Method for identifying underwater acoustic signal and method for training underwater acoustic signal identification model

Abstract

The application discloses a water acoustic signal recognition method and a training method of a water acoustic signal recognition model. The water acoustic signal recognition method comprises the following steps: acquiring a water acoustic signal; extracting a wavelet scattering feature of the water acoustic signal by using a wavelet scattering transformation model; and identifying the wavelet scattering feature based on a trained neural network model to obtain a target object corresponding to the water acoustic signal.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of underwater acoustic target recognition, and in particular to a method for recognizing underwater acoustic signals and a method for training an underwater acoustic signal recognition model. BACKGROUND

[0002] In recent years, with the progress of submarine noise reduction technology, underwater unmanned vehicles have developed rapidly, and underwater weapons such as torpedoes and mines have shown a trend of diversification, and the sea battlefield environment is more complex. Underwater acoustic target recognition is the premise of anti-submarine, torpedo defense and underwater acoustic countermeasures, and has become an important research topic. Underwater acoustic target recognition technology is an information processing technology that uses passive target radiation noise, active target echo received by sonar and other sensor information to extract target features and identify target types or ship types, providing important decision basis for human marine economy and military activities.

[0003] Underwater acoustic target recognition technology includes underwater acoustic target active recognition and underwater acoustic target passive recognition. Underwater acoustic target active recognition is based on different types of sonars equipped on different ships, submarines and torpedoes (the parameters such as frequency band, period, pulse and pulse width of different types of active sonars are different), according to the active sonar characteristics of the detected target such as working bandwidth, pulse period, lower limit frequency and upper limit frequency, the sonar type can be identified, the recognition range is narrowed by exclusion method, and further combined with other target features to complete the comprehensive identification of the target. Underwater acoustic target passive recognition is to analyze the characteristics of the underwater acoustic signals received by the sonar, and then effectively judge and identify the target type, attitude and state.

[0004] The existing underwater acoustic target recognition technology has poor recognition effect and low accuracy. In order to improve the performance of underwater acoustic target recognition, it is urgent to obtain spectral features with large inter-class differences, good robustness, high signal-to-noise ratio and environmental adaptability from target signals. SUMMARY

[0005] The purpose of the present application is to provide a method for recognizing underwater acoustic signals and a method for training an underwater acoustic signal recognition model, which at least improves the performance of underwater acoustic target recognition and achieves the beneficial effect of obtaining spectral features with large inter-class differences, good robustness, and high signal-to-noise ratio and environmental adaptability.

[0006] According to one aspect of the present application, at least one embodiment provides a method for recognizing underwater acoustic signals, comprising: obtaining an underwater acoustic signal; using a wavelet scattering transform model to extract wavelet scattering features of the underwater acoustic signal; identifying the wavelet scattering features based on a trained neural network model to obtain a target object corresponding to the underwater acoustic signal.

[0007] According to another aspect of the present application, at least one embodiment further provides a method for training a water acoustic signal recognition model, comprising: obtaining water acoustic sample data, wherein the water acoustic sample data is divided into a training set and a test set; constructing a wavelet scattering transform model, extracting training wavelet scattering features of the water acoustic sample data in the training set, and extracting test wavelet scattering features of the water acoustic sample data in the test set; training a neural network model using a convolutional neural network on the training wavelet scattering features, and testing and verifying the neural network model using the test wavelet scattering features.

[0008] According to another aspect of the present application, at least one embodiment further provides an electronic device, comprising: a processor adapted to implement instructions; and a memory adapted to store a plurality of instructions adapted to be loaded and executed by the processor: the water acoustic signal recognition method of the present application, and / or the training method of the water acoustic signal recognition model of the present application.

[0009] According to another aspect of the present application, at least one embodiment further provides a computer-readable non-volatile storage medium storing computer program instructions, which, when executed by a computer, perform the water acoustic signal recognition method of the present application, and / or the training method of the water acoustic signal recognition model of the present application.

