Deep learning-based ship acoustic image classification method and system

Abstract

The invention discloses a ship acoustic image classification method and system based on deep learning, and the method comprises the steps: carrying out the training optimization of an underwater acoustic signal sample based on sample amplification and class number equalization, generating a ship target classification and recognition model, carrying out the key feature extraction of a to-be-recognized underwater acoustic signal through employing the ship target classification and recognition model, and carrying out the recognition of the to-be-recognized underwater acoustic signal. Obtaining a ship classification result corresponding to the underwater acoustic signal to be identified according to a classification result corresponding to the extracted key features; simulated and actually measured signals are analyzed in detail by adopting a variational mode decomposition algorithm, and it is verified that the algorithm can effectively distinguish line spectrum characteristics and high-frequency noise interference in the signals. The grey wolf optimization algorithm is adopted to optimize key parameters in the VMD algorithm, modulation line spectrums and frequency multiplication components of the modulation line spectrums in actually measured signals can be accurately extracted, the purpose of signal enhancement is achieved, and therefore the effectiveness and practicability of the method are proved.

Description

Technical Field

[0001] The present invention belongs to the field of marine ship identification engineering, and particularly relates to a ship acoustic image classification method and system based on deep learning. Background Art

[0002] In underwater acoustic target recognition, the radiated noise signals emitted by ships are the main basis for passive sonar target recognition. However, these signals often have a low signal-to-noise ratio, increasing the difficulty of extracting effective information. In addition, in the underwater environment, due to the great difficulty of data collection, the amount of data available for recognition is small, which limits the feasibility of using deep learning models to improve ship target classification and recognition. Existing methods can accurately classify ship acoustic images. Therefore, conducting research on ship target feature extraction and classification recognition methods is of great significance for improving the accuracy of ship target recognition. Summary of the Invention

[0003] The purpose of the present invention is to provide a ship acoustic image classification method and system based on deep learning to overcome the problem of low accuracy in underwater ship target classification and recognition in the prior art.

[0004] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0005] A ship acoustic image classification method based on deep learning includes the following steps:

[0006] S1, training and optimizing an underwater acoustic signal sample based on sample amplification and class number equalization to generate a ship target classification and recognition model;

[0007] S2, using the ship target classification and recognition model to extract key features from the underwater acoustic signal to be recognized, and obtaining the ship classification result corresponding to the underwater acoustic signal to be recognized according to the classification result corresponding to the extracted key features.

[0008] Preferably, the underwater acoustic signal sample based on sample amplification and class number equalization includes the following steps: using continuous wavelet transform and Wasserstein generative adversarial network to generate underwater acoustic signal samples to achieve sample amplification and class number equalization.

[0009] Preferably, the one-dimensional time-domain waveform diagram of the underwater acoustic signal sample is converted into a two-dimensional time-frequency spectrum diagram by continuous wavelet transform, and then converted into an image amplification task. The complex Morlet wavelet is used as the wavelet function, and then the wavelet transform scale group is calculated, and the wavelet basis function is generated. Subsequently, the wavelet coefficients obtained by performing continuous wavelet transform are used to draw the time-frequency spectrum diagram.

[0010] Preferably, the bandwidth parameter Fb of the complex Morlet wavelet function is 3, and the center frequency ω 0 = 3.

[0011] Preferably, the variational mode decomposition method based on the grey wolf optimization algorithm is used to decompose the passive radiated noise signal of the ship target, so as to obtain the optimal modal components of each passive radiated noise signal of the ship target and the optimal component with the minimum envelope entropy value; Subsequently, calculate the energy values of each modal component, select the two modal components with the highest energy among the remaining modal components, and reconstruct the signal with the optimal component with the minimum envelope entropy value and the two modal components with the highest energy among the remaining modal components to form a new ship target signal, thus completing the extraction of key features.

[0012] Preferably, the generated underwater acoustic signal samples and underwater acoustic signal samples are used to train the Swin-Transformer network; During the training process, optimize the performance of the model by adjusting the learning rate and batch size parameters; Use the validation set to verify the trained model for the optimized model, and the classification recognition accuracy of the optimized model reaches the set value to obtain the ship target classification recognition model.

