An underwater acoustic multi-target identification method, device and computer readable storage medium

Abstract

This invention relates to a method, apparatus, and computer-readable storage medium for underwater acoustic multi-target identification, belonging to the field of underwater target identification technology. The invention includes: calculating the detection features of an underwater acoustic audio signal, and inputting the detection feature vector into a first classifier for classification to determine whether a corresponding underwater acoustic target exists in the audio signal; if so, calculating the identification features of the audio signal and inputting the identification feature vector into a second classifier to determine the type of underwater acoustic target corresponding to the audio signal. This invention can directly perform feature calculation on the underwater acoustic audio signal, offering high real-time performance and fast response; and by analyzing the underwater acoustic audio signal from multiple angles to obtain detection and identification features, it reduces information loss during feature extraction; furthermore, it divides underwater acoustic target identification into two stages: detection and identification. First, the underwater acoustic target is detected based on the detection features, and then the underwater acoustic target type is identified based on the identification features, improving the accuracy of underwater acoustic target identification.

Description

Technical Field

[0001] This invention relates to the field of underwater target recognition technology, and in particular to an underwater acoustic multi-target recognition method, apparatus, and computer-readable storage medium. Background Technology

[0002] With the continuous development and increasing intelligence of underwater equipment systems, the importance of underwater acoustic target recognition technology has significantly increased in both military and commercial applications. Underwater acoustic target recognition technology identifies different types of targets by analyzing passive sonar signals. Traditional underwater acoustic target recognition mainly relies on human observers analyzing the signals received by sonar in real time to identify the target type. Later, through the application of acoustic signal theory and modern spectral theory, the accuracy and efficiency of recognition have been improved to some extent. However, with the substantial increase in the amount of underwater acoustic information, continuing to use traditional underwater acoustic target recognition methods has become increasingly difficult and inadequate. Therefore, intelligent underwater acoustic target recognition methods have emerged. Intelligent underwater acoustic target recognition methods analyze and extract features from raw acoustic data. These features are then input into an intelligent algorithm to train a classifier, achieving automatic target recognition. This process mainly includes two parts: feature extraction and classifier design.

[0003] In existing technologies, two different directions have emerged in intelligent development regarding feature extraction and classifier design: feature-based methods and model-based methods. Feature-based methods analyze the original signal to capture significant differences in the signal. Effective features not only simplify classifier design but also directly impact recognition accuracy. However, feature extraction is essentially a dimensionality reduction process of the original data, inevitably resulting in information loss, which significantly limits recognition accuracy in complex marine environments. For example, the patent CN201910661350, which describes a big data-based intelligent underwater acoustic target recognition method, collects a large number of underwater acoustic signals emitted by underwater acoustic targets, extracts appropriate features, and then labels the data as either target-oriented or targetless, establishing a training sample set to train an artificial intelligence model. In practical applications, collecting surrounding underwater acoustic signals and extracting features, then inputting them into the trained artificial intelligence model, can determine the presence and type of underwater acoustic targets, enabling the identification of underwater acoustic targets at longer distances. However, this method can only determine the presence or absence of underwater acoustic targets and cannot achieve target type recognition in multi-target situations. Model-based methods focus on applying complex intelligent classification algorithms to underwater acoustic target recognition, such as Convolutional Neural Networks (CNNs), Deep Belief Networks (DBNs), and Sparse Autoencoders (SAEs). With sufficient data, these methods can automatically extract valuable features from training data without requiring expert knowledge. However, in underwater acoustic target recognition, the amount of available data is often insufficient for training, leading to inadequate model training or overfitting. For example, the patent CN202110753982, "U-Net Structure Generative Adversarial Network and Method for Underwater Acoustic Target Recognition," establishes a U-Net-based generative adversarial network model for underwater target recognition. This model effectively eliminates overfitting in deep learning networks with small sample data. By increasing the number of network layers, overfitting is eliminated, improving the recognition rate of underwater acoustic signals and increasing the robustness of the network model. However, solving overfitting by increasing the number of deep learning network layers increases network complexity and real-time performance, resulting in slower response times and higher computational resource consumption, failing to meet the requirements of miniaturization and real-time performance. For example, in patent CN202010427316, a method for underwater acoustic target recognition based on weighted support vector machines (SVM) is proposed. This method involves extracting features from the original underwater acoustic target signal in frames, building a weighted SVM recognition model, using a network search method to find the optimal kernel function parameters and penalty factor, and training the SVM model to obtain an SVM model with underwater target recognition capabilities, achieving a recognition accuracy of over 80%. However, single-stage feature extraction is difficult, the extracted underwater acoustic target features have poor discriminative power, and the recognition accuracy is low. Furthermore, model-based methods lack physical interpretability, cannot guarantee stability, and fail to provide observers with useful auxiliary decision-making information.Therefore, model-based methods cannot be directly used for underwater target identification.

