A Ship Identification Method and Device

Abstract

The present invention provides a ship identification method and device, including: obtaining the discrimination features of a target underwater acoustic signal; performing dimensionality reduction processing on the discrimination features to obtain a signal vector; and inputting the signal vector into a ship identification model to determine the ship type corresponding to the target underwater acoustic signal. The ship identification method and device provided by the present invention reduce redundant features or noise data by performing dimensionality reduction on the discrimination features of the target underwater acoustic signal, effectively reducing the computational amount of the ship identification model and the memory space occupied by data storage, improving the identification speed of the model, providing a basis for real-time monitoring of ships, and at the same time reducing the influence of channel interference to a certain extent in the process of collecting underwater acoustic signals of ships.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and particularly to a ship identification method and device. Background Art

[0002] Ships are important targets on the sea surface. Ship identification through underwater acoustic signals has broad application prospects in both civilian and military fields.

[0003] Existing underwater acoustic identification technologies are usually implemented based on mathematical formulas. For example, the signal is subjected to wavelet decomposition for denoising and then converted into a spectrogram, and the identification result is obtained through voiceprint comparison.

[0004] However, the above technologies require staff to perform complex formula derivations and detailed algorithm principle explanations, consuming a large amount of human resources and time, and having hysteresis. Summary of the Invention

[0005] In view of the problems existing in the prior art, embodiments of the present invention provide a ship identification method and device.

[0006] The present invention provides a ship identification method, including:

[0007] Obtaining a discrimination feature of a target underwater acoustic signal;

[0008] Performing dimensionality reduction processing on the discrimination feature to obtain a signal vector;

[0009] Inputting the signal vector into a ship identification model to determine the ship type corresponding to the target underwater acoustic signal.

[0010] According to the ship identification method provided by the present invention, the obtaining of the discrimination feature of the target underwater acoustic signal includes:

[0011] Based on the maximum likelihood estimation method, using the target underwater acoustic signal to obtain a universal background model;

[0012] Based on the maximum a posteriori probability of the universal background model, obtaining the mean supervector of the target underwater acoustic signal;

[0013] According to the mean supervector, obtaining the discrimination feature.

[0014] According to the ship identification method provided by the present invention, the performing dimensionality reduction processing on the discrimination feature to obtain a signal vector includes:

[0015] Based on the probabilistic linear discriminant analysis model, performing dimensionality reduction decomposition on the discrimination feature to obtain a dimensionality reduction vector;

[0016] The dimensionality reduction vector includes a signal vector and a noise vector;

[0017] The probability linear discriminant analysis model is described based on the conditional probability of the Gaussian distribution.

[0018] According to a ship recognition method provided by the present invention, the ship recognition model is constructed based on a long short-term memory network;

[0019] Before inputting the signal vector into the ship recognition model, the method further includes:

[0020] Obtaining a plurality of underwater acoustic signal samples and the ship type label corresponding to each underwater acoustic signal sample;

[0021] Obtaining the sample feature vector corresponding to each underwater acoustic signal sample;

[0022] Performing dimensionality reduction processing on each sample feature vector to obtain the signal vector sample corresponding to each underwater acoustic signal sample;

[0023] Taking the combination of the signal vector sample corresponding to each underwater acoustic signal sample and the ship type label as a training sample, obtaining a plurality of training samples, and training a preset recognition model by using the plurality of training samples.

[0024] According to a ship recognition method provided by the present invention, training the preset recognition model by using the plurality of training samples includes:

[0025] For any one training sample, inputting the training sample into the ship recognition model and outputting the prediction probability corresponding to the training sample;

[0026] Calculating a loss value by using a preset loss function according to the prediction probability corresponding to the training sample and the ship type label in the training sample;

[0027] If the loss value is less than a preset threshold, the training of the preset recognition model is completed, and the ship recognition model is obtained.

[0028] According to a ship recognition method provided by the present invention, the discriminant feature is dimensionally reduced and decomposed based on the following formula to obtain the reduced-dimensional vector:

[0029] w(i) = μ + Fh i + Gs i + ∈ i ;

[0030] where w(i) is the discriminant feature, i is the number of channels for collecting the target underwater acoustic signal; μ + Fh i + Gs i + ∈ i is the reduced-dimensional vector, μ + Fh i is the signal vector, Gs i + ∈ iis the noise vector;

[0031] μ is the mean of all input training data; F is the underwater acoustic class space; h i is the position of the underwater acoustic signal in the entire underwater acoustic class space; G is the error space; s i is the position of the underwater acoustic signal in the error space; ∈ i is the residual signal noise term.