[0013] A ship acoustic image classification system based on deep learning, including a training optimization module and an identification module,

[0014] The training optimization module trains and optimizes to generate a ship target classification recognition model based on underwater acoustic signal samples with sample amplification and class number equalization;

[0015] The identification module uses the ship target classification recognition model to extract key features from the underwater acoustic signal to be identified, and obtains the ship classification result corresponding to the underwater acoustic signal to be identified according to the classification result corresponding to the extracted key features.

[0016] Preferably, the underwater acoustic signal samples based on sample amplification and class number equalization include the following steps: Use continuous wavelet transform and Wasserstein generative adversarial network to generate underwater acoustic signal samples, and realize sample amplification and class number equalization.

[0017] A computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the above-mentioned ship acoustic image classification method based on deep learning are realized.

[0018] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the ship acoustic image classification method based on deep learning are realized.

[0019] Compared with the prior art, the present invention has the following beneficial technical effects:

[0020] The present invention provides a method for classifying ship acoustic images based on deep learning. A ship target classification and recognition model is trained and optimized based on underwater acoustic signal samples with sample amplification and class number equalization. The ship target classification and recognition model is used to extract key features from the underwater acoustic signal to be recognized, and the ship classification result corresponding to the underwater acoustic signal to be recognized is obtained according to the classification result corresponding to the extracted key features. The present invention uses the variational mode decomposition algorithm to conduct a detailed analysis of simulated and measured signals, verifying that the algorithm can effectively distinguish the line spectrum features and high-frequency noise interference in the signals. The grey wolf optimization algorithm is used to optimize the key parameters in the VMD algorithm, which can accurately extract the modulated line spectrum and its harmonic components in the measured signal, achieving the purpose of signal enhancement, thus proving the effectiveness and practicability of the method.

[0021] Preferably, for underwater acoustic signal samples in audio form, format conversion is performed. Using the continuous wavelet transform method, the one-dimensional time-domain waveform diagram of the underwater acoustic signal sample signal is mapped into a time-frequency spectrogram containing multi-domain features of frequency domain and time domain, forming an integrated representation method of multi-domain features. Subsequently, a high-quality fake sample is generated by the Wasserstein generative adversarial network according to the real underwater acoustic sample time-frequency diagram. The usability of the generated sample is confirmed through the network training results.

[0022] Preferably, through feature enhancement and sample amplification of the underwater acoustic signal, on this basis, the Swin-Transformer network is introduced into the field of ship target classification and recognition. The Swin-Transformer classifier is trained to obtain a ship target classification and recognition model based on multi-domain features. Subsequently, based on the data in the ShipsEar database, a classification and recognition experiment of 5 types of vehicles is completed. The classifier achieves an accuracy of 96% in classifying and recognizing 5 types of ship targets, fully demonstrating that the model can achieve accurate classification and recognition of ship targets. Description of the Drawings

[0023] Figure 1 It is a schematic flowchart of the method for classifying ship acoustic images based on deep learning in the embodiment of the present invention.

[0024] Figure 2 It is the time-domain waveform diagram and spectrogram of the original signal in the embodiment of the present invention.

[0025] Figure 3 It is the change curve diagram of the fitness function in the embodiment of the present invention.

[0026] Figure 4 It is the result diagram of the GWO algorithm optimizing the parameters of the VMD algorithm in the embodiment of the present invention.

[0027] Figure 5 It is the adaptive VMD decomposition result diagram in the embodiment of the present invention. Figure 5(a) is a time-domain waveform diagram, Figure 5 (b) is the corresponding spectrogram.

[0028] Figure 6 This is the optimal component Hilbert demodulation result diagram in the embodiment of the present invention.

[0029] Figure 7 This is the continuous wavelet transform result of the A-15 underwater acoustic signal in the embodiment of the present invention.