[0004] In summary, existing methods for underwater acoustic target recognition suffer from poor real-time performance, slow response speed, difficulty in feature extraction, and low recognition accuracy. Summary of the Invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the problems of poor real-time performance, slow response speed, high difficulty in feature extraction, and low recognition accuracy in the prior art for underwater acoustic target recognition.

[0006] To address the aforementioned technical problems, this invention provides a method for underwater acoustic multi-target identification, comprising:

[0007] Acquire underwater acoustic audio signals from the training set, and acquire the time-domain and frequency-domain signals of the underwater acoustic audio signals;

[0008] The detection features of the underwater acoustic audio signal are calculated based on the time-domain and frequency-domain signals of the underwater acoustic audio signal to obtain a detection feature vector; wherein, the detection features include, but are not limited to, the time-domain features, frequency-domain features, and time-frequency features of the underwater acoustic audio signal;

[0009] The detected feature vector is input into the first classifier for training to obtain the trained first classifier;

[0010] The identification features of the underwater acoustic audio signal are calculated based on the time-domain and frequency-domain signals of the underwater acoustic audio signal to obtain an identification feature vector; wherein, the identification features include, but are not limited to, the time-domain features, frequency-domain features, and time-frequency features of the underwater acoustic audio signal that are not included in the detection features;

[0011] The identification feature vector is input into the second classifier for training to obtain the trained second classifier.

[0012] A multi-target underwater acoustic recognition model is constructed based on the trained first classifier and the trained second classifier.

[0013] In one embodiment of the present invention, acquiring the underwater acoustic audio signal and acquiring the time-domain signal and frequency-domain signal of the underwater acoustic audio signal includes:

[0014] Acquire underwater acoustic audio signals and perform frame segmentation processing on the underwater acoustic audio signals;

[0015] The time-domain signal and frequency-domain signal of each frame of underwater acoustic audio signal are obtained respectively to obtain the time-domain signal set and frequency-domain signal set of the underwater acoustic audio signal.

[0016] In one embodiment of the present invention, the step of calculating the detection features of the underwater acoustic signal based on the time-domain and frequency-domain signals of the underwater acoustic signal to obtain the detection feature vector includes:

[0017] Calculate the temporal characteristics of each frame of the time-domain signal in the time-domain signal set of the underwater acoustic audio signal;

[0018] The time-domain features of each frame of the time-domain signal are statistically analyzed using statistical functions to obtain the time-domain feature vector of the underwater acoustic audio signal; wherein, the time-domain features include, but are not limited to, zero-crossing rate, short-time energy, and short-time autocorrelation function;

[0019] Calculate the frequency domain characteristics of each frame of the frequency domain signal in the frequency domain signal set of the underwater acoustic audio signal;

[0020] The frequency domain features of each frame of the frequency domain signal are statistically analyzed using statistical functions to obtain the frequency domain feature vector of the underwater acoustic audio signal; wherein, the frequency domain features include, but are not limited to, Mel-frequency cepstral coefficients, spectral width, and spectral difference amplitude.

[0021] The time-domain feature vector and the frequency-domain feature vector are combined to form the detection feature vector.

[0022] In one embodiment of the present invention, the step of calculating the recognition features of the underwater acoustic audio signal based on the time-domain and frequency-domain signals of the underwater acoustic audio signal to obtain the recognition feature vector includes:

[0023] Based on the time-domain signal set and the frequency-domain signal set of the underwater acoustic audio signal, the time-frequency characteristics of each frame of the underwater acoustic audio signal are calculated respectively.