[0010] In the above manner of the present application, water acoustic signal preprocessing, wavelet scattering feature extraction, and deep convolutional neural network model construction are adopted, wherein the preprocessing reduces the influence of background noise and signal irregularity on subsequent signal processing, the wavelet scattering features extracted by the wavelet scattering transform model have small intra-class difference, large inter-class difference, and strong robustness, overcoming the problem that it is difficult to extract effective features from original data using conventional time-frequency domain transform methods, and the deep convolutional neural network model constructed uses wavelet scattering features with strong stability and high recognition degree and the strong learning ability of the deep learning model, so that a good water acoustic target recognition accuracy can be obtained. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the specific embodiments of the present application, the drawings required to be used in the specific embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0012] Figure 1 is a schematic diagram of an application environment according to an embodiment of the present application;

[0013] Figure 2 is a schematic diagram of an electronic device according to an embodiment of the present application;

[0014] Figure 3A flow chart of a training method of the underwater acoustic signal recognition model according to the embodiment of the present application is shown in FIG. 1.

[0015] Figure 4 A second layer wavelet filter bank in the wavelet scattering transform model according to the embodiment of the present application is shown in FIG. 2.

[0016] Figure 5 A third layer wavelet filter bank in the wavelet scattering transform model according to the embodiment of the present application is shown in FIG. 3.

[0017] Figure 6 A recognition confusion matrix of the underwater acoustic signal recognition model according to the embodiment of the present application is shown in FIG. 4.

[0018] Figure 7 A neural network model structure diagram according to the embodiment of the present application is shown in FIG. 5.

[0019] Figure 8 A flow chart of the underwater acoustic signal recognition method according to the embodiment of the present application is shown in FIG. 6. DETAILED DESCRIPTION

[0020] The technical solutions of the present application will be described clearly and completely below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0021] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily mean a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0022] Currently, most of the underwater acoustic target recognition tasks are to map the collected one-dimensional time sequence signal to a two-dimensional spectrum, and then input the obtained two-dimensional spectrum into a neural network for training and learning, and use the trained network model to classify and recognize the underwater acoustic target. In this process, the extraction of the two-dimensional spectrum (signal spectrum feature) and the design of the neural network structure (deep learning model) are two important links. The extraction of the signal spectrum feature as the direct input of the deep learning model determines the characteristic parameters of the target recognized by the deep learning model, thereby determining the "intelligence" of the network model after training, and finally determining the good or bad of the recognition performance.

[0023] For example, among the spectrum features used in many underwater acoustic target recognition, the Mel spectrum is a relatively common one. The spectrum information corresponding to the target signal is obtained through framing, windowing, Fourier transform, Mel filter bank design and logarithmic operation. The Mel filter bank can simulate the working mechanism of the human ear and to some extent represent the auditory perception characteristics of the human ear, and is commonly used in speech signal processing. However, the inventors found that the underwater acoustic signal has a strong local structure, and some information will be lost in the Mel spectrum transformation process, which cannot completely capture the rich local structure features of the target underwater acoustic signal, limiting the classification and recognition effect. That is, although different underwater acoustic targets have certain differences in general spectrum features, in actual application scenarios, the differences in the observation sea area, real-time sea conditions and the diversity of underwater acoustic target types lead to that the general spectrum feature data of the underwater acoustic target is not obvious in distinguishability. Therefore, in order to improve the underwater acoustic target recognition performance, it is urgent to obtain a spectrum feature from the underwater acoustic signal, which has large differences between categories, good robustness, and high adaptability to signal-to-noise ratio and environment.

[0024] Based on this, at least one embodiment of the present application provides an underwater acoustic signal recognition system, which includes an electronic device, at least for executing the underwater acoustic signal recognition method developed by the present application and / or the training method of the underwater acoustic signal recognition model developed by the present application, at least to achieve the beneficial effects of improving the underwater acoustic target recognition performance, obtaining a spectrum feature with large differences between categories, good robustness, and high adaptability to signal-to-noise ratio and environment. The underwater acoustic signal recognition system can include an environment as shown in Figure 1 The above hardware environment includes an electronic device 100 and a server 200, and the electronic device 100 can operate the server 200 through corresponding instructions, so as to read, change, add data, etc.