[0032] The present invention also provides a ship identification device, including:

[0033] An acquisition module, configured to acquire the discrimination features of the target underwater acoustic signal;

[0034] A dimensionality reduction module, configured to perform dimensionality reduction processing on the discrimination features to obtain a signal vector;

[0035] A determination module, configured to input the signal vector into a ship identification model to determine the ship type corresponding to the target underwater acoustic signal.

[0036] According to a ship identification device provided by the present invention, the acquisition module is specifically configured to:

[0037] Based on the maximum likelihood estimation method, use the target underwater acoustic signal to obtain a general background model;

[0038] Based on the maximum a posteriori probability of the general background model, obtain the mean supervector of the target underwater acoustic signal;

[0039] According to the mean supervector, obtain the discrimination features.

[0040] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of any one of the above-mentioned ship identification methods are implemented.

[0041] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the above-mentioned ship identification methods are implemented.

[0042] The ship identification method and device provided by the present invention reduce redundant features or noise data by performing dimensionality reduction on the discrimination features of the target underwater acoustic signal, effectively reducing the computational amount of the ship identification model and the memory space occupied by data storage, improving the identification speed of the model, providing a basis for real-time monitoring of ships, and at the same time reducing the influence of channel interference to a certain extent in the process of collecting underwater acoustic signals of ships. Description of the Drawings

[0043] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0044] Figure 1 is a schematic flowchart of the ship identification method provided by the present invention;

[0045] Figure 2 is a schematic structural diagram of the ship identification device provided by the present invention;

[0046] Figure 3 is a schematic structural diagram of the electronic device provided by the present invention. Detailed Embodiments

[0047] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention with reference to the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0048] It should be noted that in the description of the embodiments of the present invention, the terms "include", "comprise", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device including a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article, or device. Without further limitation, the element defined by the phrase "including one..." does not exclude the existence of additional identical elements in the process, method, article, or device including the said element. The orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be construed as a limitation of the present invention. Unless otherwise clearly defined and limited, the terms "mount", "connect", and "couple" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the internal communication of two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0049] The earliest ship identification technology mainly relied on sonar engineers to passively monitor and observe ship signal patterns. This experience-based identification method was affected to a certain extent by machine performance, the psychological quality and concentration of staff, and the time-varying and random nature of ship signal transmission. This method, which relies solely on manual analysis of ship signals, has gradually shown its fatigue effect.

[0050] With the outstanding results achieved by artificial intelligence technology in target recognition, image processing, geological exploration, fault diagnosis and other fields, more and more attention has begun to shift to exploring and building deep neural networks based on intelligent methods to identify ship signals.

[0051] The multi-path effect of underwater sound transmission is one of the reasons for the complexity of underwater acoustic signals. Underwater acoustic signals in different sea areas will show different state characteristics at different times, and have the characteristics of nonlinearity, non-Gaussianity and non-stationarity. Therefore, we cannot rely solely on mathematical methods such as conditional assumptions to divide underwater acoustic signals in different time domains and spatial domains, but seek a method to explore the characteristics of underwater acoustic signals and the similarity between targets in different spatial dimensions.

[0052] The recognition of underwater acoustic signals includes four parts: data extraction, feature extraction, feature selection, and classification decision. Through a reasonable and effective feature extractor, the target effective information is analyzed from the sonar array signal, and then a suitable high-precision classifier is selected to find the unique feature expression of different types of ship signals.

[0053] The method of ship type identification through ship signals can be divided into two stages:

[0054] In the first stage, the fuzzy matching ship signal recognition technology based on the feature library obtains parameter estimates from the training samples, and then calculates the matching degree between the samples to be identified and various known samples.

[0055] Specifically, it includes: constructing a finite impulse response neural network to identify passive ship signals, and by designing a quadratic curve cross-section function neural network, shallow water ship signals can be effectively identified. By using empirical mode decomposition, signal components that reflect the characteristics of ship signals can be adaptively extracted according to the intrinsic characteristics of the ship signals themselves. For the problem of weak signals from long-range underwater targets, a hybrid intelligent recognition method is established by integrating empirical mode decomposition, feature distance estimation and fuzzy support vectors.

[0056] In the second stage, intelligent ship signal recognition technologies based on neural networks, including convolutional neural networks and recurrent neural networks, are involved. Artificial intelligence has made breakthroughs in the field of ship signal recognition due to characteristics such as large-scale parallel processing, distributed information storage, non-linear high-order feature expression, and deep network layout structure. As an end-to-end learning method, it can autonomously learn features to achieve ship signal category classification.

[0057] For example, by leveraging the characteristics of support vector machines in machine learning, a precise and fast real-time ship signal processing system is established, and an intelligent ship target intelligent recognition algorithm integrating empirical mode decomposition, feature distance evaluation technology, and combined support vector machines is proposed.

