Underwater sound target identification method, device, equipment, medium and product

Abstract

The invention provides an underwater acoustic target recognition method, device, equipment, medium and product, and the method comprises the steps: obtaining local features and global features corresponding to an underwater acoustic spectrum signal to be recognized, the local features being obtained by employing different feature extraction scales to carry out feature extraction on spectrum signals of different frequency bands in the underwater acoustic spectrum signal; fusing the local features and the global features to obtain multi-scale underwater acoustic features; and performing underwater sound target recognition based on the multi-scale underwater sound features to obtain an underwater sound target recognition result. According to the invention, the accuracy of underwater sound target recognition can be improved.

Description

Technical Field

[0001] The present application relates to the field of signal processing technology, and in particular to an underwater acoustic target recognition method, device, equipment, medium and product. Background Art

[0002] Underwater acoustic target detection and recognition is a key link in the field of underwater acoustic signal processing, and its core research object is the noise radiated by underwater targets.

[0003] At present, underwater acoustic target detection and recognition are mainly based on LOFAR (Low Frequency Analysis Recording) spectrum. However, due to the complexity of the underwater acoustic signal acquisition environment and the limitation of the working frequency band of the sonar equipment, there are differences in the distribution band of effective information of LOFAR features in different batches of data, which makes it difficult to accurately extract features and detect and recognize underwater acoustic targets. Therefore, effective feature extraction technology to improve the accuracy of underwater acoustic target detection has become a problem that needs to be solved urgently. Summary of the invention

[0004] Based on the above technical status, the present application proposes a method, device, equipment, medium and product for underwater acoustic target recognition, which can improve the accuracy of underwater acoustic target recognition.

[0005] In order to achieve the above technical objectives, this application specifically proposes the following technical solutions:

[0006] In a first aspect, the present application proposes a method for underwater acoustic target recognition, comprising: obtaining local features and global features corresponding to an underwater acoustic spectrum signal to be recognized, wherein the local features are obtained by extracting spectrum signals of different frequency bands in the underwater acoustic spectrum signal using different feature extraction scales; fusing the local features with the global features to obtain multi-scale underwater acoustic features; and performing underwater acoustic target recognition based on the multi-scale underwater acoustic features to obtain an underwater acoustic target recognition result.

[0007] In some implementations, obtaining local features corresponding to an underwater acoustic spectrum signal to be identified includes: dividing the underwater acoustic spectrum signal into a first frequency band signal and a second frequency band signal, the frequency of the first frequency band signal being greater than the frequency of the second frequency band signal; extracting coarse-grained local features from the first frequency band signal, and extracting fine-grained local features from the second frequency band signal; and splicing the coarse-grained local features and the fine-grained local features to obtain the local features corresponding to the underwater acoustic spectrum signal to be identified.

[0008] In some implementations, extracting fine-grained local features from the second frequency band signal includes: extracting fine-grained local features from the sub-signals of each frequency band in the second frequency band signal respectively, and splicing the fine-grained local features corresponding to the sub-signals of each frequency band in the second frequency band signal to obtain the fine-grained local features.

[0009] In some implementations, feature extraction scales corresponding to sub-signals in each frequency band of the second frequency band signal are the same or different.

[0010] In some implementations, the acquiring of local features and global features corresponding to the hydroacoustic spectrum signal to be identified, fusing the local features with the global features to obtain multi-scale hydroacoustic features, performing hydroacoustic target identification based on the multi-scale hydroacoustic features, and obtaining a hydroacoustic target identification result includes: inputting the hydroacoustic spectrum signal to be identified into a hydroacoustic target identification model to acquire the local features and global features corresponding to the hydroacoustic spectrum signal to be identified through the hydroacoustic target identification model, fusing the local features with the global features to obtain multi-scale hydroacoustic features, performing hydroacoustic target identification based on the multi-scale hydroacoustic features, and obtaining a hydroacoustic target identification result.

[0011] In some implementations, the hydroacoustic target recognition model includes a frequency band division unit, a coarse-grained feature extraction unit, a fine-grained feature extraction unit and a splicing unit; wherein, obtaining the local features corresponding to the hydroacoustic spectrum signal to be identified through the hydroacoustic target recognition model includes: dividing the hydroacoustic spectrum signal into a first frequency band signal and a second frequency band signal through the frequency band division unit, the frequency of the first frequency band signal being greater than the frequency of the second frequency band signal; extracting coarse-grained local features from the first frequency band signal through the coarse-grained feature extraction unit, and extracting fine-grained local features from the second frequency band signal through the fine-grained feature extraction unit; and splicing the coarse-grained local features and the fine-grained local features through the splicing unit to obtain the local features corresponding to the hydroacoustic spectrum signal to be identified.

[0012] In some implementations, the fine-grained feature extraction unit includes various fine-grained feature extraction units; wherein, the extracting of fine-grained local features from the second frequency band signal through the fine-grained feature extraction unit includes: extracting fine-grained local features from the sub-signals of each frequency band in the second frequency band signal through each fine-grained feature extraction unit, and splicing the fine-grained local features corresponding to the sub-signals of each frequency band in the second frequency band signal to obtain the fine-grained local features.