[0024] The video features of each frame of underwater acoustic audio signal are statistically analyzed using statistical functions to obtain a time-frequency feature vector, which is then used as the recognition feature vector.

[0025] The time-frequency characteristics include, but are not limited to, spectral skewness coefficient, spectral kurtosis, spectral sharpness, and spectral centroid.

[0026] In one embodiment of the present invention, the statistical function includes, but is not limited to, the maximum value function, the minimum value function, the standard deviation function, and the average value function.

[0027] In one embodiment of the present invention, the step of inputting the detected feature vector into a first classifier for training to obtain a trained first classifier includes:

[0028] The detected feature vector is input into the first classifier for training until the cross-entropy loss function of the binary classification task reaches its minimum, thus obtaining the trained first classifier.

[0029] The cross-entropy loss function for the binary classification task is:

[0030]

[0031] in, y is the probability that the predicted result is a positive sample, and y is the sample label, which is 1 for positive samples and 0 for negative samples.

[0032] In one embodiment of the present invention, the step of inputting the recognition feature vector into a second classifier for training to obtain a trained second classifier includes:

[0033] The identified feature vector is input into the second classifier for training until the cross-entropy loss function of the multi-class task reaches its minimum, thus obtaining the trained second classifier.

[0034] The cross-entropy loss function for the multi-class classification task is:

[0035]

[0036] Where K is the number of categories, and y represents the sample label. If the sample category is i, then y i =1, otherwise, y i =0; p i This represents the probability of predicting the outcome category i.

[0037] In one embodiment of the present invention, after constructing the underwater acoustic multi-target recognition model based on the trained first classifier and the trained second classifier, the method further includes:

[0038] Acquire underwater acoustic audio signals from the test set, calculate the detection feature vector of the underwater acoustic audio signals, and input the detection feature vector into the first classifier in the underwater acoustic multi-target recognition model for classification;

[0039] Based on the classification results, determine whether the underwater acoustic audio signal has a corresponding underwater acoustic target. If it does, then the underwater acoustic audio signal is taken as the target underwater acoustic audio signal.

[0040] The identification feature vector of the target underwater acoustic audio signal is calculated, and the identification feature vector is input into the second classifier in the underwater acoustic multi-target identification model for classification to determine the underwater acoustic target type corresponding to the target underwater acoustic audio signal.

[0041] The present invention also provides an underwater acoustic multi-target identification device, comprising:

[0042] The signal acquisition module is used to acquire underwater acoustic audio signals in the training set, and to acquire the time-domain signal and frequency-domain signal of the underwater acoustic audio signals;

[0043] The detection feature extraction module is used to calculate the detection features of the underwater acoustic audio signal based on the time-domain signal and the frequency-domain signal of the underwater acoustic audio signal, and obtain the detection feature vector; wherein, the detection features include, but are not limited to, the time-domain features, frequency-domain features, and time-frequency features of the underwater acoustic audio signal;

[0044] The first classification module is used to input the detection feature vector into the first classifier for training to obtain a trained first classifier.

[0045] The identification feature extraction module is used to calculate the identification features of the underwater acoustic audio signal based on the time-domain signal and the frequency-domain signal of the underwater acoustic audio signal, and obtain the identification feature vector; wherein, the identification features include, but are not limited to, the time-domain features, frequency-domain features, and time-frequency features of the underwater acoustic audio signal that are not included in the detection features;

[0046] The second classification module is used to input the recognition feature vector into the second classifier for training, so as to obtain the trained second classifier.

[0047] The model building module is used to build an underwater acoustic multi-target recognition model based on the trained first classifier and the trained second classifier.

[0048] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described underwater acoustic multi-target recognition method.