[0058] The ship recognition technology based on the above methods has at least the following defects.

[0059] Firstly, according to the traditional time-domain and frequency-domain change characteristics of the original ship signals, ship signal features are extracted. Due to various factors such as environmental complexity, a wide variety of target types, target antagonism, and target stealth features, there is a cold start problem in ship signal recognition. It is difficult to obtain an accurate confusion matrix between various sample sets, making it impossible to correct the matching algorithm between samples in a timely manner when there are certain deviations.

[0060] Secondly, during the model design process, in order to achieve the interpretability of ship signal recognition, complex formula derivations and detailed algorithm principle explanations by staff are required for each step. These processes will consume a large amount of human resources to a certain extent and often make it difficult to quickly build a real-time ship signal classification system for ship signals.

[0061] Thirdly, the underwater target features have poor generality and are vulnerable to various problems such as environmental changes, ship speeds, and channel interferences. The directly extracted low-order features are used for end-to-end learning, and different signals may present the same state in different time and space, making it impossible to achieve the best match by selecting a deep neural network.

[0062] Next, in combination with Figures 1 to 3 the ship recognition method and device provided by the embodiments of the present invention will be described.

[0063] Figure 1 is a schematic flowchart of the ship recognition method provided by the present invention. As Figure 1 shown, it includes but is not limited to the following steps:

[0064] First of all, in step S1, the discriminant features of the target underwater acoustic signal are obtained.

[0065] A sonar array can be used to collect signals from target ships within the acquisition range to obtain target underwater acoustic signals. Among them, the target underwater acoustic signals can include ship signals of the target ship and ocean noise signals. The ship signals come from the mechanical noise, propeller noise, and hydrodynamic noise of the target ship; the ocean noise signals come from the ocean environment.

[0066] The discriminative feature can be an Identity-vector (I-vector). The I-vector is a fixed-length low-dimensional vector with good direction discrimination.

[0067] Specifically, a Gaussian Mixture Model-Universal Background Model (GMM-UBM) is used to extract vectors from the target underwater acoustic signals to obtain discriminative features.

[0068] Further, in step S2, dimensionality reduction processing is performed on the discriminative features to obtain signal vectors.

[0069] For the sound source target of the target ship, Probabilistic Linear Discriminant Analysis (PLDA) can be used to perform horizontal batch processing on the discriminative features corresponding to the target underwater acoustic signals collected by the sonar array through different signal channels, which can eliminate channel differences to a certain extent; then PLDA is used to perform dimensionality reduction decomposition on the discriminative features to obtain signal vectors and noise vectors, realizing the separation of ship signals and ocean noise signals.

[0070] Further, in step S3, the signal vectors are input into the ship recognition model to determine the ship type corresponding to the target underwater acoustic signals.

[0071] The obtained signal vectors are input into the ship recognition model. The ship recognition model recognizes the signal vectors to obtain the classification results output by the ship recognition model. According to the classification results, the ship type corresponding to the target underwater acoustic signals can be obtained. In the same sea area, the ship recognition method provided by the present invention has strong robustness for ship type recognition and can effectively handle the cold start problem of ship classification.

[0072] The ship types can include multiple types such as fishing boats, law enforcement ships, cargo ships, container ships, speedboats, passenger ships, search and rescue ships, tugboats, and sailboats.

[0073] The ship identification method provided by the present invention reduces the dimensionality of the discriminative features of the target underwater acoustic signal to reduce redundant features or noise data, effectively reducing the computational load of the ship identification model and the memory space occupied by data storage, improving the identification speed of the model, providing a basis for the real-time monitoring of ships, and at the same time reducing the influence of channel interference to a certain extent in the process of collecting underwater acoustic signals of ships.

[0074] Optionally, obtaining the discriminative features of the target underwater acoustic signal includes:

[0075] Based on the maximum likelihood estimation method, using the target underwater acoustic signal to obtain a universal background model;

[0076] Based on the maximum a posteriori probability of the universal background model, obtaining the mean supervector of the target underwater acoustic signal;

[0077] According to the mean supervector, obtaining the discriminative features.

[0078] Among them, the universal background model can be GMM-UBM.

[0079] Specifically, given the target underwater acoustic signal as: Where is a string of D-dimensional feature vectors extracted from the i-th channel.

[0080] For the target underwater acoustic signal A universal background can be obtained by the maximum likelihood estimation (MLE) method based on the training of the target underwater acoustic signal, as follows:

[0081]

[0082] Where c k is the mixing coefficient, N(·; m k , R k ) is jointly represented by the D-dimensional mean vector m k and the D×D diagonal covariance matrix R k , representing the description symbol of the normal distribution of the observable data set y.