[0025] The electronic device 100 can be one or more, and a plurality of processing nodes can be included in the electronic device 100, which can be externally regarded as a whole. Optionally, the electronic device 100 can also send the acquired underwater acoustic signal to the server 200, so that the server 200 performs the underwater acoustic signal identification method developed by the present application and / or the training method of the underwater acoustic signal identification model developed by the present application. Optionally, the electronic device 100 can be connected with the server 200 through a network.

[0026] The network includes a wired network and a wireless network. The wireless network includes but is not limited to a wide area network, a metropolitan area network, a local area network or a mobile data network. Typically, the mobile data network includes but is not limited to a global system for mobile communication (GSM) network, a code division multiple access (CDMA) network, a wideband code division multiple access (WCDMA) network, a long term evolution (LTE) communication network, a WIFI network, a ZigBee network, a Bluetooth technology-based network, etc. Different types of communication networks can be operated by different operators. The type of communication network does not constitute a limitation on the embodiments of the present application.

[0027] The electronic device 100, as shown in Figure 2 The electronic device 100, as shown in

[0028] The processor 202 can be various applicable processors, for example, implemented in the form of a central processor, a microprocessor, an embedded processor, etc., and can adopt an X86, ARM, etc. architecture. The memory 204 can be various applicable storage devices, for example, a non-volatile storage device, including but not limited to a magnetic storage device, a semiconductor storage device, an optical storage device, etc., and can be arranged as a single storage device, a storage device array or a distributed storage device, and the embodiments of the present application do not limit these.

[0029] Those skilled in the art can understand that the structure of the electronic device 100 described above is only schematic, and does not limit the structure of the electronic device 100. For example, the electronic device 100 can also include more or fewer components than those described above. Figure 2More or less components (such as a transmission device) can be shown in the figures. The transmission device is used to receive or send data via a network. In one example, the transmission device is a radio frequency (RF) module, which is used to communicate with the Internet in a wireless manner.

[0030] Under the above operating environment, the training method of the underwater acoustic signal recognition model is proposed in at least one embodiment of the present application, which can be loaded and executed by the processor 202. As shown in the flowchart of the training method of the underwater acoustic signal recognition model, it should be noted that the steps shown in the flowchart can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from here. The method can include the following steps: Figure 3

[0031] Step S301, obtaining underwater acoustic sample data, wherein the underwater acoustic sample data is divided into a training set and a test set;

[0032] Step S303, constructing a wavelet scattering transform model, extracting training wavelet scattering features of the underwater acoustic sample data in the training set, and extracting test wavelet scattering features of the underwater acoustic sample data in the test set;

[0033] Step S305, training a neural network model using a convolutional neural network on the training wavelet scattering features, and testing and verifying the neural network model using the test wavelet scattering features.

[0034] It can be seen that the present application proposes an underwater acoustic target recognition method based on wavelet scattering features and deep learning. Compared with the Fourier transform, constant-Q transform, MFCC transform and the like in the prior art, the underwater acoustic signal recognition model has the advantages of high shift invariance, micro-variable stability and recognition ability, can extract robust features of underwater acoustic target signals, effectively improves the correct rate of recognition, and adopts a deep convolutional neural network model with a multi-layer structure, which can represent the input data in a form that is easier to separate or recognize.

[0035] In step S301, underwater acoustic sample data is obtained, which is divided into a training set and a test set. Optionally, the underwater acoustic sample data is divided into a training set and a test set in the following manner: the underwater acoustic sample data is randomly divided into a training set and a test set; the underwater acoustic sample data in the training set and the test set is segmented into equal-length time-domain frame sequences, and corresponding label information of the frame sequences is generated, wherein the label information is the object category to which the frame sequence belongs; the time-domain frame sequences are preprocessed, and the preprocessing includes filtering, detrending and normalization processing.