[0013] In some implementations, the hydroacoustic target recognition model is trained using the following steps: obtaining local feature samples and global feature samples corresponding to hydroacoustic spectrum signal samples through the hydroacoustic target recognition model to be trained, wherein the local feature samples are obtained by extracting features from spectrum signal samples of different frequency bands in the hydroacoustic spectrum signal samples using different feature extraction scales; fusing the local feature samples with the global feature samples to obtain multi-scale hydroacoustic feature samples, performing hydroacoustic target prediction based on the multi-scale hydroacoustic feature samples to obtain a predicted hydroacoustic target category; determining a hydroacoustic target prediction loss based on the difference between the predicted hydroacoustic target category and the labeled hydroacoustic target category, and converging the hydroacoustic target recognition model to be trained based on the hydroacoustic target prediction loss to obtain the hydroacoustic target recognition model.

[0014] The second aspect of the present application proposes an underwater acoustic target recognition device, comprising: an acquisition unit, used to acquire local features and global features corresponding to an underwater acoustic spectrum signal to be recognized, wherein the local features are obtained by extracting spectrum signals of different frequency bands in the underwater acoustic spectrum signal using different feature extraction scales; a fusion unit, used to fuse the local features with the global features to obtain multi-scale underwater acoustic features; and an identification unit, used to perform underwater acoustic target recognition based on the multi-scale underwater acoustic features to obtain an underwater acoustic target recognition result.

[0015] The third aspect of the present application proposes an electronic device, comprising a memory and a processor; the memory is connected to the processor and is used to store programs; the processor is used to implement the underwater acoustic target identification method described in the first aspect and any one of the implementation methods of the first aspect by running the program in the memory.

[0016] In a fourth aspect, the present application proposes a storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for underwater acoustic target recognition described in the first aspect and any one of the implementation methods of the first aspect is implemented.

[0017] In a fifth aspect, the present application proposes a computer program product, comprising computer program instructions, which, when executed by a processor, enable the processor to implement the underwater acoustic target recognition method described in the first aspect and any one of the implementation methods of the first aspect.

[0018] The method, device, equipment, medium and product for underwater acoustic target recognition proposed in the present application obtain local features and global features corresponding to the underwater acoustic spectrum signal to be identified, wherein the local features are obtained by extracting features from spectrum signals of different frequency bands in the underwater acoustic spectrum signal using different feature extraction scales; the local features are merged with the global features to obtain multi-scale underwater acoustic features; and underwater acoustic target recognition is performed based on the multi-scale underwater acoustic features to obtain underwater acoustic target recognition results. Since local features are obtained by extracting features from spectrum signals of different frequency bands in the underwater acoustic spectrum signal using different feature extraction scales, the integrity of feature extraction can be guaranteed; in addition, simultaneous monitoring of global features and local features can provide richer feature information and improve the diversity and robustness of features. When underwater acoustic target recognition is performed based on the multi-scale underwater acoustic features, the accuracy of underwater acoustic target recognition can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0020] Figure 1 A schematic diagram of a network model of a channel domain attention mechanism in related technologies;

[0021] Figure 2 A schematic diagram of a flow chart of an underwater acoustic target recognition method provided in an embodiment of the present application;

[0022] Figure 3 A schematic diagram of the framework of the underwater acoustic target recognition model provided in the embodiment of the present application;

[0023] Figure 4 A schematic diagram of a flow chart of a training process of an underwater acoustic target recognition model provided in an embodiment of the present application;

[0024] Figure 5 A schematic diagram of the structure of an underwater acoustic target recognition device provided in an embodiment of the present application;

[0025] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The technical solution proposed in the embodiment of the present application is applicable to any underwater acoustic target recognition application scenario, such as submarine detection and tracking, seabed topography mapping, mineral resource detection, marine life monitoring, marine pollution monitoring, underwater communication, underwater navigation, underwater search and rescue, etc. The technical solution of the embodiment of the present application can improve the accuracy of underwater acoustic target recognition.

[0027] The technical solution provided in the embodiments of the present application can be exemplarily applied to hardware devices such as processors, electronic devices, servers (including cloud servers), or packaged into software programs to be run. When the hardware device executes the processing of the technical solution of the embodiments of the present application, or the above-mentioned software program is run, the automatic splitting of the target task and the automatic calling of the application program interface required for the task can be achieved to complete the purpose of the target task. The embodiments of the present application only exemplarily introduce the specific processing of the technical solution of the present application, and do not limit the specific implementation form of the technical solution of the present application. Any technical implementation form that can execute the processing of the technical solution of the present application can be adopted by the embodiments of the present application.