[0083] It can be obtained that θ = {c k , m k , R k | k = 1,..., K}, as a set of parameters of the universal background model.

[0084] Assume that a certain type of underwater acoustic signal Y i , can be represented by the mean supervector M(i) of the signal and the block component R kIt is represented by a D·K×D·K block-diagonal matrix R0 formed.

[0085] After that, using the maximum a posteriori probability of GMM-UBM, the mean supervector M(i) describing the target underwater acoustic signal is obtained as follows:

[0086] M(i) = M0 + Tw(i);

[0087] where M0 is the m of the universal background model k concatenated into a D·K-dimensional supervector; T is a low-rank matrix of D·K×F (F << D·K) dimensions, called the total variability matrix; w(i) is an F-dimensional random vector with a prior distribution that satisfies the standard normal distribution N(·; 0, I).

[0088] where the mean supervector M(i) is the superposition of all D-dimensional mean vectors m k superposition.

[0089] Given Y i , θ and T, the solution formula for the I-vector is as follows:

[0090]

[0091]

[0092] where is the Kth D-dimensional supervector of M(i); is the optimal solution of w(i) that meets the experimental data requirements, making the data of the underwater acoustic signal maximally separable.

[0093] Specifically, the closed-form solution of the i-vector extraction formula can be as follows:

[0094]

[0095]

[0096]

[0097] where m is the mean that conforms to the trainable data set in the entire data set.

[0098] The universal background model is trained, and then the target underwater acoustic signal is input into the trained model to obtain the discriminative feature w(i). Among them, the discriminative feature w(i) is an F-dimensional random vector.

[0099] According to the ship identification method provided by the present invention, a general background model is used to reduce the dimension and extract features of the target underwater acoustic signal, so as to obtain discriminant features convenient for classification, providing a basis for the dimensionality reduction decomposition of the discriminant features and the identification of ships.

[0100] The predefined length vectors extracted from each underwater acoustic signal can be used as the input of a standard pattern recognition algorithm. Before being passed to the input layer of the ship identification model, a PLDA model can also be used to provide a probability framework suitable for the predefined length input vectors.

[0101] Optionally, the dimensionality reduction processing of the discriminant features to obtain a signal vector includes:

[0102] Based on the probabilistic linear discriminant analysis model, the discriminant features are decomposed for dimensionality reduction to obtain a dimensionality reduction vector;

[0103] The dimensionality reduction vector includes a signal vector and a noise vector;

[0104] The probabilistic linear discriminant analysis model is described based on the conditional probability of the Gaussian distribution.

[0105] Multiple channels in the sonar array simultaneously collect underwater acoustic signals of the target ship, and the D-dimensional feature vector Yi extracted from the i-th channel is given i and the set w(i) of its i vectors.

[0106] Optionally, the discriminant features are decomposed for dimensionality reduction based on the following formula to obtain the dimensionality reduction vector:

[0107] w(i) = μ + Fh i + Gs i + ∈ i ;

[0108] where w(i) is the discriminant feature, and i is the number of channels for collecting the target underwater acoustic signal; μ + Fh i + Gs i + ∈ i is the dimensionality reduction vector, μ + Fh i is the signal vector, Gs i + ∈ i is the noise vector;

[0109] μ is the mean of all input training data; F is the underwater acoustic category space; h i is the position of the underwater acoustic signal in the entire underwater acoustic category space; G is the error space; s i is the position of the underwater acoustic signal in the error space; ∈ i is the residual signal noise term.

[0110] The underwater acoustic category space F contains information that can be used to represent various underwater acoustic categories; h i can be regarded as a specific underwater acoustic category or the position of this section of the underwater acoustic signal in the entire underwater acoustic category space; the error space G is used to represent the change information of the underwater acoustic signals of the same category under different states and environments

[0111] Among them, the dimensions of F and G need to be preset in advance and form a parameter set Φ = {μ, F, G, ∑}, and ∑ is defined as a Gaussian with diagonal covariance.

[0112] The signal vector μ + Fh i , depends only on Y i the identity of the category, and not on the specific acoustics; the noise vector Gs i +∈ i , is different for each acoustics of the individual and represents the in-individual noise, that is, the ocean environmental background noise.

[0113] The PLDA model is described using the conditional probability based on the Gaussian distribution as follows:

[0114] P r (w(i)|h i ,s i ,Φ) = N(μ + Fh i + Gs i ,∑);

[0115] P r (h i ) = N(0, I);

[0116] P r (s i ) = N(0, I);

[0117] Among them, preset values can be defined for the latent variable h i and the latent variable s i .