[0036] ​For example, the underwater acoustic sample data includes three types of signals of passenger ships, motorboats and background noise, each with 14 original samples, and the sampling rate is 52734 Hz, and the sample time length is from 30 s to 400 s. For example, the original samples are randomly divided into 80% training set and 20% test set according to the ratio of 9:1; the data in the training set and the test set are divided into equal-length time domain frame sequences, the length is about 5.94 s, and the corresponding labeled information of the frame sequence is generated, the number of frame sequences is the number of samples of each type of target, and the labeled information is the category to which the frame sequence belongs, and finally the number of samples in the training set and the test set is shown in Table 1; the pre-processing of the framed data is performed, that is, each frame of data is filtered by designing a low-pass filter to filter out high-frequency noise above 3 kHz, and the filtered signal is de-trended and normalized. In addition, since the sampling rate of the underwater acoustic sample data is high, the present application can also perform downsampling processing to make the sampling rate 22050 Hz.

[0037] Table 1 Number of samples of each type of target in the training set and the test set after frame processing

[0038] Training set sample number Test set sample number Passenger ship 356 27 Motorboat 359 27 Background noise 356 28

[0039] In step S303, a wavelet scattering transform model is constructed, training wavelet scattering features of the underwater acoustic sample data in the training set are extracted, and test wavelet scattering features of the underwater acoustic sample data in the test set are extracted. The wavelet scattering transform model can include a multi-layer wavelet scattering transform, and the construction of the wavelet scattering transform model can include: obtaining wavelet scattering transform parameters, wherein the wavelet scattering transform parameters include input signal length L, sampling rate Fs, quality factor Q, and invariant scale T, the dimension of the quality factor Q = the number of layers of the wavelet scattering transform = the number of wavelet filter banks, and the invariant scale T is used to represent the time scale of the scale filter; determining the multi-layer wavelet scattering transform based on the wavelet function ψ(ω) and the scale function φ(ω), wherein, ω is the angular frequency, σ is the multi-resolution parameter, ω c is the central angular frequency, δ is the scale parameter. It should be noted that the wavelet scattering transform model can be constructed by Morlet wavelet, Haar wavelet, Gammatone wavelet and the like, and the present application takes Morlet wavelet as an example to describe the method of constructing the wavelet scattering transform model.

[0040] It can be seen that the wavelet scattering transform model of the present application can construct a multi-layer wavelet scattering transform. Taking the construction of M(M=1, 2, 3, …) layer wavelet scattering transform architecture as an example, the wavelet scattering transform parameters are designed, including L, Fs, Q, T, ψ(ω) and φ(ω) are determined, and the m(m=1, 2, …, M) layer wavelet filter set (ψ1, ψ2, ψ3, …, ψ N) and scale filter φ s , N represents the number of filters in the filter bank. In the subsequent description of the training method of the developed underwater acoustic signal recognition model and / or the developed underwater acoustic signal recognition method, the designed wavelet scattering transform parameters are: M = 3, L = 5.94s, Fs = 22050Hz, Q =

[081] , T = 0.12s.

[0041] Based on the above-mentioned wavelet scattering transform model, the training wavelet scattering features of the underwater acoustic sample data in the training set are extracted, and the test wavelet scattering features of the underwater acoustic sample data in the test set are extracted. As shown in Figures 4-5 When M = 3, the wavelet scattering features of the labeled training set and test set samples in step S301 are extracted by using 3-layer wavelet scattering transform. For example, in the first layer wavelet scattering transform, the scale filter is used to filter the sample signals in the training set and the test set, and the first layer scattering coefficient is generated; in the second layer wavelet scattering transform, the second layer wavelet filter bank is obtained according to the designed wavelet scattering transform parameters, the second layer wavelet filter bank is used to perform continuous wavelet transform on the sample signal, and the wavelet transform coefficient is obtained and is subjected to a modulus operation, and then the scale filter is used to filter the operation result, and the second layer scattering coefficient is generated, and the modulus operation result is saved as the third layer input; in the third layer wavelet scattering transform, the third layer wavelet filter bank is obtained according to the designed wavelet scattering transform parameters, the third layer wavelet filter bank is used to perform continuous wavelet transform on the modulus operation result of the second layer, and the modulus operation is performed, and then the scale filter is used to filter, and the third layer scattering coefficient is generated; after the wavelet scattering transform is completed, the wavelet scattering features are composed of the wavelet scattering coefficients of each layer, and in general application, the wavelet scattering coefficients can be down-sampled to reduce the operation amount of subsequent training in combination with the performance of the computing device.