[0028] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0029] Before introducing this application solution, the relevant technologies are first introduced:

[0030] Hydroacoustic target detection and recognition is an important part of the field of hydroacoustic signal processing. As an important means of modern ocean exploration, its main research object is the noise radiated by underwater targets, including mechanical noise, propeller noise and hydrodynamic noise. These noise sources specifically include mechanical noise: generated by the operation of mechanical equipment inside underwater targets, such as engines, pumps and generators; propeller noise: caused by the rotation of propellers, with periodic characteristics, is one of the main noise sources of ships and submarines; hydrodynamic noise: generated by the interaction between water flow and target surface, affected by flow velocity and target shape.

[0031] The detection and recognition of underwater acoustic targets are mainly based on two methods: traditional methods and deep learning. The underwater acoustic target recognition based on traditional methods is a commonly used technology at present, and its core lies in feature extraction and classifier design. Feature extraction mainly includes time-frequency features (such as time-frequency graphs) and auditory features (such as Mel cepstral coefficients, MFCC). The time-frequency features generate time-frequency graphs through time-frequency analysis methods (such as short-time Fourier transform, wavelet transform, etc.) to capture the changes of signals in time and frequency. Auditory features use the characteristics of the human auditory system to extract Mel cepstral coefficients. By performing Mel frequency filtering and discrete cosine transform on the signal, features related to human auditory perception are extracted. Classifier design includes methods such as support vector machine (SVM), Gaussian mixture model (GMM), decision tree and random forest, and these classifiers are used to classify different categories of underwater acoustic signals.

[0032] The deep learning-based method automatically learns complex feature representations to identify underwater acoustic targets. Feature extraction methods include convolutional neural networks (CNN), recurrent neural networks (RNN) and their variants (such as LSTM, GRU) and Transformer. CNN automatically extracts high-level features of underwater acoustic signals through multi-layer convolution and pooling operations; RNN and its variants capture the temporal dependencies of signals through memory mechanisms and are suitable for processing time series data; Transformer captures long-range dependencies in signals through self-attention mechanisms and is suitable for processing complex time series data. The classifier design uses fully connected layers and softmax layers to classify the extracted features.

[0033] In the field of deep learning, the network architecture based on the attention mechanism is currently the most widely used multi-scale solution. The attention module in the neural network is usually an additional neural network that can distinguish different parts of the input as needed and automatically assign different weights. The attention mechanism enables the neural network to distinguish key information and improve the processing of information that is helpful for classification and positioning in underwater acoustic signals, thereby improving the performance of underwater acoustic target detection networks.

[0034] The channel attention mechanism is one of the more typical attention mechanisms. The basic idea of ​​the self-attention mechanism is that when processing sequence data, each element can be associated with other elements in the sequence, rather than just relying on elements in adjacent positions. It adaptively captures the long-range dependencies between elements by calculating the relative importance between elements.

[0035] SENet (Squeeze and Excitation Networks) is a network model with a channel-domain attention mechanism. It models the importance of each channel in the feature map, reconstructs the channel features, pays more attention to the channels in the feature map that are more valuable to the detection task, and pays less attention to the channels that are less helpful or redundant to the detection task, thereby improving the performance of the model. SENet mainly consists of three parts: compression (Squeeze), activation (Excitation) and reconstruction (Scale). Reconstruct the original feature map U'=U·M c .in, represents the original feature map, Represents the feature map after channel attention reconstruction, and its network model is as follows Figure 1 As shown:

[0036] Compression part F sq (·): The purpose is to obtain a numerical value that can represent the importance of each channel. The implementation method is to give an input feature map U and let it pass through F sq (·) Do a transformation to get an output As shown in the following formula:

[0037]

[0038] F sq (·) The specific operation is to use global average pooling (GAP) on the input feature map, average the feature map of each channel to represent the entire channel, which is equivalent to having a global receptive field, that is, to achieve global low-dimensional embedding of U, by compressing the spatial information of each channel into the corresponding channel and converting it into a value. In this way, each pixel represents a channel, and the dimension of the final data becomes 1×1×C.

[0039] Activate Section F ex (·): Given an input Z, let it pass through F ex (·) to get an output Mc as shown in the following formula:

[0040] M c =F ex (Z,W)=σ(g(z,w))=σ(w2δ(w1 z))

[0041] Where δ(·) is the ReLu activation function, σ(·) is the activation function Sigmoid, and the final M c That is, it is the weight matrix that represents the importance of the channels in the original feature map. The specific operation is to pass Z obtained from the compressed part through two fully connected layers, corresponding to w1 and w2 in the figure. Two fully connected layers are used here to increase nonlinear ability and fitting ability.

[0042] Reconstruction Part F scale (·):The weight matrix M obtained using the activation part c , reconstruct the original feature map to get the output U', apply the obtained weight to each channel of U, that is, multiply each channel of U by the channel weight, and the reconstruction part of the feature map is completed, as shown in the following formula:

[0043] U'=F scale (U, Mc) = Mc × U;

[0044] In pattern recognition based on neural networks, the introduction of the attention mechanism allows the model to focus on key information and ignore irrelevant information. By applying a soft attention mechanism to the feature channel domain, the interdependence between feature channels is explicitly modeled. Specifically, the importance of each feature channel is automatically obtained through learning, and then useful features are enhanced according to this importance and features that are not very useful for the current task are suppressed. With the attention mechanism as the core, by studying the network model enhanced based on the attention mechanism, the key features of difficult samples can be captured and fine-grained and refined modeling can be achieved.

