Ship classification and identification method and device based on multi-channel underwater acoustic signal features
By reconstructing multi-channel underwater acoustic signals and performing ordinal mode frequency statistics in phase space, the problem of integrating multi-channel underwater acoustic signal information was solved, improving the accuracy and efficiency of ship classification and identification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NORTHWESTERN POLYTECHNICAL UNIV
- Filing Date
- 2026-02-06
- Publication Date
- 2026-05-15
AI Technical Summary
Existing technologies struggle to effectively integrate information from multi-channel underwater acoustic signals, resulting in low accuracy in ship classification and identification.
By coarsely processing the data of each channel of the multi-channel underwater acoustic signal and reconstructing it in phase space, using time delay vector and embedding dimension vector for dimensionality increase processing, statistically analyzing ordinal pattern frequencies, and integrating features using Cartesian product operations, ship classification and recognition can be performed.
It effectively integrates the features of multi-channel underwater acoustic signals, improves the accuracy of ship classification and identification, and reduces the computational load of the algorithm.
Smart Images
Figure CN121687117B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of underwater acoustic signal processing, and specifically to a method and apparatus for ship classification and identification based on multi-channel underwater acoustic signal characteristics. Background Technology
[0002] Underwater acoustic signals typically exhibit significant nonstationarity, nongaussianity, and nonlinearity. Their complex dynamic characteristics are crucial for underwater target identification, signal classification, and environmental monitoring. Nonlinear dynamics methods provide effective tools, revealing the underlying dynamic laws of signals through phase space reconstruction and complexity measurements. Commonly used methods include multiscale permutation entropy, improved multiscale permutation entropy, multivariate multiscale permutation entropy, and multivariate multiscale sample entropy.
[0003] Multi-scale permutation entropy and its improved counterpart can only quantify the complexity of single-channel data. Therefore, they cannot effectively distinguish signals with varying correlations between channels. Multivariate multi-scale permutation entropy and multivariate multi-scale sample entropy are nonlinear feature algorithms suitable for multi-channel data processing. However, multivariate multi-scale sample entropy exhibits unstable performance when processing real-world data, and the extracted features typically have high variance, leading to low classification accuracy. Furthermore, multivariate multi-scale permutation entropy essentially calculates the multi-scale permutation entropy for each channel separately and then fuses the results, thus failing to effectively distinguish signals with varying correlations between channels, resulting in low accuracy for target recognition. Summary of the Invention
[0004] To address the aforementioned shortcomings in existing technologies, the ship classification and identification method and apparatus based on multi-channel underwater acoustic signal features provided by this invention solves the problem that existing underwater acoustic signal feature extraction methods struggle to effectively integrate information from various channels, resulting in low accuracy in subsequent classification and identification.
[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:
[0006] A method for ship classification and identification based on multi-channel underwater acoustic signal features is provided, which includes the following steps:
[0007] Each channel of the multi-channel underwater acoustic signal is coarsely processed to decompose all channel data to the same scale, resulting in the decomposed channel data.
[0008] By setting a time delay vector and an embedding dimension vector, the phase space of the decomposed channel data is reconstructed to obtain an embedding delay vector that corresponds one-to-one with each decomposed channel data. The number of elements contained in the time delay vector and the embedding dimension vector are equal to the number of channels. The time delay and embedding dimension components of each channel are equal.
[0009] Each embedded delay vector is divided into multiple sub-vectors according to the length of the corresponding channel. The data in each sub-vector is arranged in ascending order. The element values in the ascending order are replaced with their indices in the sub-vector before the ascending order, thus obtaining the ordinal pattern corresponding to the sub-vector.
[0010] The ordinal pattern corresponding to a single channel is regarded as a vector, and an ordinal pattern matrix is constructed for all the ordinal patterns that can appear in a single channel; the number of columns in the ordinal pattern matrix corresponding to a single channel is the embedding dimension component of that channel, and the number of rows is the maximum number of ordinal patterns that can appear in that channel.
[0011] Perform a Cartesian product operation on the ordinal pattern matrix corresponding to each channel to obtain several composite ordinal patterns; the maximum number of composite ordinal patterns is the result of an exponentiation operation with the factorial of the embedded dimension components as the base and the number of channels as the exponent.