[0030] Figure 8 This is the diagram of the generation process of the radiated noise signal image of ship type A in the embodiment of the present invention; Figure 8 (a) is a partial real signal image, Figure 8 (b) is the generated signal image (epoch = 180), Figure 8 (c) is the generated signal image (epoch = 20000).

[0031] Figure 9 This is the schematic diagram of the training result of the Swin-Transformer network in the embodiment of the present invention. Detailed implementation manners

[0032] In order to enable those skilled in the art of the present technology to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0033] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order different from those illustrated or described here. In addition, the terms "including" and "having" 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 may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0034] As Figure 1 shown, the present invention provides a ship acoustic image classification method based on deep learning, which specifically includes the following steps:

[0035] S1. Optimize and generate a ship target classification and recognition model based on underwater acoustic signal sample amplification and class number equalization.

[0036] S2. Use the ship target classification and recognition model to extract key features from the underwater acoustic signal to be recognized, and obtain the ship classification result corresponding to the underwater acoustic signal to be recognized according to the classification result corresponding to the extracted key features.

[0037] Specifically, in the specific implementation manner of this application, aiming at the problems of insufficient underwater acoustic signal samples and unbalanced sample class numbers, based on the underwater acoustic signal samples of sample amplification and class number equalization, the specific acquisition method includes the following steps:

[0038] Use continuous wavelet transform and Wasserstein generative adversarial network (WGAN) to generate underwater acoustic signal samples, and achieve sample amplification and class number equalization.

[0039] Specifically, it includes the following steps:

[0040] Use continuous wavelet transform to convert the one-dimensional time-domain waveform diagram of the underwater acoustic signal sample into a two-dimensional time-frequency spectrum diagram, and then transform it into an image amplification task. Select the complex Morlet wavelet as the wavelet function as shown in Equation (3), and set its bandwidth parameter Fb = 3 and center frequency ω 0 = 3.

[0041]

[0042] In the formula: ψ(t) - complex Morlet wavelet function; t - time variable; σ - Gaussian standard deviation; ω 0 - center frequency of the wavelet.

[0043]

[0044] In the formula: scales - wavelet transform scale; f - signal frequency; totalscale - total wavelet transform scale; k - attenuation factor, gradually decreasing from the total scale to 1.

[0045] Calculate the wavelet transform scale group through Equation (4), and generate the wavelet basis function. Subsequently, draw the time-frequency spectrum diagram based on the wavelet coefficients obtained by performing continuous wavelet transform.

[0046] Wasserstein generative adversarial network (WGAN): Use WGAN to generate underwater acoustic signal samples with the same feature distribution to achieve sample amplification. WGAN optimizes the generator by minimizing the Wasserstein distance between the generated samples and the real samples, so as to generate more real and diverse samples.

[0047] Specifically, the variational mode decomposition (VMD) method based on the grey wolf optimization algorithm is used to decompose the passive radiated noise signal of the ship target, obtaining the optimal modal components of each ship target passive radiated noise signal and the optimal component with the minimum envelope entropy value. Subsequently, the energy values of each modal component are calculated, and the two modal components with the highest energy among the remaining modal components are selected. The optimal component with the minimum envelope entropy value and the two modal components with the highest energy among the remaining modal components are used for signal reconstruction to form a new ship target signal, completing the extraction of key features.

[0048] In the present invention, the grey wolf optimization algorithm (GWO algorithm) is used to optimize the parameters of the VMD algorithm. Two key parameters in the VMD algorithm (the preset number of decomposition modes K and the penalty factor α) are set as the position coordinates for grey wolf search, and the local minimum envelope entropy value E emin is selected as the fitness function, and minimizing it is set as the optimization goal. Through the iterative search process, the GWO algorithm is used to find the optimal parameter combination that enables the IMF to contain the most modulation information, thereby determining the optimal number of modal decompositions K best and the penalty factor α best . This optimization strategy ensures that the most abundant modulation feature information can be accurately extracted when processing signals. The envelope signal is extracted from the original signal through Hilbert demodulation, and calculating the information entropy of this envelope function is called envelope entropy. The calculation formula of the envelope entropy E e is shown in formula (1):

[0049]

[0050] In the formula: p j —— The normalized form of the envelope signal.