[0042] In step S305, the training wavelet scattering features are trained by using the convolutional neural network, and the test wavelet scattering features are used to test and verify the neural network model. That is, the wavelet scattering features of the underwater acoustic signal obtained from the training set are used as learning features, the wavelet scattering features of the underwater acoustic signal are input into the network model based on the designed multi-layer structure of the convolutional neural network, i.e. the deep convolutional neural network (DCNN) model, the wavelet scattering features of the underwater acoustic signal are trained, and the underwater acoustic target recognition model is output; the obtained test set wavelet scattering features are used to test the underwater acoustic target recognition model, the effectiveness is verified, and the recognition accuracy of the model is obtained. The confusion matrix of the test result is shown in Figure 6 The average recognition accuracy of the model is about 93.90%.

[0043] Optionally, in the process of training the underwater acoustic target classification network model, the DCNN network is designed as a 5-layer neural network structure, specifically, an input layer is connected with 5 convolutional networks, each convolutional network includes a convolutional layer, a batch normalization layer, an activation function, and a pooling layer, a Dropout layer is added after the multi-layer convolutional network to improve the generalization, and finally a fully connected layer, a Softmax layer and an output layer are connected to realize target classification. The 5-layer deep convolutional neural network structure and its parameters are shown in Figure 7 , wherein the dropout rate of the Dropout layer is set to 0.2, and the learning rate in the training process is set to 10-4.

[0044] Through the above manner of the present application, the original underwater acoustic target data collected is preprocessed, the influence of background noise and signal zero drift on subsequent signal processing is reduced, a wavelet scattering transform model suitable for underwater acoustic signals is designed, compared with commonly used Fourier transform, constant-Q transform, MFCC transform and the like, the model has the advantages of high shift invariance, micro-variable stability and recognition ability, can extract the robust features of the underwater acoustic target signal, and effectively improves the correct rate of recognition, in addition, a deep convolutional neural network model with a multi-layer structure is used, which can represent the input data in a form that is easier to separate or identify.

[0045] Under the above running environment, at least one embodiment of the present application proposes an underwater acoustic signal recognition method, which can be loaded and executed by the processor 202. As shown in the flowchart of the underwater acoustic signal recognition method, Figure 8 , it should be noted that the steps shown in the flowchart of the drawing can be executed in a computer system such as a group of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here, the method can include the following steps:

[0046] Step S802, acquiring an underwater acoustic signal;

[0047] Step S804, extracting wavelet scattering features of the underwater acoustic signal by using a wavelet scattering transform model;

[0048] Step S806, identifying the wavelet scattering features based on the trained neural network model to obtain a target object corresponding to the underwater acoustic signal.

[0049] It should be noted that, in order to facilitate understanding, the water acoustic signal recognition model of the present application is split into a wavelet scattering transform model and a neural network model to describe the water acoustic signal recognition method of the present application. The above trained neural network model includes an input layer, a multi-layer convolutional network, a Dropout layer, a fully connected layer, a Softmax layer and an output layer. The above (trained) wavelet scattering transform model includes a multi-layer wavelet scattering transform, which can include a first layer wavelet scattering transform, a second layer wavelet scattering transform and a third layer wavelet scattering transform.

[0050] In step S802, a water acoustic signal is acquired. For example, the water acoustic signal is received by a sonar.

[0051] In step S804, a wavelet scattering feature of the water acoustic signal is extracted by using the wavelet scattering transform model. For example, the water acoustic signal is first filtered by using the first layer wavelet scattering transform to generate first scattering coefficients; the water acoustic signal is subjected to a second continuous wavelet transform and second filtering by using the second layer wavelet scattering transform to generate second scattering coefficients; the result of the second continuous wavelet transform is subjected to a third continuous wavelet transform and third filtering by using the third layer wavelet scattering transform to generate third scattering coefficients. That is, the wavelet scattering feature includes a scattering coefficient matrix composed of the first scattering coefficients, the second scattering coefficients and the third scattering coefficients.