[0012] The frequency of each composite ordinal mode in all composite ordinal modes is counted, and the characteristics are calculated to obtain the characteristics of the multi-channel underwater acoustic signal.
[0013] The multi-channel underwater acoustic signal features are used as input to a pre-trained classification model for ship classification, resulting in ship classification and recognition results.
[0014] Furthermore, the data from each channel of the multi-channel underwater acoustic signal is coarse-grained, so that all channel data are decomposed to the same scale. The corresponding expression is:
[0015]
[0016] in Indicates the first The first channel in the Data collected at various times is decomposed at a scale of The corresponding decomposed data; For the first The first channel in the Data collected in real time; For decomposition scale; for The result is rounded down, where N is the total number of sampling times.
[0017] Furthermore, the expression corresponding to phase space reconstruction is:
[0018]
[0019] in For the first Phase space reconstruction results of the channel data after time decomposition; For embedding dimension vectors, , For the embedding dimension vector of the th One element; , representing the time delay vector, The first element in the time delay vector Each element.
[0020] Furthermore, the expression for the frequency of each composite ordinal pattern in all composite ordinal patterns is as follows:
[0021]
[0022] in Indicates the first Composite ordinal patterns Frequency of occurrence; The number of embedded delay vectors after phase space reconstruction; This represents the process of mapping each embedded delay vector to a composite ordinal pattern; The cardinality of a set.
[0023] Furthermore, feature calculation is a multi-scale permutation entropy calculation, the expression of which is:
[0024]
[0025] in This represents the result of multi-scale permutation entropy calculation; Indicates the maximum number of compound ordinal patterns. The factorial of the embedded dimension components; It is the natural logarithm.
[0026] A ship classification and identification device based on multi-channel underwater acoustic signal characteristics is provided, comprising:
[0027] A multi-channel underwater acoustic acquisition module is used to acquire underwater acoustic signals and obtain multi-channel underwater acoustic signals;
[0028] The data decomposition module is used to coarsely process the data of each channel of the multi-channel underwater acoustic signal, so that all channel data are decomposed to the same scale, and the decomposed channel data is obtained.
[0029] The phase space reconstruction module is used to reconstruct the phase space of the decomposed channel data by setting the time delay vector and the embedding dimension vector, so as to obtain the embedding delay vector corresponding to each decomposed channel data. The number of elements contained in the time delay vector and the embedding dimension vector are equal to the number of channels. The time delay and embedding dimension components of each channel are equal.
[0030] The ordinal pattern mapping module is used to divide each embedded delay vector into multiple sub-vectors according to the length of the corresponding channel, sort the data in each sub-vector in ascending order, and replace the element value in the ascending order with its index in the sub-vector before the ascending order to obtain the ordinal pattern corresponding to the sub-vector.
[0031] The composite ordinal pattern calculation module treats the ordinal pattern corresponding to a single channel as a vector, constructs an ordinal pattern matrix for all ordinal patterns that can appear in a single channel, and performs a Cartesian product operation on the ordinal pattern matrix corresponding to each channel to obtain several composite ordinal patterns. The number of columns in the ordinal pattern matrix corresponding to a single channel is the embedding dimension component of that channel, and the number of rows is the maximum number of ordinal patterns that can appear in that channel. The maximum number of composite ordinal patterns is the result of an exponentiation of the factorial of the embedding dimension component to the number of channels.
[0032] The feature extraction module is used to count the frequency of each composite ordinal mode in all composite ordinal modes and perform feature calculations to obtain the features of the multi-channel underwater acoustic signal.
[0033] The ship classification and recognition module is used to classify ships by taking the features of multi-channel underwater acoustic signals as input to a pre-trained classification model and obtaining the ship classification and recognition results.
[0034] Furthermore, the data from each channel of the multi-channel underwater acoustic signal is coarse-grained, so that all channel data are decomposed to the same scale. The corresponding expression is:
[0035]
[0036] in Indicates the first The first channel in the Data collected at various times is decomposed at a scale of The corresponding decomposed data; For the first The first channel in the Data collected in real time; For decomposition scale; for The result is rounded down, where N is the total number of sampling times.