[0049] The underwater acoustic multi-target recognition method of this invention includes: calculating the detection features of the underwater acoustic audio signal to obtain a detection feature vector, and inputting the detection feature vector into a first classifier for classification to determine whether the underwater acoustic audio signal has a corresponding underwater acoustic target. If it does, the method calculates the recognition features of the underwater acoustic audio signal to obtain a recognition feature vector, and inputs the recognition feature vector into a second classifier to identify the type of underwater acoustic target corresponding to the underwater acoustic audio signal based on the classification result. This invention can directly perform feature calculation on the underwater acoustic audio signal, has high real-time performance, and fast response speed. Furthermore, both the detection features and recognition features include, but are not limited to, the time-domain features, frequency-domain features, and time-frequency features of the underwater acoustic audio signal. By analyzing the underwater acoustic audio signal from multiple angles, detection features and recognition features are obtained, reducing information loss during feature extraction. In addition, by first detecting the underwater acoustic target based on the detection features and then recognizing the underwater acoustic target based on the recognition features, the accuracy of underwater acoustic target recognition is improved. Attached Figure Description

[0050] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein...

[0051] Figure 1 This is a schematic diagram of the underwater acoustic multi-target recognition principle provided by the present invention;

[0052] Figure 2 This is a flowchart of the underwater acoustic multi-target recognition method provided by the present invention;

[0053] Figure 3 This is a schematic diagram illustrating the training and testing of the underwater acoustic multi-target recognition model provided by the present invention;

[0054] Figure 4 This is a schematic diagram of the underwater acoustic signal feature extraction strategy provided by the present invention;

[0055] Figure 5 This is a schematic diagram of the underwater acoustic multi-target recognition device provided by the present invention. Detailed Implementation

[0056] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0057] Example 1:

[0058] The underwater acoustic multi-target recognition method provided in this application mainly includes two stages. The first stage is the target detection stage, which generates a detection feature set (TSHD) for detecting the presence of underwater acoustic targets. The second stage generates a recognition feature set (TSHR) for identifying the type of underwater acoustic target. Figure 1 The diagram shown illustrates the principle of underwater acoustic multi-target recognition provided in this application. When an underwater acoustic audio signal to be detected is acquired, the TSHD feature set of the underwater acoustic audio signal is first extracted and input into the classifier to determine whether an underwater target exists. If an underwater target exists, the TSHR feature set of the underwater acoustic audio signal is re-extracted and input into the classifier to identify the type of underwater target.

[0059] like Figure 2 The flowchart shown is a diagram of the underwater acoustic multi-target identification method provided by the present invention, including:

[0060] S10: Obtain the underwater acoustic audio signals from the training set, and obtain the time-domain and frequency-domain signals of the underwater acoustic audio signals.

[0061] In some embodiments, step S10 is specifically implemented as follows: acquiring underwater acoustic audio signals from the training set, performing frame-by-frame processing on them, and acquiring the time-domain signal and frequency-domain signal of each frame of underwater acoustic audio signal to obtain a time-domain signal set and a frequency-domain signal set.

[0062] For example, in this embodiment, the underwater acoustic audio signal is divided into several frames with a duration of one second. This allows for better extraction of the features of each frame of the underwater acoustic audio signal, resulting in more comprehensive feature extraction and improved model recognition accuracy.

[0063] Alternatively, the frequency domain signal of the underwater acoustic audio signal can be obtained using the Fourier Transform (FFT) method.

[0064] S20: Calculate the detection characteristics of the underwater acoustic audio signal based on the time-domain and frequency-domain signals of the underwater acoustic audio signal, and obtain the detection feature vector.

[0065] Specifically, the detection features include, but are not limited to, the time-domain features, frequency-domain features, and time-frequency features of the underwater acoustic audio signal.

[0066] In some embodiments, step S20 is specifically implemented as follows:

[0067] S200: Calculate the time-domain characteristics of the time-domain signal for each frame.

[0068] Optionally, the time-domain characteristics include, but are not limited to, the zero-crossing rate, short-time energy, and short-time autocorrelation function of the time-domain signal.

[0069] S201: Use statistical functions to statistically analyze the temporal characteristics of each frame of the time-domain signal to obtain the temporal feature vector of the underwater acoustic audio signal.

[0070] For example, in this embodiment, the zero-crossing rate of the time-domain signal of each frame is calculated as a time-domain feature, and a time-domain feature vector is obtained by statistically analyzing the zero-crossing rate of the time-domain signal of each frame using a statistical function.

[0071] Specifically, the zero-crossing rate refers to the number of times an audio signal crosses a zero point (i.e., changes from positive to negative or from negative to positive), and its expression is:

[0072]

[0073] Where sgn[] is the sign function, N is the length of a frame, and n is the number of frames;

[0074]

[0075] In other embodiments of this application, the calculated time-domain features may also be the short-time energy, short-time autocorrelation function, or other time-domain parameter information of the time-domain signal, and this application does not limit this.