[0045] Current multi-scale technology based on attention mechanism often finds it difficult to capture the relationship between local and global variables for underwater acoustic signals with extremely low signal-to-noise ratio.

[0046] In summary, the underwater acoustic feature extraction method in the related art is relatively simple, resulting in a low accuracy rate of underwater acoustic target recognition. Therefore, how to accurately extract underwater acoustic signal features and improve the accuracy of underwater acoustic target detection and recognition is an important issue that needs to be solved.

[0047] In view of this, the present application proposes a method, device, equipment, medium and product for underwater acoustic target recognition. Based on the sensitivity of underwater acoustic signals in different frequency bands, effective features are processed through different feature extraction scales to extract the characterization information of the signal at a local fine-grained level, and at the same time, global information on all frequency bands is extracted to ensure that both local and global features of the model can be monitored during training, thereby effectively improving the accuracy of the model's underwater acoustic target recognition, and further improving the accuracy of underwater acoustic target recognition.

[0048] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0049] The present application embodiment first proposes a method for underwater acoustic target recognition, see Figure 2 As shown, the method includes steps S101 to S103:

[0050] S101, obtaining local features and global features corresponding to the underwater acoustic spectrum signal to be identified.

[0051] The underwater acoustic spectrum signal refers to the spectrum signal obtained from the underwater acoustic time domain signal to be identified through time-frequency conversion methods (such as Fourier transform). The underwater acoustic spectrum signal reflects the energy distribution of the underwater acoustic signal at different frequencies and is the basis for underwater acoustic target recognition.

[0052] The local features and global features corresponding to the underwater acoustic spectrum signal are both referred to as lofar features. Lofar features can still detect the radiation noise of underwater targets under low signal-to-noise ratio conditions, so they can be used as feature extraction of underwater acoustic signals.

[0053] Among them, local features are obtained by extracting features from spectrum signals of different frequency bands in the underwater acoustic spectrum signal using different feature extraction scales. Specifically, local features focus on the detailed information of the underwater acoustic spectrum signal in a specific frequency band. Global features refer to the features obtained by extracting features from the entire underwater acoustic spectrum signal. Global features focus on the overall information of the spectrum signal and reflect the overall characteristics of the underwater acoustic signal.

[0054] In some embodiments, obtaining the local features corresponding to the underwater acoustic spectrum signal to be identified includes the following steps a1 to a3:

[0055] Step a1: dividing the underwater acoustic spectrum signal into a first frequency band signal and a second frequency band signal, wherein the frequency of the first frequency band signal is greater than the frequency of the second frequency band signal.

[0056] Among them, the first frequency band signal can be understood as a high frequency band signal, and the second frequency band signal can be understood as a low frequency band signal. By dividing the underwater acoustic spectrum signal at a ratio of 1:1, the first frequency band signal and the second frequency band signal can be obtained. For example, if the spectrum range of the underwater acoustic spectrum signal is 200Hz-2MHz, then the underwater acoustic spectrum signal is divided at a ratio of 1:1 to obtain a second frequency band signal of 200Hz-999900Hz and a first frequency band signal of 999901Hz-2MHz.

[0057] Step a2: extracting coarse-grained local features from the first frequency band signal, and extracting fine-grained local features from the second frequency band signal.

[0058] Specifically, different feature extraction scales may be used to extract features from the first frequency band signal and the second frequency band signal, respectively, so as to obtain coarse-grained local features corresponding to the first frequency band signal and fine-grained local features corresponding to the second frequency band signal.

[0059] The different feature extraction scales include a first scale parameter and a second scale parameter. The first scale parameter is greater than the second scale parameter. The coarse-grained local features are extracted from the first frequency band signal, and the fine-grained local features are extracted from the second frequency band signal, including: extracting the coarse-grained local features from the first frequency band signal by the first scale parameter, and extracting the fine-grained local features from the second frequency band signal by the second scale parameter.

[0060] Specifically, extracting fine-grained local features from the second frequency band signal includes: extracting fine-grained local features from sub-signals of each frequency band in the second frequency band signal, respectively, and concatenating the fine-grained local features corresponding to the sub-signals of each frequency band in the second frequency band signal to obtain fine-grained local features. The feature extraction scales corresponding to the sub-signals of each frequency band in the second frequency band signal are the same or different.

[0061] Assuming that the feature extraction scale is the convolution window length, coarse-grained local features can be extracted through a coarse analysis method. For example, the spectrum of the first frequency band signal is segmented using a larger convolution window length, with each window covering a frequency range of 100 Hz and a time range of 1 second. In this way, the general trend of the underwater acoustic signal can be captured, such as the overall intensity change of a specific frequency component within a certain period of time, which can provide information about the macroscopic characteristics of the underwater acoustic signal.