[0051] The calculation method of p j is shown in formula (2):

[0052]

[0053] In the formula: a(j) —— The envelope signal after the original signal is demodulated by Hilbert; N —— The signal length.

[0054] In the present invention, the grey wolf optimization algorithm is used to optimize the parameters of the VMD algorithm, which can reduce the noise of the underwater acoustic signal, lower the signal-to-noise ratio of the underwater acoustic signal, enhance the proportion of the characteristic components in the underwater acoustic signal, weaken the interference of the noise, and thus enhance the radiation noise characteristics of the ship target.

[0055] The present invention decomposes the modal components with the most vehicle characteristics through the VMD algorithm optimized by GWO parameters, screens out the components containing background noise, thereby eliminating the interference of background noise, and realizes the process of enhancing the characteristics of the radiated noise of the vehicle, laying a foundation for subsequent feature signal sample amplification and classification recognition, especially improving the accuracy of the classification model.

[0056] In this application, a variational mode decomposition (VMD) method based on the grey wolf optimization algorithm is used to decompose the passive radiated noise signal of the ship target, which specifically includes the following steps:

[0057] S1, Initialize the parameters of the GWO (grey wolf optimization algorithm) algorithm, as well as the parameter ranges of the number of modal decompositions K and the quadratic penalty factor α, initialize the population, and the positions of each search wolf (grey wolf).

[0058] S2, Enter the loop, and update the iteration number n = n + 1;

[0059] S3, Take the positions of each search wolf as the input of the variational mode decomposition (VMD) algorithm, and use the VMD algorithm to decompose the signal to obtain K IMF (intrinsic mode function) components;

[0060] S4, Calculate the envelope entropy of each IMF component, screen out the minimum envelope (i.e., the IMF component or component set with the minimum envelope entropy), and take the entropy value of the screened minimum envelope as the fitness value of the current search wolf.

[0061] S5, According to the fitness values of the search wolves, select the first three search wolves with the optimal fitness values (i.e., α wolf, β wolf, δ wolf), and update the positions of the remaining search wolves (ω wolf) and the search range and step size of the GWO algorithm according to the social hierarchy and hunting behavior of the grey wolves.

[0062] S6, Determine whether the termination condition is satisfied, that is, whether the global optimal solution has been found or the maximum number of iterations has been reached. If not, repeat steps S2 - S5;

[0063] S7, If the termination condition is satisfied, end the algorithm and output the number K best of the optimal decomposition mode best and the corresponding penalty factor α.

[0064] Train and optimize to generate a ship target classification and recognition model:

[0065] The Swin-Transformer network is adopted as the basic architecture of the ship target classification and recognition model; Swin-Transformer is a vision Transformer model based on self-attention, with powerful feature representation capabilities. Through a multi-level self-attention mechanism, Swin-Transformer can capture global and local information in the image, thus achieving accurate recognition of ship targets;

[0066] The generated underwater acoustic signal samples and underwater acoustic signal samples are used to train the Swin-Transformer network. During the training process, the performance of the model is optimized by adjusting parameters such as the learning rate and batch size; the optimized model is verified using the validation set to evaluate the accuracy and recall rate indicators of its classification and recognition. By continuously iterating the training process, the generalization ability and robustness of the model are improved. In this application, the classification accuracy of the method of the present invention on 5 types of underwater acoustic targets reaches 96%, which is higher than other common deep learning models. The following Table 1 shows the comparison of the classification accuracies of different methods on 5 types of underwater acoustic targets.