[0076] S202: Calculate the frequency domain characteristics of each frame of the frequency domain signal.

[0077] Optionally, frequency domain features include, but are not limited to, Mel-frequency cepstral coefficients, spectral width, and spectral difference amplitude.

[0078] S203: Use statistical functions to statistically analyze the frequency domain characteristics of each frame of the frequency domain signal to obtain the frequency domain feature vector of the underwater acoustic audio signal.

[0079] For example, in this embodiment, the Mel-Cepstral Coefficients of each frame of the frequency domain signal are calculated as frequency domain features, and statistical functions are used to statistically analyze the Mel-Cepstral Coefficients of each frame of the frequency domain signal to obtain a frequency domain feature vector.

[0080] Specifically, each frame of the spectrum signal is passed through a Mel filter to obtain the Mel spectrum. Cepstral analysis is then performed on the Mel spectrum (logarithm and inverse transform are taken; the inverse transform is usually implemented by DCT discrete cosine transform, and the 2nd to 13th coefficients after DCT are taken as Mel cepstral coefficients) to obtain the Mel cepstral coefficients.

[0081] In other embodiments of this application, the calculated frequency domain features may also be spectral width, spectral differential amplitude, or other frequency domain parameter information, which are not limited in this application.

[0082] S204: Combine the time-domain and frequency-domain feature vectors of the underwater acoustic audio signal as the detection feature vector.

[0083] In this embodiment, the time-domain and frequency-domain characteristics of the underwater acoustic audio signal are calculated as the detection features. In other embodiments, one or more of the time-domain characteristics, frequency-domain characteristics, time-frequency characteristics, and special audio features of the underwater acoustic audio signal can be selected as the detection features.

[0084] S30: Input the probe feature vector into the first classifier for training to obtain the trained first classifier.

[0085] In some embodiments, step S30 is specifically implemented as follows: the probe feature vector is input into the first classifier for training until the cross-entropy loss function of the binary classification task reaches its minimum, and the trained first classifier is obtained.

[0086] Specifically, the cross-entropy loss function for the binary classification task is:

[0087]

[0088] in, y is the probability that the predicted result is a positive sample, and y is the sample label, which is 1 for positive samples and 0 for negative samples.

[0089] For example, if the underwater acoustic audio signal has a corresponding underwater acoustic target, the sample label is a positive sample; if the underwater acoustic audio signal does not have a corresponding underwater acoustic target, the sample label is a negative sample.

[0090] S40: Calculate the recognition features of the underwater acoustic audio signal based on the time-domain and frequency-domain signals of the underwater acoustic audio signal, and obtain the recognition feature vector of the underwater acoustic audio signal.

[0091] Specifically, the identification features include, but are not limited to, the time-domain features, frequency-domain features, and time-frequency features of underwater acoustic signals not included in the detection features.

[0092] In some embodiments, step S40 is specifically implemented as follows: based on the time-domain signal set and frequency-domain signal set of the underwater acoustic audio signal, the time-frequency characteristics of each frame of the underwater acoustic audio signal are calculated respectively, and the time-frequency characteristics of each frame of the underwater acoustic audio signal are statistically analyzed using a statistical function to obtain a time-frequency feature vector, and the time-frequency feature vector is used as the recognition feature vector.

[0093] Optionally, the time-frequency characteristics include, but are not limited to, spectral skewness coefficient, spectral kurtosis, spectral sharpness, and spectral centroid.

[0094] In other embodiments of this application, one or more of the spectral skewness coefficient, spectral kurtosis, spectral sharpness, and spectral centroid may be used as time-frequency features, and this application does not limit this.

[0095] In this embodiment, the frequency domain features of the underwater acoustic audio signal are used as the identification features. In other embodiments, one or more of the time domain features, frequency domain features, time-frequency features, and special audio features not included in the detection features can also be selected as identification features.

[0096] S50: Input the recognition feature vector into the second classifier for training to obtain the trained second classifier.