[0037] Furthermore, the expression corresponding to phase space reconstruction is:
[0038]
[0039] in For the first Phase space reconstruction results of the channel data after time decomposition; For embedding dimension vectors, , For the embedding dimension vector of the th One element; , representing the time delay vector, The first element in the time delay vector Each element.
[0040] Furthermore, the expression for the frequency of each composite ordinal pattern in all composite ordinal patterns is as follows:
[0041]
[0042] in Indicates the first Composite ordinal patterns Frequency of occurrence; The number of embedded delay vectors after phase space reconstruction; This represents the process of mapping each embedded delay vector to a composite ordinal pattern; The cardinality of a set.
[0043] Furthermore, feature calculation is a multi-scale permutation entropy calculation, the expression of which is:
[0044]
[0045] in This represents the result of multi-scale permutation entropy calculation; Indicates the maximum number of compound ordinal patterns. The factorial of the embedded dimension components; It is the natural logarithm.
[0046] The beneficial effects of this invention are as follows: By reconstructing multi-channel data in phase space through dimensionality increase, this invention effectively integrates the information of each channel and performs ordinal pattern statistics using cross-channel joint event sets. Simultaneously, by using Cartesian products to rationally divide cross-channel joint events, the computational load of the algorithm is reduced, and the effective extraction of the complexity features of multi-channel underwater acoustic signals is achieved, providing a new and effective means for underwater acoustic signal feature extraction and related applications. Attached Figure Description
[0047] Figure 1 This is a flowchart illustrating the method.
[0048] Figure 2 It is a simulated signal curve calculated by the multivariable multiscale permutation entropy algorithm;
[0049] Figure 3 It is a simulated signal curve calculated by the cross-channel multi-scale permutation entropy algorithm;
[0050] Figure 4The simulated signal curve is obtained by calculating the improved multivariable multiscale sample entropy algorithm;
[0051] Figure 5 The measured ship radiated noise curve is obtained by the traditional multivariable multiscale permutation entropy algorithm.
[0052] Figure 6 It is the measured ship radiated noise curve obtained by improving the multivariate multiscale sample entropy calculation;
[0053] Figure 7 It is the measured ship radiated noise curve obtained by cross-channel multi-scale permutation entropy algorithm. Detailed Implementation
[0054] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.
[0055] like Figure 1 As shown, the ship classification and identification method based on multi-channel underwater acoustic signal features includes the following steps:
[0056] S1. Coarse-grained processing is performed on the data of each channel of the multi-channel underwater acoustic signal to decompose all channel data to the same scale, resulting in the decomposed channel data.
[0057] S2. By setting the time delay vector and the embedding dimension vector, the phase space of the decomposed channel data is reconstructed to obtain the embedding delay vector corresponding to each decomposed channel data. The number of elements contained in the time delay vector and the embedding dimension vector are equal to the number of channels. The time delay and embedding dimension components of each channel are equal.
[0058] S3. Divide each embedded delay vector into multiple sub-vectors according to the length of the corresponding channel, sort the data in each sub-vector in ascending order, and replace the element value in the ascending order with its index in the sub-vector before the ascending order to obtain the ordinal pattern corresponding to the sub-vector.
[0059] S4. Treat the ordinal pattern corresponding to a single channel as a vector, and construct an ordinal pattern matrix for all ordinal patterns that can appear in a single channel; where the number of columns in the ordinal pattern matrix corresponding to a single channel is the embedding dimension component of that channel, and the number of rows is the maximum number of ordinal patterns that can appear in that channel.
[0060] S5. Perform a Cartesian product operation on the ordinal pattern matrix corresponding to each channel to obtain several composite ordinal patterns; the maximum number of composite ordinal patterns is the result of an exponentiation operation with the factorial of the embedded dimension components as the base and the number of channels as the exponent.
[0061] S6. Count the frequency of each composite ordinal mode in all composite ordinal modes, and perform feature calculation to obtain the multi-channel underwater acoustic signal characteristics.
[0062] S7. Use the multi-channel underwater acoustic signal features as input to the pre-trained classification model to classify ships and obtain the ship classification and recognition results.