[0067] Table 1 Classification and recognition results of different methods for the ShipsEar library

[0068]

[0069] Embodiment

[0070] In a specific embodiment of this application, the measured ship signal samples in the ShipsEar library are used to experiment with this method. The time sampling length of the original underwater radiated noise signal is 1 s, and the sampling frequency is 8 kHz. Its time-domain waveform diagram and frequency spectrum diagram are as Figure 2 shown. The GWO algorithm is used to optimize the VMD parameters for the signal samples. The parameters of the GWO algorithm used are shown in Table 2: the number of search wolves is 8, the maximum number of iterations is 10, the range of the number of decomposition modes K is [3, 10], and the range of the penalty factor α is [300, 6000]. According to these parameters, the adaptive VMD algorithm is executed to obtain the fitness function change curve as Figure 3 shown. The optimal individual fitness E min = 24359, and the optimization result is as Figure 4 shown. The optimal number of modal decomposition layers K best = 10 and the optimal penalty factor α best = 3332.

[0071] Table 2 GWO algorithm parameters

[0072]

[0073] Decompose the signal using the VMD algorithm according to the optimized parameters. The envelope entropy value after decomposition is the smallest, and the decomposition result is as shown in Figure 5 . It can be seen that the frequency bands on each component are concentrated within a relatively narrow bandwidth, and there is a center frequency, and no mode mixing phenomenon occurs. The envelope entropy values of each component are shown in Table 3. The envelope entropy value of IMF10 is the smallest, which is 4.2435. Therefore, this component is selected as the optimal component.

[0074] Table 3 IMF envelope entropy values

[0075]

[0076] Perform Hilbert envelope demodulation on IMF10, and the extracted envelope spectrum line results are as shown in Figure 6 . From Figure 6 , it can be clearly observed that there are spectral peaks in multiples. The spectral peak values are 8.5 hz, 17 hz, and 25.5 hz respectively. This result indicates that the fundamental frequency signal of the propeller of this vehicle is 8.5 hz.

[0077] In order to amplify samples of different categories, first, intercept the audio in the database with a length of 10 s. Then, merge 11 types of ship types into 4 experimental categories according to the ship size, and add 1 type of natural background noise to construct a classification and recognition sample dataset as shown in Table 4. First, perform signal enhancement processing on it.

[0078] Table 4 Classification and recognition dataset

[0079]

[0080] In order to realize the amplification of samples of different categories by WGAN, use continuous wavelet transform to convert the one-dimensional time-domain waveform diagram of the audio into a two-dimensional time-frequency spectrum diagram, and then transform it into an amplification task for images. Select the complex Morlet wavelet as the wavelet function as shown in Equation (5), and set its bandwidth parameter Fb = 3 and the center frequency ω 0 = 3.

[0081]

[0082] In the formula: ψ(t) —— complex Morlet wavelet function; t —— time variable; σ —— Gaussian standard deviation; ω 0 —— center frequency of the wavelet.

[0083]

[0084] In the formula: scales —— wavelet transform scale; f —— signal frequency; totalscale —— total wavelet transform scale; k —— attenuation factor, gradually decreasing from the total scale to 1.

[0085] Calculate the wavelet transform scale group through formula (6) and generate wavelet basis functions. Subsequently, perform continuous wavelet transform on the obtained wavelet coefficients to plot the time-frequency spectrogram.

[0086] Perform continuous wavelet transform on the 90 audio-form data records in the ShipsEar library after classification and recognition to obtain time-frequency spectrograms, as Figure 7 The example shows the signal processing result of id = 15 in class A, and will Figure 7 Convert the time-domain waveform diagram in the form of the left figure in Figure 7 to the spectrogram shown in the right figure in

[0087] Perform the above preprocessing to convert the 90 audio records in the ShipsEar library into time-frequency spectrograms containing time-domain and frequency-domain features, forming an image dataset for 5-class classification. Use WGAN for data augmentation for each class of data, and set the training hyperparameters as shown in Table 5 below. The results are as Figure 8 shown, intuitively demonstrating the entire image augmentation process and its results.

[0088] Table 5 WGAN Hyperparameter Settings

[0089]

[0090]

[0091] In the initial stage of model training, when the number of training epochs reaches 180, it is observed that Figure 8 (b) The generated images already initially possess the general characteristics of underwater acoustic signals. However, there are still some obvious noise points in these images, which not only affect the clarity of the images but also hinder the model's effective capture and learning of the underwater acoustic signal characteristics. Therefore, it is necessary to further improve the quality of the generated images to better support the model training process.

