rates of the adaptive transition scattering entropy for ocean environmental noise are shown in Table 1.

[0075] Table 1 Recognition rates of the adaptive transition scattering entropy for ocean environmental noise

[0076]

[0077] Therefore, the present invention adopts the above-mentioned method for extracting the characteristics of ocean environmental noise based on adaptive transfer dispersion entropy. Starting from dispersion entropy, this method uses an adaptive mapping to replace the single mapping method of dispersion entropy, which can improve the accuracy of entropy values. By using the transition probability between non-adjacent dispersion patterns to replace the dispersion pattern of dispersion entropy, the introduced state transition information can improve the accuracy of feature extraction.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for extracting features of ocean environmental noise based on adaptive transfer spread entropy, characterized by: The following steps are involved: Step 1: For the ocean ambient noise sequence, divide each type of sequence into M samples, each sample length is n, for any sample sequence X = {x i , i=1,2,…,N}, calculate the skewness of the ocean ambient noise sequence and perform adaptive mapping based on the skewness; Step 2: Use phase space reconstruction to construct the mapped time series into a scattered pattern; Step 3: Calculate the transition probability between non-adjacent scattering patterns to obtain a transition probability matrix, and count the frequency of occurrence of each transition probability; Step 4: Calculate the adaptive transfer spread entropy value according to the information entropy definition as the characteristic of the ocean environmental noise sequence; Step 5: Input the features into the extreme learning machine classifier for classification verification.

2. The method for extracting ocean environmental noise features based on adaptive transfer spread entropy according to claim 1 is characterized in that: The calculation formula for skewness in step 1 is: In the formula, σ and μ are the standard deviation and mean of the sequence X, respectively. i is the i-th sample point of sequence X; The sequence X = {x i ,i=1,2,…,N} is mapped to Y={y i ,i=1,2,…,N},y i It can be expressed as: In the formula, y i is the i-th sample point of sequence Y, and t represents all possible value points in the normal distribution.

3. The method for extracting ocean environmental noise features based on adaptive transfer spread entropy according to claim 2 is characterized in that: The sequence obtained in step 1 uses adaptive mapping instead of the single mapping method of distributed entropy to improve the accuracy of entropy estimation.

4. The method for extracting ocean environmental noise features based on adaptive transfer spread entropy according to claim 3 is characterized in that: Step 2 includes the following steps: S1. Convert the sequence Y into a positive integer sequence in the interval [1, c] through the rounding function It is expressed as follows: In the formula, round represents the rounding function, c is the number of categories, is the sequence Z c The i-th sample point; S2, yes Reconstruct phase space and construct embedding vector As shown below: In the formula, m and d are the embedding dimension and delay time respectively, and d is usually 1; S3, each embedding vector Each corresponds to a scattering pattern Where v0 represents v1 means v m-1 express 5. The method for extracting ocean environmental noise features based on adaptive transfer spread entropy according to claim 4 is characterized in that: The transition probability in step 3 The calculation formula is as follows: where k is the step size parameter, satisfying 1 < k < L; L = N - (m - 1)d - m, and number{*} represents the number of times * appears; denotes that is denoted as the spreading pattern π α , denotes that is denoted as the spreading pattern π β ; denotes the transition probability between the spreading pattern π α with a step interval of k steps and π β .

6. The method for extracting ocean environmental noise features based on adaptive transfer spread entropy according to claim 5 is characterized in that: The transition probability matrix P in step 3 can be expressed as:

7. The method for extracting ocean environmental noise features based on adaptive transfer spread entropy according to claim 6 is characterized in that: The calculation formula of the adaptive transfer spread entropy in step 4 is:

Images