[0062] At the same time, fine-grained local features can also be extracted through sophisticated analysis methods. For example, the spectrum of the second frequency band signal is segmented using a smaller convolution window length, with each window covering a frequency range of 10 Hz and a time range of 0.1 seconds. In this way, small changes and details in the underwater acoustic signal can be captured, such as rapid fluctuations of specific frequency components in a short period of time. Fine-grained features help improve the accuracy of underwater acoustic target recognition.

[0063] Step a3: splice the coarse-grained local features and the fine-grained local features to obtain the local features corresponding to the underwater acoustic spectrum signal to be identified.

[0064] For example, assuming that a frequency band of a 10-second 1000Hz underwater acoustic spectrum signal is selected as the analysis object on the spectrum graph, the final local features include: coarse-grained features extracted from 10 seconds 1000Hz based on the first window length, and fine-grained features extracted from 10 seconds 1000Hz based on the second window length. By combining these two parts of features, the local features of the underwater acoustic spectrum signal can be obtained. The first window length is greater than the second window length.

[0065] By splicing coarse-grained local features and fine-grained local features, it is possible to retain the characteristics at a larger scale (coarse-grained features) and add detailed information at a smaller scale (fine-grained features) for each position on the spectrum. In this way, a more comprehensive and rich representation of underwater acoustic signal features can be obtained, which can improve the accuracy of underwater acoustic target recognition.

[0066] S102: Fusing local features and global features to obtain multi-scale underwater acoustic features.

[0067] Local features and global features are obtained by extracting features of underwater acoustic signals based on different feature extraction scales, so multi-scale underwater acoustic features can be obtained. Multi-scale underwater acoustic features can more comprehensively describe the characteristics of underwater acoustic signals from different angles and enhance the expressiveness of features.

[0068] S103, performing underwater acoustic target recognition based on multi-scale underwater acoustic features to obtain an underwater acoustic target recognition result.

[0069] Among them, step S101, step S102 and step S103 can be implemented by an underwater acoustic target recognition model, which specifically includes: inputting the underwater acoustic spectrum signal to be recognized into the underwater acoustic target recognition model, so as to obtain the local features and global features corresponding to the underwater acoustic spectrum signal to be recognized through the underwater acoustic target recognition model, fusing the local features and the global features to obtain multi-scale underwater acoustic features, and performing underwater acoustic target recognition based on the multi-scale underwater acoustic features to obtain an underwater acoustic target recognition result.

[0070] This embodiment divides the underwater acoustic spectrum signal into high-frequency bands and low-frequency bands, extracts coarse-grained local features and fine-grained local features respectively, and then splices these features to obtain multi-scale underwater acoustic features. Through multi-scale feature fusion, a more comprehensive and rich feature representation can be provided, thereby improving the accuracy of underwater acoustic target recognition.

[0071] The detailed process of implementing the above steps S101, S102 and S103 by the underwater acoustic target recognition model is introduced in sequence below in conjunction with the framework schematic diagram of the underwater acoustic target recognition model:

[0072] Figure 3 A schematic diagram of the structure of an underwater acoustic target recognition model provided in an embodiment of the present application. Figure 3 As shown, the underwater acoustic target recognition model includes a frequency band division unit 31, a coarse-grained feature extraction unit 32, a fine-grained feature extraction unit 33 and a splicing unit 34.

[0073] In step S101, local features corresponding to the underwater acoustic spectrum signal to be identified are obtained through an underwater acoustic target recognition model, including steps b1, b2 and b3:

[0074] Step b1, dividing the underwater acoustic spectrum signal into a first frequency band signal and a second frequency band signal by the frequency band dividing unit 31, wherein the frequency of the first frequency band signal is greater than the frequency of the second frequency band signal.

[0075] Before step b1, it is necessary to extract the lofar feature from the underwater acoustic time domain signal, and the lofar feature is a spectrum feature. After that, the lofar feature is vectorized by the vectorization and frequency band division module NN-block_1 to obtain the Embedding, and the Embedding corresponding to the underwater acoustic spectrum signal is divided into the Embedding corresponding to the first frequency band signal and the Embedding corresponding to the second frequency band signal. The first frequency band signal and the second frequency band signal are respectively used as the local coarse granularity band and the local fine granularity band.

[0076] The division rule of the first frequency band signal and the second frequency band signal can be found in the introduction of the aforementioned embodiment, which will not be described again here.

[0077] Step b2: extracting coarse-grained local features from the first frequency band signal through the coarse-grained feature extraction unit 32, and extracting fine-grained local features from the second frequency band signal through the fine-grained feature extraction unit 33.

[0078] Among them, the coarse-grained feature extraction unit 32 and the fine-grained feature extraction unit 33 can be a pooling layer, BatchNormalization, Rule, or a convolution layer.

[0079] Taking the pooling layer as an example, the coarse-grained feature extraction unit 32 includes the pooling layer. The lofar features corresponding to the local coarse-grained frequency band signal are input into the pooling layer for feature extraction, and a coarse-grained local feature can be obtained, which is recorded as Embedding_seg_1.