[0097] In some embodiments, step S50 is specifically implemented as follows: the recognition feature vector is input into the second classifier for training until the cross-entropy loss function of the multi-class task reaches its minimum, and the trained second classifier is obtained.

[0098] Specifically, the cross-entropy loss function for multi-class classification tasks is:

[0099]

[0100] Where K is the number of categories, and y represents the sample label. If the sample category is i, then y i =1, otherwise, y i =0; p i This represents the probability of predicting the outcome category i.

[0101] S60: Construct an underwater acoustic multi-target recognition model based on the trained first classifier and the trained second classifier.

[0102] Optionally, the statistical functions in steps S20 and S40 can be maximum value functions, minimum value functions, standard deviation functions, average value functions, etc.

[0103] like Figure 3 The diagram shown illustrates the training and testing process of the underwater acoustic multi-target recognition model provided in this application.

[0104] Optionally, after constructing the underwater acoustic multi-target recognition model, this embodiment further includes testing the underwater acoustic multi-target recognition model using underwater acoustic audio signals from a test set. The specific process includes:

[0105] Acquire underwater acoustic audio signals from the test set, and calculate the detection feature vector of the underwater acoustic audio signal based on the time domain signal and frequency domain signal of the underwater acoustic audio signal. Input the detection feature vector into the first classifier in the underwater acoustic multi-target recognition model for classification.

[0106] Based on the classification results, determine whether there is a corresponding underwater acoustic target for the underwater acoustic audio signal. If there is, then the underwater acoustic audio signal is taken as the target underwater acoustic audio signal.

[0107] Based on the time-domain and frequency-domain signals of the target underwater acoustic audio signal, the identification feature vector of the target underwater acoustic audio signal is calculated. The identification feature vector is then input into the second classifier in the underwater acoustic multi-target identification model for classification, so as to determine the underwater acoustic target type corresponding to the target underwater acoustic audio signal based on the classification result.

[0108] The underwater acoustic multi-target recognition method provided in this application can directly calculate features from underwater acoustic audio signals without using feature extraction networks, resulting in higher real-time performance and faster response speed. By directly calculating the corresponding time-domain, frequency-domain, and time-frequency features from the time-domain and frequency-domain signals, the difficulty of feature extraction is reduced. Dividing underwater acoustic target recognition into two stages—target detection and target recognition—also improves the accuracy of underwater acoustic target recognition.

[0109] Example 2:

[0110] Based on Embodiment 1 above, this application proposes a strategy for extracting underwater acoustic signal features, such as... Figure 4 As shown, the corresponding time-domain features, frequency-domain features, and time-frequency features are extracted from the time-domain and frequency-domain signals of the underwater acoustic audio signal, respectively. These features are statistically analyzed using statistical functions to obtain the corresponding feature vectors. These feature vectors are then used as the basis for classifying the underwater acoustic audio signal by the classifier.

[0111] Based on this feature extraction strategy, this application embodiment performs feature extraction on a set of marine data, as shown in Table 1. This includes TSHD and TSHR. TSHD is a detection feature set. In this embodiment, TSHD includes time-domain and frequency-domain features of the underwater acoustic audio signal, which is used to detect whether the underwater acoustic audio signal has a corresponding underwater acoustic target. TSHR is a recognition feature set. In this embodiment, TSHR includes time-frequency features, which is used to identify the type of underwater acoustic target corresponding to the underwater acoustic audio signal.

[0112] Table 1

[0113]

[0114] Where ZCR is the zero-crossing rate, MFCC is the Mel-Cepstral coefficient, Spectral skewness is the spectral skewness coefficient, Spectral kurtosis is the spectral kurtosis, Spectral sharpness is the spectral sharpness, and Spectral centroid is the spectral centroid.

[0115] Statistical functions are used to statistically analyze the time-domain features, frequency-domain features, and time-frequency features in Table 1 to obtain the detection feature vector and the identification feature vector. Table 2 shows the statistical functions used in the embodiments of this application.

[0116] Table 2

[0117]

[0118] Different feature vectors can be obtained by using different statistical functions to perform statistics on the features. Therefore, TSHD contains 15 features (3*5) and TSHR contains 52 features (4*13).