[0118] According to the ship identification method provided by the present invention, by separately modeling the ship signal and the noise signal, the high-dimensional sequence input data I-vector is reduced to a low-dimensional feature vector with a predefined length, and the noise background model is extracted to eliminate the influence brought by the poor environment while retaining most of the relevant information.

[0119] Optionally, the ship identification model is constructed based on a long short-term memory network;

[0120] Before inputting the signal vector into the ship identification model, the method further includes:

[0121] Obtain a plurality of underwater acoustic signal samples and the corresponding ship type labels for each underwater acoustic signal sample;

[0122] Obtain the sample feature vector corresponding to each underwater acoustic signal sample;

[0123] Perform dimensionality reduction processing on each sample feature vector to obtain the signal vector sample corresponding to each underwater acoustic signal sample;

[0124] Take the combination of the signal vector sample corresponding to each underwater acoustic signal sample and the ship type label as a training sample, obtain multiple training samples, and use the multiple training samples to train a preset recognition model.

[0125] Long short-term memory is a special structure of a recurrent neural network, which can solve sequences with time dependence, has the ability of global processing, and there is a certain time span in the element-level correspondence between input and output. To a certain extent, it can avoid the long-dependence problem. The default behavior of different time-scale memory information is for long and short time series, and there is no need to specifically design different gate structures.

[0126] Use a long short-term memory network (LSTM) to model the signal vector, which can find the characteristics and trends of variable changes from the time dimension, and ensure the long-term retention effect on the input vector of a predefined length. The preset recognition model is constructed based on LSTM.

[0127] Obtain multiple underwater acoustic signal samples and the ship type label corresponding to each underwater acoustic signal sample, that is, the ship type corresponding to each underwater acoustic signal sample is known and has been labeled by the ship type label.

[0128] On this basis, obtain the sample feature vector corresponding to each underwater acoustic signal sample; perform dimensionality reduction decomposition on each sample feature vector to obtain the signal vector sample corresponding to each underwater acoustic signal sample.

[0129] Furthermore, take the combination of the signal vector sample corresponding to each underwater acoustic signal sample and the ship type label as a training sample, that is, take each signal vector sample with a ship type label as a training sample, and thus multiple training samples can be obtained. After obtaining multiple training samples, then input the multiple training samples into the ship recognition model in sequence, that is, input the signal vector sample and the ship type label in each training sample into the ship recognition model at the same time, and adjust the model parameters in the ship recognition model according to each output result of the ship recognition model, and finally complete the training process of the ship recognition model.

[0130] According to the ship identification method provided by the present invention, the ship identification model is trained based on the idea of deep learning, so that the ship identification model learns the characteristics of the signal vectors corresponding to different ship types, which is conducive to using the trained ship identification model to identify the ship types of underwater acoustic signals.

[0131] Optionally, the training of the preset identification model using multiple training samples includes:

[0132] For any one training sample, input the training sample into the ship identification model, and output the prediction probability corresponding to the training sample;

[0133] Use a preset loss function to calculate the loss value according to the prediction probability corresponding to the training sample and the ship type label in the training sample;

[0134] If the loss value is less than the preset threshold, the training of the preset identification model is completed, and the ship identification model is obtained.

[0135] The preset loss function can be expressed as:

[0136]

[0137] where W is the parameter for model training; b is the offset vector, which can prevent overfitting; y is the actual label; is the ship type obtained through the ship identification model.

[0138] The preset threshold is set to 0.5. The preset identification model after training is used as the ship identification model, and the accuracy rate of ship type classification can reach more than 70%.

[0139] According to the ship identification method provided by the present invention, by training the ship identification model, it is beneficial to control the loss value of the ship identification model within a preset range, thereby facilitating the improvement of the accuracy of the ship identification model for ship type identification.

[0140] Figure 2 is a schematic structural diagram of the ship identification device provided by the present invention, as Figure 2 shown, including but not limited to:

[0141] An acquisition module 201, configured to acquire the discrimination features of the target underwater acoustic signal;

[0142] A dimensionality reduction module 202, configured to perform dimensionality reduction processing on the discrimination features to obtain a signal vector;

[0143] A determination module 203, configured to input the signal vector into the ship identification model to determine the ship type corresponding to the target underwater acoustic signal.

[0144] During the operation of the device, the acquisition module 201 acquires the discriminative features of the target underwater acoustic signal; the dimensionality reduction module 202 performs dimensionality reduction processing on the discriminative features to obtain a signal vector; the determination module 203 inputs the signal vector into a ship recognition model to determine the ship type corresponding to the target underwater acoustic signal.

[0145] First, the acquisition module 201 acquires the discriminative features of the target underwater acoustic signal.