[0063] Corresponding to the method, the ship classification and identification device based on multi-channel underwater acoustic signal characteristics includes:
[0064] A multi-channel underwater acoustic acquisition module is used to acquire underwater acoustic signals and obtain multi-channel underwater acoustic signals;
[0065] The data decomposition module is used to coarsely process the data of each channel of the multi-channel underwater acoustic signal, so that all channel data are decomposed to the same scale, and the decomposed channel data is obtained.
[0066] The phase space reconstruction module is used to reconstruct the phase space of the decomposed channel data by setting the time delay vector and the embedding dimension vector, so as to obtain the embedding delay vector corresponding to each decomposed channel data. The number of elements contained in the time delay vector and the embedding dimension vector are equal to the number of channels. The time delay and embedding dimension components of each channel are equal.
[0067] The ordinal pattern mapping module is used to divide each embedded delay vector into multiple sub-vectors according to the length of the corresponding channel, sort the data in each sub-vector in ascending order, and replace the element value in the ascending order with its index in the sub-vector before the ascending order to obtain the ordinal pattern corresponding to the sub-vector.
[0068] The composite ordinal pattern calculation module treats the ordinal pattern corresponding to a single channel as a vector, constructs an ordinal pattern matrix for all ordinal patterns that can appear in a single channel, and performs a Cartesian product operation on the ordinal pattern matrix corresponding to each channel to obtain several composite ordinal patterns. The number of columns in the ordinal pattern matrix corresponding to a single channel is the embedding dimension component of that channel, and the number of rows is the maximum number of ordinal patterns that can appear in that channel. The maximum number of composite ordinal patterns is the result of an exponentiation of the factorial of the embedding dimension component to the number of channels.
[0069] The feature extraction module is used to count the frequency of each composite ordinal mode in all composite ordinal modes and perform feature calculations to obtain the features of the multi-channel underwater acoustic signal.
[0070] The ship classification and recognition module is used to classify ships by taking the features of multi-channel underwater acoustic signals as input to a pre-trained classification model and obtaining the ship classification and recognition results.
[0071] In the specific implementation process, the data of each channel of the multi-channel underwater acoustic signal is coarsely processed so that all channel data are decomposed to the same scale. The corresponding expression is:
[0072]
[0073] in Indicates the first The first channel in the Data collected at various times is decomposed at a scale of The corresponding decomposed data; For the first The first channel in the Data collected in real time; For decomposition scale; for The result is rounded down, where N is the total number of sampling times.
[0074] The expression corresponding to phase space reconstruction is:
[0075]
[0076] in For the first Phase space reconstruction results of the channel data after time decomposition; For embedding dimension vectors, , For the embedding dimension vector of the th One element; , representing the time delay vector, The first element in the time delay vector Each element is a component.
[0077] Theoretically, the embedding dimension and time delay components of each channel can be unequal, but setting them to be equal simplifies the theoretical description without affecting the algorithm's properties and calculation process. Therefore, in this embodiment, it is assumed that the embedding dimension and time delay components of each channel are equal, i.e. as well as Based on this, the length of the embedded delay vector after phase space reconstruction can be obtained as follows: .
[0078] In this embodiment, for a length of Given a vector whose elements are all of different sizes, the total number of ordinal patterns that can appear in the corresponding channel is: For a specific vector, all elements in the vector are sorted in ascending order, and their original indices are recorded. For example, the original vector is [30, 8, 24], which becomes [8, 24, 30] after being sorted in ascending order. The indices of the elements in the sorted vector in the original vector are [2, 3, 1] ([2, 3, 1] is an ordinal pattern). Based on this method, all embedded delay vectors in the phase space can be mapped to the ordinal pattern space. When the embedding dimension is... At that time, there were a total of The ordinal patterns that can appear (in terms of) Explanation: There are a total of There are a total of 6 possible ordinal patterns.
[0079] In this embodiment, the embedded delay vector obtained after phase space reconstruction It contains data from each channel, and the total length of the vector is [length missing]. The data length of each channel is For a specific channel (that is, the result of dividing each embedded delay vector into multiple sub-vectors according to the length of the corresponding channel), since its length is... Therefore, the above mapping method is used to embed the delay vector. This channel is partially mapped to ordinal patterns, and there are a total of ordinal patterns that can appear. If we consider the ordinal pattern as a vector, then all possible ordinal patterns for that channel constitute a... OK The matrix of columns. Perform a Cartesian product operation on the ordinal pattern matrices corresponding to different channels to obtain the final... Possible composite ordinal patterns, represented by the symbol express.