[0080] The fine-grained feature extraction unit 33 includes a pooling layer. The lofar features corresponding to the local fine-grained frequency band signals are input into the pooling layer for feature extraction, and fine-grained local features can be obtained, which are recorded as Embedding_seg_2.

[0081] Step b3: The coarse-grained local features and the fine-grained local features are spliced ​​by the splicing unit 34 to obtain the local features corresponding to the underwater acoustic spectrum signal to be identified.

[0082] In some embodiments, the concatenation unit 34 may be a concatenation network. The concatenation unit 34 concatenates Embedding_seg_1 and Embedding_seg_2 to obtain a local feature, which is recorded as Embedding_seg.

[0083] The fine-grained feature extraction unit 33 includes various fine-grained feature extraction units 331. The fine-grained feature extraction unit extracts fine-grained local features from the second frequency band signal, including: extracting fine-grained local features from the sub-signals of each frequency band in the second frequency band signal respectively through each fine-grained feature extraction unit 331, and splicing the fine-grained local features corresponding to the sub-signals of each frequency band in the second frequency band signal to obtain fine-grained local features. The feature extraction scales of various fine-grained feature extraction units 331 can be completely the same, partially the same, partially different, or completely different.

[0084] Continue reading Figure 3 Each fine-grained feature extraction unit 331 includes a pooling layer. The lofar features corresponding to the local fine-grained frequency band signal are input into each pooling layer for feature extraction, and each fine-grained local feature can be obtained, which is recorded as Embedding_seg_2. Then, the Embedding_seg_1 and each Embedding_seg_2 are spliced ​​by the splicing unit 34 to obtain a local feature, which is recorded as Embedding_seg.

[0085] The above embodiment introduces the process of extracting local features. In addition, the whole Embedding can also be subjected to feature extraction through a pooling layer to obtain a global feature, which is recorded as Embedding_whol.

[0086] Afterwards, by fusing the local feature Embedding_seg and the global feature Embedding_whol, a multi-scale hydroacoustic feature can be obtained. The multi-scale hydroacoustic feature is input into NN-block_2 for hydroacoustic target recognition, and the hydroacoustic target recognition result can be obtained.

[0087] In some embodiments, NN-block_2 may be a classifier.

[0088] This embodiment divides the underwater acoustic spectrum signal into a first frequency band signal and a second frequency band signal, and uses different feature extraction scales for the first frequency band signal and the second frequency band signal to perform feature extraction, thereby making full use of the characteristics of different frequency bands, making the features of the low frequency band and the high frequency band complementary, and providing more comprehensive and richer feature information, thereby improving the accuracy and robustness of underwater acoustic target recognition.

[0089] The above embodiment introduces the reasoning and application process of the underwater acoustic target recognition model. Before the reasoning and application of the underwater acoustic target recognition model, it is necessary to train it. The following is an introduction to the training process of the underwater acoustic target recognition model in conjunction with the accompanying drawings:

[0090] Figure 4The following is a flow chart of the training process of the underwater acoustic target recognition model provided in the embodiment of the present application. Figure 4 As shown, the training process of the underwater acoustic target recognition model includes the following steps S401 to S403:

[0091] S401. Obtain local feature samples and global feature samples corresponding to underwater acoustic spectrum signal samples through the underwater acoustic target recognition model to be trained.

[0092] The local feature samples are obtained by extracting features from spectrum signal samples in different frequency bands of the underwater acoustic spectrum signal samples using different feature extraction scales.

[0093] In this embodiment, the local feature samples and the global feature samples corresponding to the underwater acoustic spectrum signal samples both refer to lofar features.

[0094] Among them, the framework of the underwater acoustic target recognition model to be trained can be referred to Figure 3 Framework of the underwater acoustic target recognition model.

[0095] Step S401 specifically includes the following steps c1 to c3:

[0096] Step c1: dividing the underwater acoustic spectrum signal samples into first frequency band signal samples and second frequency band signal samples by a frequency band dividing unit, wherein the frequency of the first frequency band signal samples is greater than the frequency of the second frequency band signal samples.

[0097] Before step c1, it is necessary to extract lofar feature samples from the underwater acoustic time domain signal samples, and the lofar feature samples are spectrum feature samples. Afterwards, the lofar feature samples are vectorized by the vectorization and frequency band division module NN-block_1' to be trained to obtain Embedding', and the Embedding' corresponding to the underwater acoustic spectrum signal sample is divided into Embedding' corresponding to the first frequency band signal sample and Embedding' corresponding to the second frequency band signal sample. The first frequency band signal sample and the second frequency band signal sample are respectively used as local coarse-grained frequency band samples and local fine-grained frequency band samples.

[0098] The division rule of the first frequency band signal samples and the second frequency band signal samples can be found in the introduction of the above-mentioned embodiment, which will not be described in detail here.

[0099] Step c2: extracting coarse-grained local feature samples from the first frequency band signal samples by the coarse-grained feature extraction unit to be trained, and extracting fine-grained local feature samples from the second frequency band signal samples by the fine-grained feature extraction unit to be trained.