[0146] A sonar array can be used to collect signals from target ships within the acquisition range to obtain target underwater acoustic signals. Among them, the target underwater acoustic signal can include the ship signal of the target ship and the ocean noise signal. The ship signal comes from the mechanical noise, propeller noise, and hydrodynamic noise of the target ship; the ocean noise signal comes from the ocean environment.

[0147] The discriminative feature can be an I-vector. The I-vector is a fixed-length low-dimensional vector with good direction discrimination.

[0148] Specifically, GMM-UBM is used to extract vectors from the target underwater acoustic signal to obtain discriminative features.

[0149] Furthermore, the dimensionality reduction module 202 performs dimensionality reduction processing on the discriminative features to obtain a signal vector.

[0150] For the sound source target of the target ship, PLDA can be used to perform horizontal batch processing on the discriminative features corresponding to the target underwater acoustic signals collected by the sonar array through different signal channels, which can eliminate channel differences to a certain extent; then PLDA is used to perform dimensionality reduction decomposition on the discriminative features to obtain a signal vector and a noise vector, realizing the separation of the ship signal and the ocean noise signal.

[0151] Furthermore, the determination module 203 inputs the signal vector into a ship recognition model to determine the ship type corresponding to the target underwater acoustic signal.

[0152] The obtained signal vector is input into the ship recognition model. The ship recognition model recognizes the signal vector to obtain the classification result output by the ship recognition model. According to the classification result, the ship type corresponding to the target underwater acoustic signal can be obtained. In the same sea area, the ship recognition method provided by the present invention has strong robustness for ship type recognition and can effectively handle the cold start problem of ship classification.

[0153] The ship types can include multiple types such as fishing boats, law enforcement ships, cargo ships, container ships, speedboats, passenger ships, search and rescue ships, tugboats, and sailboats.

[0154] The ship identification device provided by the present invention reduces the dimensionality of the discriminative features of the target underwater acoustic signal to reduce redundant features or noise data, effectively reducing the computational load of the ship identification model and the memory space occupied by data storage, improving the identification speed of the model, providing a basis for the real-time monitoring of ships, and at the same time reducing the influence of channel interference to a certain extent in the process of collecting the underwater acoustic signals of ships.

[0155] Optionally, the obtaining module is specifically configured to:

[0156] Based on the maximum likelihood estimation method, obtain a universal background model by using the target underwater acoustic signal;

[0157] Based on the maximum a posteriori probability of the universal background model, obtain the mean supervector of the target underwater acoustic signal;

[0158] Obtain the discriminative features according to the mean supervector.

[0159] Among them, the universal background model can be GMM-UBM.

[0160] Specifically, given the target underwater acoustic signal as: Where is a string of D-dimensional feature vectors extracted from the i-th channel.

[0161] For the target underwater acoustic signal A universal background can be obtained by training based on the target underwater acoustic signal using the maximum likelihood estimation (MLE) method, as follows:

[0162]

[0163] Where c k is the mixing coefficient, and N(·; m k , R k ) is jointly represented by the D-dimensional mean vector m k and the D×D diagonal covariance matrix R k , which is a description symbol representing the normal distribution of the observable data set y.

[0164] We can obtain θ = {c k , m k , R k |k = 1,..., K} as a set of parameters of the universal background model.

[0165] Assume that a certain type of underwater acoustic signal Y i , can be represented by the mean supervector M(i) of the signal and the block component R kIt is represented by a D·K×D·K block-diagonal matrix R0 formed.

[0166] After that, using the maximum a posteriori probability of GMM-UBM, the mean supervector M(i) describing the target underwater acoustic signal is obtained as follows:

[0167] M(i) = M0 + Tw(i);

[0168] where M0 is the m of the general background model k concatenated into a D·K-dimensional supervector; T is a low-rank matrix of D·K×F (F << D·K) dimensions, called the total variability matrix; w(i) is an F-dimensional random vector with a prior distribution that satisfies the standard normal distribution N(·; 0, I).

[0169] where the mean supervector M(i) is the superposition of all D-dimensional mean vectors m k of.

[0170] Given Y i , θ and T, the solution formula for the I-vector is as follows:

[0171]

[0172]

[0173] where is the K-th D-dimensional supervector of M(i); is the optimal solution of w(i) that meets the experimental data requirements, making the data of the underwater acoustic signal maximally separable.

[0174] Specifically, the closed-form solution of the i-vector extraction formula can be as follows:

[0175]

[0176]

[0177]

[0178] where m is the mean that conforms to the trainable data set under the entire data set.

[0179] Train the general background model, and then input the target underwater acoustic signal into the trained model to obtain the discriminative feature w(i).

[0180] where the discriminative feature w(i) is an F-dimensional random vector.