[0119] This application also provides an underwater acoustic multi-target recognition device, such as... Figure 5 As shown, it includes:

[0120] The signal acquisition module 10 is used to acquire underwater acoustic audio signals in the training set, and to acquire the time-domain signal and frequency-domain signal of the underwater acoustic audio signals.

[0121] The detection feature extraction module 20 is used to calculate the detection features of the underwater acoustic audio signal based on the time-domain and frequency-domain signals of the underwater acoustic audio signal, and obtain the detection feature vector.

[0122] The first classification module 30 is used to input the detection feature vector into the first classifier for training to obtain the trained first classifier.

[0123] The feature extraction module 40 is used to calculate the recognition features of the underwater acoustic audio signal based on the time-domain and frequency-domain signals of the underwater acoustic audio signal, and obtain the recognition feature vector.

[0124] The second classification module 50 is used to input the recognition feature vector into the second classifier for training, so as to obtain the trained second classifier.

[0125] The model building module 60 is used to build an underwater acoustic multi-target recognition model based on the trained first classifier and the trained second classifier.

[0126] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described underwater acoustic multi-target recognition method.

[0127] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0128] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0129] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0130] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0131] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for underwater acoustic multi-target recognition, characterized in that, include: Acquire underwater acoustic audio signals from the training set, and acquire the time-domain and frequency-domain signals of the underwater acoustic audio signals; The detection features of the underwater acoustic audio signal are calculated based on the time-domain and frequency-domain signals of the underwater acoustic audio signal to obtain the detection feature vector; Specifically, it includes: Calculate the temporal characteristics of each frame of the time-domain signal in the time-domain signal set of the underwater acoustic audio signal; The time-domain features of each frame of the time-domain signal are statistically analyzed using statistical functions to obtain the time-domain feature vector of the underwater acoustic audio signal; wherein, the time-domain features include zero-crossing rate, short-time energy, and short-time autocorrelation function; Calculate the frequency domain characteristics of each frame of the frequency domain signal in the frequency domain signal set of the underwater acoustic audio signal; The frequency domain features of each frame of the frequency domain signal are statistically analyzed using statistical functions to obtain the frequency domain feature vector of the underwater acoustic audio signal; wherein, the frequency domain features include Mel-frequency cepstral coefficients, spectral width, and spectral difference amplitude; The time-domain feature vector and the frequency-domain feature vector are combined to form the detection feature vector; wherein, the detection features include the time-domain features, frequency-domain features, and time-frequency features of the underwater acoustic audio signal; The detected feature vector is input into the first classifier for training to obtain the trained first classifier; The recognition features of the underwater acoustic audio signal are calculated based on the time-domain and frequency-domain signals to obtain a recognition feature vector; specifically, this includes: Based on the time-domain signal set and the frequency-domain signal set of the underwater acoustic audio signal, the time-frequency characteristics of each frame of the underwater acoustic audio signal are calculated respectively. The time-frequency characteristics of each frame of underwater acoustic audio signal are statistically analyzed using statistical functions to obtain a time-frequency feature vector, which is then used as the recognition feature vector. The time-frequency features include spectral skewness coefficient, spectral kurtosis, spectral sharpness, and spectral centroid; the identification features include time-domain features, frequency-domain features, and time-frequency features of the underwater acoustic audio signal that are not included in the detection features. The identification feature vector is input into the second classifier for training to obtain the trained second classifier. A multi-target underwater acoustic recognition model is constructed based on the trained first classifier and the trained second classifier.

2. The underwater acoustic multi-target recognition method according to claim 1, characterized in that, The step of acquiring underwater acoustic audio signals from the training set, and acquiring the time-domain and frequency-domain signals of the underwater acoustic audio signals, includes: Acquire underwater acoustic audio signals and perform frame segmentation processing on the underwater acoustic audio signals; The time-domain signal and frequency-domain signal of each frame of underwater acoustic audio signal are obtained respectively to obtain the time-domain signal set and frequency-domain signal set of the underwater acoustic audio signal.

3. The underwater acoustic multi-target recognition method according to claim 1, characterized in that, The statistical functions include the maximum value function, the minimum value function, the standard deviation function, and the average value function.