[0100] Among them, the coarse-grained feature extraction unit to be trained and the fine-grained feature extraction unit to be trained can be a pooling layer, a Batch Normalization, a Rule, and a convolutional layer.

[0101] Taking the pooling layer as an example, the coarse-grained feature extraction unit to be trained includes the pooling layer. The lofar feature samples corresponding to the local coarse-grained frequency band signal samples are input into the pooling layer for feature extraction, and a coarse-grained local feature sample can be obtained, which is recorded as Embedding_seg_1'.

[0102] The fine-grained feature extraction unit to be trained includes a pooling layer. The lofar feature samples corresponding to the local fine-grained frequency band signal samples are input into the pooling layer for feature extraction, and fine-grained local feature samples can be obtained, which are recorded as Embedding_seg_2'.

[0103] Step c3: splicing the coarse-grained local feature samples and the fine-grained local feature samples through the splicing unit to be trained to obtain the local feature samples corresponding to the underwater acoustic spectrum signal samples to be identified.

[0104] In some embodiments, the concatenation unit to be trained may be a concatenation network. By concatenating Embedding_seg_1' and Embedding_seg_2' through the concatenation unit to be trained, a local feature sample may be obtained, which is recorded as Embedding_seg'.

[0105] The fine-grained feature extraction unit to be trained includes each fine-grained feature extraction unit to be trained. The fine-grained feature extraction unit to be trained extracts fine-grained local feature samples from the second frequency band signal sample, including: extracting fine-grained local feature samples from the sub-signal samples of each frequency band in the second frequency band signal sample respectively through each fine-grained feature extraction unit to be trained, and splicing the fine-grained local feature samples corresponding to the sub-signal samples of each frequency band in the second frequency band signal sample to obtain fine-grained local feature samples.

[0106] Each fine-grained feature extraction unit to be trained includes a pooling layer. The lofar feature samples corresponding to the local fine-grained frequency band signal samples are input into each pooling layer for feature extraction, and each fine-grained local feature sample can be obtained, which is recorded as Embedding_seg_2'. Then, the Embedding_seg_1' and each Embedding_seg_2' are spliced ​​through the splicing unit to be trained to obtain a local feature sample, which is recorded as Embedding_seg'.

[0107] The above embodiment introduces the extraction process of local feature samples. In addition, the whole Embedding sample can also be subjected to feature extraction through a pooling layer to obtain a global feature sample, which is recorded as Embedding_whol'.

[0108] In some embodiments, the NN-block_2 to be trained may be a classifier to be trained.

[0109] S402: Fusing the local feature samples and the global feature samples to obtain multi-scale hydroacoustic feature samples, and performing hydroacoustic target prediction based on the multi-scale hydroacoustic feature samples to obtain a predicted hydroacoustic target category.

[0110] By fusing the local feature sample Embedding_seg' and the global feature sample Embedding_whol', a multi-scale hydroacoustic feature sample can be obtained. The multi-scale hydroacoustic feature sample is input into the NN-block_2 to be trained for prediction of hydroacoustic target recognition, and the predicted hydroacoustic target recognition result can be obtained.

[0111] S403, determining a hydroacoustic target prediction loss based on a difference between the predicted hydroacoustic target category and the labeled hydroacoustic target category, and converging a hydroacoustic target recognition model to be trained based on the hydroacoustic target prediction loss to obtain a hydroacoustic target recognition model.

[0112] By calculating the difference between the predicted underwater acoustic target category and the actual labeled category, the prediction error of the model can be quantified. This quantified error is the "underwater acoustic target prediction loss". This loss value not only reflects the prediction ability of the model in the current state, but also is an important basis for guiding the subsequent learning direction of the model.

[0113] In order to minimize this prediction loss, optimization algorithms such as gradient descent can be used to gradually adjust the parameters in the model so that the model can better fit the characteristics of the underwater acoustic target in the training data set. As the training process continues, the prediction loss of the underwater acoustic target will gradually decrease, indicating that the model's recognition accuracy of the underwater acoustic target is improving. When the recognition accuracy of the model reaches the preset accuracy or the change in the loss value tends to be stable, the model is considered to have converged, and the model obtained at this time is an optimized and trained underwater acoustic target recognition model.

[0114] In summary, the technical solution proposed in the present application starts from the information of different frequency bands of the LOFAR features of the underwater acoustic signal, and adopts different feature extraction scales to extract LOFAR features from local and global frequency band features, thereby ensuring the integrity of feature information extraction at different granularities and improving the richness of the underwater acoustic feature representation, so that both global and local features can be monitored during the model training process. Among them, local features help the model capture subtle changes and local patterns in the underwater acoustic signal, and global features help the model understand the overall structure and trend of the underwater acoustic signal. Combining local features with global features can improve the richness of features, thereby improving the accuracy and reliability of underwater acoustic target detection and recognition.