[0092] As the training process progresses, gradually adjust and optimize the model's parameters and structure. When the number of training epochs increases to 20000, Figure 8 (c) The generated images shown present significant improvements. These images are more similar to the original images in structure, the waveforms and characteristics of the underwater acoustic signals are clearer, and there are no obvious noise interference points. This indicates that the model has been able to accurately capture the characteristics of the underwater acoustic signals and successfully generate high-quality images.

[0093] To more objectively evaluate the quality of the generated images, the Wasserstein distance is used as an evaluation metric, and the changes in the real sample scores and generated sample scores with the increase in the number of training epochs are plotted (as Figure 9 shown). FromFigure 9 It can be seen that as the number of training rounds increases, the scores of the real samples and the generated samples gradually approach. Especially when the number of training rounds reaches 2500, the score difference between the two gradually converges and approaches 0. This result indicates that the generated underwater acoustic signal image samples are already highly similar to the real samples in terms of quality and have reached a relatively high level.

[0094] In summary, through a large amount of training and optimization, the amplification process of underwater acoustic signal images has been successfully realized, and high-quality ship radiated noise signal images have been generated.

[0095] This invention uses the dataset after feature enhancement and data amplification for classification and recognition. Classification and recognition experiment of ship targets based on Swin-Transformer In this section, the research on classification and recognition methods is still carried out based on the data in the ShipsEar library. 90 original signals in the ShipsEar library are converted into spectrograms, and the signals of 5 categories are amplified to 500 time-frequency spectrogram images each. The amplified underwater acoustic sample dataset is divided into a training set, a validation set, and a test set according to the ratio of 7:2:1, and trained using the Swin-Transformer network. The training objective is to achieve a 5-classification recognition task, and the number of training rounds is 100. The training results using the Swin-Transformer network are as Figure 9 , Figure 9 The left figure in it depicts the change of the accuracy on the training set and the validation set as the number of training rounds increases. It can be seen that it converges around 60 rounds, reaches an accuracy of 100% on the test set, and reaches an accuracy of 95.8% on the validation set; Figure 9 The right figure in it describes the change of the loss function with the number of training rounds. It can be seen that the loss function converges on both the training set and the test set.

[0096] In another embodiment of this invention, a ship acoustic image classification system based on deep learning is provided, including a training and optimization module and a recognition module.

[0097] The training and optimization module trains and optimizes to generate a ship target classification and recognition model based on underwater acoustic signal samples with sample amplification and class number equalization.

[0098] The recognition module uses the ship target classification and recognition model to extract key features from the underwater acoustic signal to be recognized, and obtains the ship classification result corresponding to the underwater acoustic signal to be recognized according to the classification result corresponding to the extracted key features.

[0099] In another embodiment of the present invention, a terminal device is provided. The terminal device includes a processor and a memory. The memory is used to store a computer program, and the computer program includes program instructions. The processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function. The processor described in the embodiment of the present invention can be used for the operation of the ship acoustic image classification method based on deep learning, including the following steps: training and optimizing to generate a ship target classification and recognition model based on the underwater acoustic signal samples with sample amplification and class number equalization, using the ship target classification and recognition model to extract key features from the underwater acoustic signal to be recognized, and obtaining the ship classification result corresponding to the underwater acoustic signal to be recognized according to the classification result corresponding to the extracted key features.

[0100] In another embodiment of the present invention, a storage medium is also provided, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in the terminal device and is used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and, of course, the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, and the operating system of the terminal is stored in this storage space. And, one or more instructions suitable for being loaded and executed by the processor are also stored in this storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory.