[0080] For each vector in the phase space Each can be assigned a corresponding composite ordinal pattern. Therefore, the expression for counting the frequency of each composite ordinal pattern in all composite ordinal patterns is:
[0081]
[0082] in Indicates the first Composite ordinal patterns Frequency of occurrence; The number of embedded delay vectors after phase space reconstruction; This represents the process of mapping each embedded delay vector to a composite ordinal pattern; The cardinality of the set is represented by the numerator. The numerator in the formula represents the number of vectors among all embedded delayed vectors that map to the composite ordinal pattern.
[0083] After statistically obtaining the frequency of all possible composite ordinal patterns, taking the calculation of multi-scale permutation entropy as an example, the expression is as follows:
[0084]
[0085] in This represents the result of multi-scale permutation entropy calculation; Indicates the maximum number of compound ordinal patterns; It is the natural logarithm.
[0086] The obtained cross-channel multi-scale permutation entropy result is a curve with scale as the abscissa and cross-channel permutation entropy value as the ordinate.
[0087] In one embodiment of the present invention, a simulation experiment was conducted to verify the effectiveness of the present invention. Figure 2 , Figure 3 and Figure 4 The simulation signals used consisted of multi-channel data containing both correlated and uncorrelated Gaussian white noise between channels. Each channel contained 10,000 data points, with each component of the embedding dimension set to 3, each component of the time delay set to 1, and a scale range of [1, 15]. The complexity of the multi-channel data was calculated using multivariate multiscale permutation entropy, cross-channel multiscale permutation entropy (the name of the method used in this study to extract features from multi-channel underwater acoustic signals), and an improved multivariate multiscale sample entropy.
[0088] from Figure 2 As can be seen, the multivariate multiscale permutation entropy algorithm fails to extract the implicit correlation information between channels, resulting in... Figure 2 The entropy curves of different types of signals showed severe overlap, further rendering it impossible to effectively distinguish between different signals using multivariate, multi-scale entropy arrangements. Figure 3 and Figure 4 The improved multivariate multiscale sample entropy and cross-channel multiscale permutation entropy algorithms effectively distinguish between the two types of signals. Figure 3 The multi-scale arrangement entropy of the mid-channel array shows good distinguishability between the two types of signals, while the entropy value of uncorrelated Gaussian white noise between channels is significantly higher than that of correlated Gaussian white noise between channels. In contrast, Figure 4 The improved multivariate multiscale sample entropy has relatively poor distinguishability between the two types of signals, and the entropy curves of the two signals are quite similar and difficult to distinguish.
[0089] Figure 5 , Figure 6 and Figure 7The data processing results for ship radiated noise with different nonlinear characteristics are presented. Two sets of multi-channel ship radiated noise data are used, with 3 channels and an array spacing of 5 meters. The array sampling rate is 9960Hz. The embedding dimension of each channel for the multivariate multi-scale permutation entropy and the cross-channel multi-scale permutation entropy are set to 2, and the time delay is set to 5. The scale range is [1, 15]. The embedding dimension of each channel for the improved multivariate multi-scale sample entropy is set to 2, the time delay is set to 1, and the threshold coefficient is set to 0.2. The multivariate multi-scale permutation entropy curve results are shown below. Figure 5 As shown, the improved multivariate multiscale sample entropy curve results are as follows: Figure 6 As shown, the entropy results of cross-channel multi-scale permutation are as follows: Figure 7 As shown.
[0090] contrast Figure 5 and Figure 7 It can be seen that the two sets of ship radiated noise entropy curves calculated using the multivariate multiscale permutation entropy algorithm exhibit significant fluctuations in entropy values with scale changes. Furthermore, the means of the two entropy curves are similar, but their variances are large, causing the multivariate multiscale permutation entropy algorithm to be unable to effectively distinguish between the two different types of ship radiated noise. Figure 7 In the cross-channel multi-scale arrangement entropy curves, the entropy value of ship 1 decreases continuously with increasing scale, while the entropy curve of ship 2 shows a slow upward trend with increasing scale. Meanwhile, the variances of the entropy curves of the three types of ship radiated noise are relatively small, making it possible to effectively distinguish different types of ship radiated noise using cross-channel multi-scale arrangement entropy. On the other hand, Figure 6 While the improved multivariate multiscale sample entropy curves of radiated noise from different types of ships can distinguish between different types of signals to some extent, the overall mean values of the curves are too close and overlap to some extent, affecting the accuracy of subsequent classification and recognition.