[0052] When M = 3, L = 5.94s, Fs = 22050Hz, Q =

[081] , T = 0.12s, the first layer wavelet scattering transform can include a first scale filter, the second layer wavelet scattering transform can include a second wavelet filter bank and a second scale filter, and the third layer wavelet scattering transform can include a third wavelet filter bank and a third scale filter, wherein the dimension of Q is equal to the number of layers of the wavelet scattering transform, and also equal to the number of wavelet filter banks, the elements of Q are quality factors of each filter bank, and the number of filters in each octave is represented by T, usually the first element of Q is 0, indicating that the first layer does not use the wavelet filter.

[0053] Therefore, the first filtering of the water acoustic signal by using the first layer wavelet scattering transform can include filtering the water acoustic signal by using the first scale filter. The second continuous wavelet transform and the second filtering of the water acoustic signal by using the second layer wavelet scattering transform can include performing the second continuous wavelet transform on the water acoustic signal to obtain second wavelet transform coefficients by using the second wavelet filter bank, and filtering the second wavelet transform coefficients by using the second scale filter. The third continuous wavelet transform and the third filtering of the result of the second continuous wavelet transform by using the third layer wavelet scattering transform can include performing the continuous wavelet transform on the second wavelet transform coefficients to obtain third wavelet transform coefficients by using the third wavelet filter bank, and filtering the third wavelet transform coefficients by using the third scale filter.

[0054] In step S806, the wavelet scattering feature is identified based on the trained neural network model to obtain the target object corresponding to the underwater acoustic signal. Optionally, the wavelet scattering feature is processed by an input layer, a multi-layer convolution network, a Dropout layer, a full connection layer, and a Softmax layer in sequence, and is output by an output layer, wherein each convolution network of the multi-layer convolution network further comprises a convolution layer, a batch normalization layer, an activation function, and a pooling layer.

[0055] Through the above manner of the present application, the underwater acoustic signal preprocessing, the wavelet scattering feature extraction, and the deep convolution neural network model construction are realized, wherein the preprocessing reduces the influence of background noise and signal irregularity on subsequent signal processing, the wavelet scattering feature extracted by the wavelet scattering transformation model has small intra-class difference, large inter-class difference, and strong robustness, and overcomes the problem that it is difficult to extract effective features from original data by using conventional time-frequency domain transformation methods, the deep convolution neural network model constructed by using the wavelet scattering feature with strong stability and high recognition degree and the strong learning ability of the deep learning model can obtain a good underwater acoustic target recognition accuracy.

[0056] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method of identifying an underwater acoustic signal, characterized by, The method comprises: obtaining an underwater acoustic signal; extracting a wavelet scattering feature of the underwater acoustic signal by using a wavelet scattering transform model, wherein the wavelet scattering transform model comprises a multi-layer wavelet scattering transform, and the multi-layer wavelet scattering transform comprises an mth-layer wavelet scattering transform, and the mth-layer wavelet scattering transform is determined based on a length L, a sampling rate Fs, a quality factor Q and an invariant scale T of the underwater acoustic signal; identifying the wavelet scattering feature based on a trained neural network model to obtain a target object corresponding to the underwater acoustic signal. Specifically, the multi-layer wavelet scattering transform comprises a first-layer wavelet scattering transform, a second-layer wavelet scattering transform and a third-layer wavelet scattering transform, and the extracting of the wavelet scattering feature of the underwater acoustic signal comprises: performing first filtering on the underwater acoustic signal by using the first-layer wavelet scattering transform to generate first scattering coefficients; performing second continuous wavelet transform and second filtering on the underwater acoustic signal by using the second-layer wavelet scattering transform to generate second scattering coefficients; performing third continuous wavelet transform and third filtering on a result of the second continuous wavelet transform by using the third-layer wavelet scattering transform to generate third scattering coefficients. 2.The identification method of claim 1, wherein the trained neural network model comprises an input layer, a multi-layer convolutional network, a Dropout layer, a fully connected layer, a Softmax layer, and an output layer, and wherein, The identifying of the wavelet scattering feature based on the trained neural network model comprises: the wavelet scattering feature is processed in sequence by an input layer, a multi-layer convolution network, a Dropout layer, a full connection layer, a Softmax layer and an output layer, wherein each layer of the multi-layer convolution network further comprises a convolution layer, a batch normalization layer, an activation function and a pooling layer.