[0101] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the method for classifying ship acoustic images based on deep learning in the above embodiments; the one or more instructions in the computer-readable storage medium are loaded and executed by the processor to perform the following steps: training and optimizing an underwater acoustic signal sample to generate a ship target classification and recognition model based on sample amplification and class number equalization, using the ship target classification and recognition model to extract key features from the underwater acoustic signal to be recognized, and obtaining the ship classification result corresponding to the underwater acoustic signal to be recognized according to the classification result corresponding to the extracted key features.

[0102] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented 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.

[0103] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0104] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0105] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are performed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide for implementing the functions in Figure 1 one flow or multiple flows and / or blocksFigure 1 Steps of functions specified in one or more boxes.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: the specific implementation manners of the present invention can still be modified or equivalently replaced, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A ship acoustic image classification method based on deep learning, characterized in that: The following steps are involved: S1, based on sample amplification and equalization of the number of categories, the underwater acoustic signal sample training optimization generates a ship target classification and recognition model; S2, using the ship target classification and recognition model to extract key features of the underwater acoustic signal to be identified, and obtaining the ship classification result corresponding to the underwater acoustic signal to be identified according to the classification result corresponding to the extracted key features.

2. According to a method for classifying ship acoustic images based on deep learning in claim 1, it is characterized in that: The underwater acoustic signal sample based on sample amplification and category quantity equalization includes the following steps: using continuous wavelet transform and Wasserstein generative adversarial network to generate underwater acoustic signal samples, and realizing sample amplification and category quantity equalization.

3. According to claim 2, a ship acoustic image classification method based on deep learning is characterized in that: Continuous wavelet transform is used to convert the one-dimensional time domain waveform of the underwater acoustic signal sample into a two-dimensional time-frequency spectrum, which is then converted into an image amplification task. The complex Morlet wavelet is used as the wavelet function, and then the wavelet transform scale group is calculated and the wavelet basis function is generated. Subsequently, the wavelet coefficients obtained by continuous wavelet transform are executed to draw the time-frequency spectrum.

4. The ship acoustic image classification method based on deep learning according to claim 3 is characterized in that: The bandwidth parameter of the complex Morlet wavelet function is Fb=3, and the center frequency ω0=3.

5. According to a method for classifying ship acoustic images based on deep learning according to claim 1, it is characterized in that: The variational modal decomposition method based on the Grey Wolf optimization algorithm is used to decompose the passive radiation noise signal of the ship target, and the optimal modal component of each ship target passive radiation noise signal is obtained, and the optimal component with the minimum envelope entropy value is obtained; Then the energy value of each modal component is calculated, and the two modal components with the highest energy in the remaining modal components are selected. The optimal component with the smallest envelope entropy value and the two modal components with the highest energy in the remaining modal components are reconstructed to form a new ship target signal and complete the key feature extraction.

6. The method for classifying ship acoustic images based on deep learning according to claim 2, characterized in that: The generated underwater acoustic signal samples and underwater acoustic signal samples are used to train the Swin-Transformer network; during the training process, the performance of the model is optimized by adjusting the learning rate and batch size parameters; the optimized model is verified by using the validation set to verify the trained model, and the ship target classification and recognition model can be obtained when the classification and recognition accuracy of the optimized model reaches the set value.

7. A ship acoustic image classification system based on deep learning, characterized in that: Including training optimization module and recognition module, Training optimization module, which generates ship target classification and recognition model based on the optimization of underwater acoustic signal sample training with sample amplification and equalization of the number of categories; The recognition module uses the ship target classification and recognition model to extract key features of the hydroacoustic signal to be identified, and obtains the ship classification result corresponding to the hydroacoustic signal to be identified based on the classification result corresponding to the extracted key features.

8. The ship acoustic image classification system based on deep learning according to claim 7, characterized in that: The underwater acoustic signal sample based on sample amplification and category quantity equalization includes the following steps: using continuous wavelet transform and Wasserstein generative adversarial network to generate underwater acoustic signal samples, and realizing sample amplification and category quantity equalization.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the ship acoustic image classification method based on deep learning as described in any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the ship acoustic image classification method based on deep learning as described in any one of claims 1 to 6 are implemented.

Images