[0091] In this embodiment, the classification model can be a probabilistic neural network (PNN) or other existing classifiers. The specific structure of the classification model is not innovative in this application and will not be described in detail.
[0092] The following is given Figures 5-7 The measured data were used to identify and classify ships using a probabilistic neural network. 100 results were calculated for each curve, with 70 used as the training set and 30 as the test set. The classification accuracy of the features obtained from the three methods for ships is shown in Tables 1, 2, and 3.
[0093] Table 1: Classification accuracy of the improved multivariate multiscale sample entropy
[0094]
[0095] Table 2: Classification accuracy of multivariate multiscale permutation entropy
[0096]
[0097] Table 3: Classification and Recognition Accuracy of Cross-Channel Multi-Scale Permutation Entropy
[0098]
[0099] In summary, this invention effectively integrates information from each channel by performing dimensionality-up reconstruction of multi-channel data in phase space, and utilizes cross-channel joint event sets for ordinal pattern statistics. Simultaneously, it employs Cartesian products to rationally partition cross-channel joint events, reducing computational complexity and effectively extracting the complexity features of multi-channel underwater acoustic signals, thus providing a new and effective method for underwater acoustic signal feature extraction and ship classification and identification.
Claims
1. A ship classification and identification method based on multi-channel underwater acoustic signal features, characterized in that, Includes the following steps: Each channel of the multi-channel underwater acoustic signal is coarsely processed to decompose all channel data to the same scale, resulting in the decomposed channel data. By setting a time delay vector and an embedding dimension vector, the phase space of the decomposed channel data is reconstructed to obtain an embedding delay vector that corresponds one-to-one with each decomposed channel data. The number of elements contained in the time delay vector and the embedding dimension vector are equal to the number of channels. The time delay and embedding dimension components of each channel are equal. Each embedded delay vector is divided into multiple sub-vectors according to the length of the corresponding channel. The data in each sub-vector is arranged in ascending order. The element values in the ascending order are replaced with their indices in the sub-vector before the ascending order, thus obtaining the ordinal pattern corresponding to the sub-vector. Treat the ordinal pattern corresponding to a single channel as a vector, and construct an ordinal pattern matrix for all ordinal patterns that can appear in a single channel. The number of columns in the ordinal pattern matrix corresponding to a single channel is the embedding dimension component of that channel, and the number of rows is the maximum number of ordinal patterns that can appear in that channel. Perform a Cartesian product operation on the ordinal pattern matrix corresponding to each channel to obtain several composite ordinal patterns; the maximum number of composite ordinal patterns is the result of an exponentiation operation with the factorial of the embedded dimension components as the base and the number of channels as the exponent. The frequency of each composite ordinal mode in all composite ordinal modes is counted, and the characteristics are calculated to obtain the characteristics of the multi-channel underwater acoustic signal. The multi-channel underwater acoustic signal features are used as input to a pre-trained classification model for ship classification, resulting in ship classification and recognition results.
2. The ship classification and identification method based on multi-channel underwater acoustic signal features according to claim 1, characterized in that, The expression for coarsely processing each channel of the multi-channel underwater acoustic signal to decompose all channel data to the same scale is: in Indicates the first The first channel in the The data collected at each time point is decomposed at a scale of The corresponding decomposed data; For the first The first channel in the Data collected in real time; For decomposition scale; for The result is rounded down, where N is the total number of sampling times.
3. The ship classification and identification method based on multi-channel underwater acoustic signal features according to claim 2, characterized in that, The expression corresponding to phase space reconstruction is: in For the first Phase space reconstruction results of the channel data after time decomposition; For embedding dimension vectors, , For the embedding dimension vector of the th One element; , representing the time delay vector, The first element in the time delay vector Each element.