[0181] According to the ship identification device provided by the present invention, a general background model is used to reduce the dimension and extract features of the target underwater acoustic signal, so as to obtain discriminative features convenient for classification, providing a basis for the dimensionality reduction decomposition of the discriminative features and the identification of ships.

[0182] It should be noted that the ship identification device provided by the embodiments of the present invention can be implemented based on the ship identification method described in any of the above embodiments when specifically executed, and this embodiment will not be elaborated here.

[0183] Figure 3 is a schematic structural diagram of the electronic device provided by the present invention. As Figure 3 shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330, and a communication bus 340. Among them, the processor 310, the communication interface 320, and the memory 330 communicate with each other through the communication bus 340. The processor 310 can call the logical instructions in the memory 330 to execute the ship identification method, and the method includes: obtaining the discriminative features of the target underwater acoustic signal; performing dimensionality reduction processing on the discriminative features to obtain a signal vector; inputting the signal vector into the ship identification model to determine the ship type corresponding to the target underwater acoustic signal.

[0184] In addition, when the logical instructions in the above-mentioned memory 330 are implemented in the form of software functional units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs that can store program codes.

[0185] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the ship identification method provided by each of the above methods, and the method includes: obtaining the discrimination features of a target underwater acoustic signal; performing dimensionality reduction processing on the discrimination features to obtain a signal vector; and inputting the signal vector into a ship identification model to determine the ship type corresponding to the target underwater acoustic signal.

[0186] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the ship identification method provided by each of the above embodiments, and the method includes: obtaining the discrimination features of a target underwater acoustic signal; performing dimensionality reduction processing on the discrimination features to obtain a signal vector; and inputting the signal vector into a ship identification model to determine the ship type corresponding to the target underwater acoustic signal.

[0187] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0188] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0189] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in each of the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present invention.

Claims

1. A ship identification method, characterized in that, Including: Obtain the discriminant features of the target underwater acoustic signal; Perform dimensionality reduction processing on the discriminant features to obtain a signal vector; Input the signal vector into a ship recognition model to determine the ship type corresponding to the target underwater acoustic signal; The obtaining of the discriminant features of the target underwater acoustic signal includes: Based on the maximum likelihood estimation method, use the target underwater acoustic signal to obtain a general background model; Based on the maximum a posteriori probability of the general background model, obtain the mean supervector of the target underwater acoustic signal; According to the mean supervector, obtain the discriminant features; Wherein, the general background model is a GMM-UBM model, and the target underwater acoustic signal is a string of D-dimensional feature vectors extracted from the i-th channel, and its expression is: Among them, The specific calculation formula for obtaining the general background model by using the target underwater acoustic signal based on the maximum likelihood estimation method is: Among them, c k is the mixing coefficient; N(y; m k , R k ) is the joint representation of the D-dimensional mean vector m k and the D×D diagonal covariance matrix R k , which is the descriptive symbol of the normal distribution representing the set of observable data y; It is calculated that θ = {c k , m k , R k | k = 1, ..., K}, as a set of parameters of the general background model; One type of underwater acoustic signal Y i is represented by the D·K×D·K block diagonal matrix R0 composed of the mean supervector M(i) and the block component R k to obtain the maximum a posteriori probability using the universal background model, and the expression for the mean supervector M(i) of the target underwater acoustic signal is obtained as follows: M(i) = M0 + Tw(i); Among them, M0 is composed of m of the general background model k concatenated into a D·K-dimensional supervector; T is a low-rank matrix of dimension D·K×F (F << D·K); w(i) is an F-dimensional random vector with a prior distribution that satisfies the standard normal distribution N(·; 0, I); the mean supervector M(i) is the superposition of all D-dimensional mean vectors m k ; The obtaining of the discriminant features according to the mean supervector specifically includes: Given Y i , θ, and T, the solution formula for the I-vector is as follows: Among them, is the K-th D-dimensional supervector of the mean supervector M(i); is the optimal solution for w(i) to meet the requirements of experimental data, making the data of the target underwater acoustic signal maximally separable; The closed-form solution of the I-vector extraction formula is specifically: Wherein, m is the mean that conforms to the trainable data set in the entire data set; Train the general background model, and then input the target underwater acoustic signal into the trained general background model to obtain the discriminant feature w(i); wherein, the discriminant feature w(i) is an F-dimensional random vector.

2. The ship identification method according to claim 1, characterized in that, The performing of dimensionality reduction processing on the discriminant features to obtain a signal vector includes: Based on the probabilistic linear discriminant analysis model, perform dimensionality reduction decomposition on the discriminant features to obtain a dimensionality reduction vector; The dimensionality reduction vector includes a signal vector and a noise vector; The probabilistic linear discriminant analysis model is described based on the conditional probability of the Gaussian distribution.