3. The identification method according to claim 1, wherein the wavelet scattering feature comprises first scattering coefficients, second scattering coefficients and third scattering coefficients, the first-layer wavelet scattering transform comprises a first scale filter, the second-layer wavelet scattering transform comprises a second wavelet filter bank and a second scale filter, and the third-layer wavelet scattering transform comprises a third wavelet filter bank and a third scale filter, and the method further comprises: the first filtering on the underwater acoustic signal by using the first-layer wavelet scattering transform comprises filtering the underwater acoustic signal by using the first scale filter; the second continuous wavelet transform and the second filtering on the underwater acoustic signal by using the second-layer wavelet scattering transform comprise performing second continuous wavelet transform on the underwater acoustic signal by using the second wavelet filter bank to obtain second wavelet transform coefficients, and filtering the second wavelet transform coefficients by using the second scale filter; the third continuous wavelet transform and the third filtering on the result of the second continuous wavelet transform by using the third-layer wavelet scattering transform comprise performing continuous wavelet transform on the second wavelet transform coefficients by using the third wavelet filter bank to obtain third wavelet transform coefficients, and filtering the third wavelet transform coefficients by using the third scale filter.

4. The identification method according to claim 1, characterized in that, The obtaining of the underwater acoustic signal comprises: receiving the underwater acoustic signal by a sonar.

5. The method of claim 1, wherein the water acoustic signal recognition model is trained by using a plurality of water acoustic signals. The method comprises: obtaining underwater acoustic sample data, wherein the underwater acoustic sample data is divided into a training set and a test set; constructing a wavelet scattering transform model, extracting training wavelet scattering features of the underwater acoustic sample data in the training set, and extracting test wavelet scattering features of the underwater acoustic sample data in the test set, wherein the wavelet scattering transform model comprises a multi-layer wavelet scattering transform, and the multi-layer wavelet scattering transform comprises an m-th layer wavelet scattering transform, and the m-th layer wavelet scattering transform is determined based on a length L of the underwater acoustic signal, a sampling rate Fs, a quality factor Q, and an invariance scale T; training a convolutional neural network model using the wavelet scattering features, and testing and verifying the convolutional neural network model using the test wavelet scattering features; wherein the wavelet scattering transform model comprises a multi-layer wavelet scattering transform, and constructing the wavelet scattering transform model comprises: obtaining wavelet scattering transform parameters, wherein the wavelet scattering transform parameters comprise an input signal length L, a sampling rate Fs, a quality factor Q, and an invariance scale T, the quality factor Q has a dimension = a number of layers of the wavelet scattering transform = a number of wavelet filter banks, and the invariance scale T is used to represent a time scale of a scale filter; A multi-layer wavelet scattering transform is determined based on a wavelet function ψ(ω) and a scale function φ(ω), where ω is an angular frequency, σ is a multi-resolution parameter, ω c is a central angular frequency, δ is a scale parameter.

6. The training method of claim 5, wherein the underwater acoustic sample data is divided into the training set and the test set comprises: randomly dividing the underwater acoustic sample data into the training set and the test set; segmenting the underwater acoustic sample data in the training set and the test set into equal-length time domain frame sequences, and generating corresponding labeled information of the frame sequences, wherein the labeled information is a class of an object to which the frame sequence belongs; preprocessing the time domain frame sequences, and the preprocessing comprises filtering, detrending, and normalization processing.

7. An electronic device comprising: a processor adapted to implement instructions; and a memory adapted to store a plurality of instructions adapted to be loaded and executed by the processor, the instructions comprising: a method for identifying an underwater acoustic signal according to any one of claims 1-4, and / or a method for training an underwater acoustic signal identification model according to any one of claims 5-6.

8. A computer-readable nonvolatile storage medium storing computer program instructions, when executed by a computer, performing: a method for identifying an underwater acoustic signal according to any one of claims 1-4, and / or a method for training an underwater acoustic signal identification model according to any one of claims 5-6. ​

Images