4. The ship classification and identification method based on multi-channel underwater acoustic signal features according to claim 3, characterized in that, The expression for the frequency of each composite ordinal pattern in all composite ordinal patterns is: in Indicates the first Composite ordinal patterns Frequency of occurrence; The number of embedded delay vectors after phase space reconstruction; This represents the process of mapping each embedded delay vector to a composite ordinal pattern; The cardinality of a set.
5. The ship classification and identification method based on multi-channel underwater acoustic signal features according to claim 4, characterized in that, Feature calculation is a multi-scale permutation entropy calculation, and its expression is: in This represents the result of multi-scale permutation entropy calculation; Indicates the maximum number of compound ordinal patterns. The factorial of the embedded dimension components; It is the natural logarithm.
6. A ship classification and identification device based on multi-channel underwater acoustic signal characteristics, characterized in that, include: A multi-channel underwater acoustic acquisition module is used to acquire underwater acoustic signals and obtain multi-channel underwater acoustic signals; The data decomposition module is used to coarsely process the data of each channel of the multi-channel underwater acoustic signal, so that all channel data are decomposed to the same scale, and the decomposed channel data is obtained. The phase space reconstruction module is used to reconstruct the phase space of the decomposed channel data by setting the time delay vector and the embedding dimension vector, so as to obtain the embedding delay vector corresponding to each decomposed channel data. The number of elements contained in the time delay vector and the embedding dimension vector are equal to the number of channels. The time delay and embedding dimension components of each channel are equal. The ordinal pattern mapping module is used to divide each embedded delay vector into multiple sub-vectors according to the length of the corresponding channel, sort the data in each sub-vector in ascending order, and replace the element value in the ascending order with its index in the sub-vector before the ascending order to obtain the ordinal pattern corresponding to the sub-vector. The composite ordinal pattern calculation module treats the ordinal pattern corresponding to a single channel as a vector, constructs an ordinal pattern matrix for all ordinal patterns that can appear in a single channel, and performs a Cartesian product operation on the ordinal pattern matrix corresponding to each channel to obtain several composite ordinal patterns. The number of columns in the ordinal pattern matrix corresponding to a single channel is the embedding dimension component of that channel, and the number of rows is the maximum number of ordinal patterns that can appear in that channel. The maximum number of composite ordinal patterns is the result of an exponentiation of the factorial of the embedding dimension component to the number of channels. The feature extraction module is used to count the frequency of each composite ordinal mode in all composite ordinal modes and perform feature calculations to obtain the features of the multi-channel underwater acoustic signal. The ship classification and recognition module is used to classify ships by taking the features of multi-channel underwater acoustic signals as input to a pre-trained classification model and obtaining the ship classification and recognition results.
7. The ship classification and identification device based on multi-channel underwater acoustic signal characteristics according to claim 6, characterized in that, The expression for coarsely processing each channel of the multi-channel underwater acoustic signal to decompose all channel data to the same scale is: in Indicates the first The first channel in the The data collected at each time point is decomposed at a scale of The corresponding decomposed data; For the first The first channel in the Data collected in real time; For decomposition scale; for The result is rounded down, where N is the total number of sampling times.
8. The ship classification and identification device based on multi-channel underwater acoustic signal characteristics according to claim 7, characterized in that, The expression corresponding to phase space reconstruction is: in For the first Phase space reconstruction results of the channel data after time decomposition; For embedding dimension vectors, , For the embedding dimension vector of the th One element; , representing the time delay vector, The first element in the time delay vector Each element.
9. The ship classification and identification device based on multi-channel underwater acoustic signal characteristics according to claim 8, characterized in that, The expression for the frequency of each composite ordinal pattern in all composite ordinal patterns is: in Indicates the first Composite ordinal patterns Frequency of occurrence; The number of embedded delay vectors after phase space reconstruction; This represents the process of mapping each embedded delay vector to a composite ordinal pattern; The cardinality of a set.
10. The ship classification and identification device based on multi-channel underwater acoustic signal characteristics according to claim 9, characterized in that, Feature calculation is a multi-scale permutation entropy calculation, and its expression is: in This represents the result of multi-scale permutation entropy calculation; Indicates the maximum number of compound ordinal patterns. The factorial of the embedded dimension components; It is the natural logarithm.