3. The ship identification method according to claim 1, characterized in that, The ship recognition model is constructed based on a long short-term memory network; Before inputting the signal vector into the ship recognition model, the method further includes: Obtain multiple underwater acoustic signal samples and the ship type labels corresponding to each underwater acoustic signal sample; Obtain the sample feature vector corresponding to each underwater acoustic signal sample; Perform dimensionality reduction processing on each sample feature vector to obtain the signal vector sample corresponding to each underwater acoustic signal sample; Use the combination of the signal vector sample corresponding to each underwater acoustic signal sample and the ship type label as a training sample, obtain multiple training samples, and use the multiple training samples to train a preset recognition model.

4. The ship identification method according to claim 3, wherein, The training of the preset recognition model by using multiple training samples includes: For any one training sample, input the training sample into the preset recognition model and output the prediction probability corresponding to the training sample; Use a preset loss function to calculate a loss value according to the prediction probability corresponding to the training sample and the ship type label in the training sample; If the loss value is less than a preset threshold, the training of the preset recognition model is completed, and the ship recognition model is obtained.

5. The ship identification method according to claim 2, wherein, Perform dimensionality reduction decomposition on the discriminant features based on the following formula to obtain the dimensionality reduction vector: w(i) = μ + Fh i + Gs i + ∈ i ; Among them, w(i) is the discrimination feature, and i is the number of channels for collecting the target underwater acoustic signal; μ + Fh i + Gs i + ∈ i is the dimensionality reduction vector, μ + Fh i is the signal vector, Gs i + ∈ i is the noise vector; μ is the mean of all input training data; F is the underwater acoustic category space; h i is the position of the underwater acoustic signal in the entire underwater acoustic category space; G is the error space; s i is the position of the underwater acoustic signal in the error space; ∈ i is the residual signal noise term.

6. A ship identification device, wherein, Including: An obtaining module, configured to obtain the discriminant features of the target underwater acoustic signal; A dimensionality reduction module, configured to perform dimensionality reduction processing on the discriminant features to obtain a signal vector; A determination module, configured to input the signal vector into a ship recognition model to determine the ship type corresponding to the target underwater acoustic signal; The obtaining of the discriminative features of the target underwater acoustic signal includes: Based on the maximum likelihood estimation method, using the target underwater acoustic signal to obtain a universal background model; Based on the maximum a posteriori probability of the universal background model, obtaining the mean supervector of the target underwater acoustic signal; According to the mean supervector, obtaining the discriminative features; Wherein, the universal background model is a GMM-UBM model, and the target underwater acoustic signal is a string of D-dimensional feature vectors extracted from the i-th channel, and its expression is: Among them, The specific calculation formula for obtaining the universal background model by using the target underwater acoustic signal based on the maximum likelihood estimation method is: where c k is the mixing coefficient; N(y; m k , R k ) is the joint representation of the D-dimensional mean vector m k and the D×D diagonal covariance matrix R k , which is the descriptive symbol of the normal distribution representing the set of observable data y; Calculated to obtain θ = {c k , m k , R k | k = 1, ..., K}, as a set of parameters of the general background model; One type of underwater acoustic signal Y i is represented by the D·K×D·K block diagonal matrix R0 composed of the mean supervector M(i) and the block component R k To obtain the maximum a posteriori probability using the general background model, the expression for the mean supervector M(i) of the target underwater acoustic signal is: M(i) = M0 + Tw(i); where M0 is m of the general background model k concatenated into a D·K-dimensional supervector; T is a low-rank matrix of dimension D·K×F (F << D·K); w(i) is an F-dimensional random vector with a prior distribution that satisfies the standard normal distribution N(·; 0, I); the mean supervector M(i) is the superposition of all D-dimensional mean vectors m k ; The obtaining of the discriminative features according to the mean supervector specifically includes: Given Y i , θ, and T, the solution formula for the I-vector is as follows: Among them, is the K-th D-dimensional supervector of the mean supervector M(i); is the optimal solution for w(i) to meet the requirements of experimental data, making the data of the target underwater acoustic signal maximally separable; The closed-form solution of the I-vector extraction formula is specifically: Wherein, m is the mean that conforms to the trainable data set in the entire data set; Training the universal background model, and then inputting the target underwater acoustic signal into the trained universal background model to obtain the discriminative feature w(i); wherein, the discriminative feature w(i) is an F-dimensional random vector.

7. An electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, When the processor executes the computer program, the steps of the ship recognition method according to any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the ship recognition method according to any one of claims 1 to 5 are implemented.

Images