4. The underwater acoustic multi-target recognition method according to claim 1, characterized in that, The step of inputting the detected feature vector into the first classifier for training to obtain the trained first classifier includes: The detected feature vector is input into the first classifier for training until the cross-entropy loss function of the binary classification task reaches its minimum, thus obtaining the trained first classifier. The cross-entropy loss function for the binary classification task is: ， in, It is the probability that the predicted result is a positive sample. represents the sample label, with 1 for positive samples and 0 for negative samples.

5. The underwater acoustic multi-target recognition method according to claim 1, characterized in that, The identified feature vector is input into the second classifier for training, and the trained second classifier includes: The identified feature vector is input into the second classifier for training until the cross-entropy loss function of the multi-class task reaches its minimum, thus obtaining the trained second classifier. The cross-entropy loss function for the multi-class classification task is: ， in, For the number of categories, This represents the sample label; if the sample category is... ,but ,otherwise, ; For the predicted result category The probability of.

6. The underwater acoustic multi-target recognition method according to claim 1, characterized in that, After constructing the underwater acoustic multi-target recognition model based on the trained first classifier and the trained second classifier, the method further includes: Acquire underwater acoustic audio signals from the test set, calculate the detection feature vector of the underwater acoustic audio signals, and input the detection feature vector into the first classifier in the underwater acoustic multi-target recognition model for classification; Based on the classification results, determine whether the underwater acoustic audio signal has a corresponding underwater acoustic target. If it does, then the underwater acoustic audio signal is taken as the target underwater acoustic audio signal. The identification feature vector of the target underwater acoustic audio signal is calculated, and the identification feature vector is input into the second classifier in the underwater acoustic multi-target identification model for classification to determine the underwater acoustic target type corresponding to the target underwater acoustic audio signal.

7. A multi-target underwater acoustic recognition device, characterized in that, include: The signal acquisition module is used to acquire underwater acoustic audio signals in the training set, and to acquire the time-domain signal and frequency-domain signal of the underwater acoustic audio signals; The detection feature extraction module is used to calculate the detection features of the underwater acoustic audio signal based on the time domain signal and the frequency domain signal of the underwater acoustic audio signal, and obtain the detection feature vector; Specifically, it includes: Calculate the temporal characteristics of each frame of the time-domain signal in the time-domain signal set of the underwater acoustic audio signal; The time-domain features of each frame of the time-domain signal are statistically analyzed using statistical functions to obtain the time-domain feature vector of the underwater acoustic audio signal; wherein, the time-domain features include zero-crossing rate, short-time energy, and short-time autocorrelation function; Calculate the frequency domain characteristics of each frame of the frequency domain signal in the frequency domain signal set of the underwater acoustic audio signal; The frequency domain features of each frame of the frequency domain signal are statistically analyzed using statistical functions to obtain the frequency domain feature vector of the underwater acoustic audio signal; wherein, the frequency domain features include Mel-frequency cepstral coefficients, spectral width, and spectral difference amplitude; The time-domain feature vector and the frequency-domain feature vector are combined to form the detection feature vector; wherein, the detection features include the time-domain features, frequency-domain features, and time-frequency features of the underwater acoustic audio signal; The first classification module is used to input the detection feature vector into the first classifier for training to obtain a trained first classifier. The feature extraction module is used to calculate the recognition features of the underwater acoustic audio signal based on the time-domain and frequency-domain signals of the underwater acoustic audio signal, and obtain the recognition feature vector; specifically, it includes: Based on the time-domain signal set and the frequency-domain signal set of the underwater acoustic audio signal, the time-frequency characteristics of each frame of the underwater acoustic audio signal are calculated respectively. The time-frequency characteristics of each frame of underwater acoustic audio signal are statistically analyzed using statistical functions to obtain a time-frequency feature vector, which is then used as the recognition feature vector. The time-frequency features include spectral skewness coefficient, spectral kurtosis, spectral sharpness, and spectral centroid; the identification features include time-domain features, frequency-domain features, and time-frequency features of the underwater acoustic audio signal that are not included in the detection features. The second classification module is used to input the recognition feature vector into the second classifier for training, so as to obtain the trained second classifier. The model building module is used to build an underwater acoustic multi-target recognition model based on the trained first classifier and the trained second classifier.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the underwater acoustic multi-target recognition method according to any one of claims 1-6.

Images