[0115] Corresponding to the above-mentioned underwater acoustic target recognition method, the embodiment of the present application also proposes an underwater acoustic target recognition device, such as Figure 5 As shown, the device includes: an acquisition unit 501, a fusion unit 502 and an identification unit 503; wherein the acquisition unit 501 is used to acquire local features and global features corresponding to the hydroacoustic spectrum signal to be identified, wherein the local features are obtained by extracting the spectrum signals of different frequency bands in the hydroacoustic spectrum signal using different feature extraction scales; the fusion unit 502 is used to fuse the local features with the global features to obtain multi-scale hydroacoustic features; the identification unit 503 is used to perform hydroacoustic target identification based on the multi-scale hydroacoustic features to obtain a hydroacoustic target identification result.

[0116] In some embodiments, the acquisition unit 501 is used to acquire local features corresponding to the underwater acoustic spectrum signal to be identified, specifically including: dividing the underwater acoustic spectrum signal into a first frequency band signal and a second frequency band signal, the frequency of the first frequency band signal being greater than the frequency of the second frequency band signal; extracting coarse-grained local features from the first frequency band signal, and extracting fine-grained local features from the second frequency band signal; splicing the coarse-grained local features and the fine-grained local features to obtain the local features corresponding to the underwater acoustic spectrum signal to be identified.

[0117] In some embodiments, the acquisition unit 501 is used to extract fine-grained local features from the second frequency band signal, including: extracting fine-grained local features from the sub-signals of each frequency band in the second frequency band signal respectively, and splicing the fine-grained local features corresponding to the sub-signals of each frequency band in the second frequency band signal to obtain the fine-grained local features.

[0118] In some embodiments, the feature extraction scales corresponding to the sub-signals of each frequency band in the second frequency band signal are the same or different.

[0119] In some embodiments, the acquisition unit 501 is used to acquire local features and global features corresponding to the hydroacoustic spectrum signal to be identified, the fusion unit 502 is used to fuse the local features with the global features to obtain multi-scale hydroacoustic features, and the recognition unit 503 is used to perform hydroacoustic target recognition based on the multi-scale hydroacoustic features to obtain hydroacoustic target recognition results, including: inputting the hydroacoustic spectrum signal to be identified into a hydroacoustic target recognition model to acquire local features and global features corresponding to the hydroacoustic spectrum signal to be identified through the hydroacoustic target recognition model, fusing the local features with the global features to obtain multi-scale hydroacoustic features, and performing hydroacoustic target recognition based on the multi-scale hydroacoustic features to obtain hydroacoustic target recognition results.

[0120] In some embodiments, the hydroacoustic target recognition model includes a frequency band division unit, a coarse-grained feature extraction unit, a fine-grained feature extraction unit and a splicing unit; wherein the acquisition unit 501 is used to obtain the local features corresponding to the hydroacoustic spectrum signal to be identified through the hydroacoustic target recognition model, including: dividing the hydroacoustic spectrum signal into a first frequency band signal and a second frequency band signal through the frequency band division unit, the frequency of the first frequency band signal being greater than the frequency of the second frequency band signal; extracting coarse-grained local features from the first frequency band signal through the coarse-grained feature extraction unit, and extracting fine-grained local features from the second frequency band signal through the fine-grained feature extraction unit; splicing the coarse-grained local features and the fine-grained local features through the splicing unit to obtain the local features corresponding to the hydroacoustic spectrum signal to be identified.

[0121] In some embodiments, the fine-grained feature extraction unit includes various fine-grained feature extraction units; wherein, extracting fine-grained local features from the second frequency band signal through the fine-grained feature extraction unit includes: extracting fine-grained local features from the sub-signals of each frequency band in the second frequency band signal through each fine-grained feature extraction unit, and splicing the fine-grained local features corresponding to the sub-signals of each frequency band in the second frequency band signal to obtain the fine-grained local features.

[0122] In some embodiments, the hydroacoustic target recognition model is trained by the following steps: obtaining local feature samples and global feature samples corresponding to the hydroacoustic spectrum signal samples through the hydroacoustic target recognition model to be trained, wherein the local feature samples are obtained by extracting features from spectrum signal samples of different frequency bands in the hydroacoustic spectrum signal samples using different feature extraction scales; fusing the local feature samples with the global feature samples to obtain multi-scale hydroacoustic feature samples, performing hydroacoustic target prediction based on the multi-scale hydroacoustic feature samples to obtain a predicted hydroacoustic target category; determining the hydroacoustic target prediction loss based on the difference between the predicted hydroacoustic target category and the labeled hydroacoustic target category, and converging the hydroacoustic target recognition model to be trained based on the hydroacoustic target prediction loss to obtain the hydroacoustic target recognition model.

[0123] The underwater acoustic target recognition device provided in this embodiment belongs to the same application concept as the underwater acoustic target recognition method provided in the above embodiments of this application, and can execute the underwater acoustic target recognition method provided in any of the above embodiments of this application, and has the corresponding functional modules and beneficial effects of executing the underwater acoustic target recognition method. For technical details not fully described in this embodiment, please refer to the specific processing content of the underwater acoustic target recognition method provided in the above embodiments of this application, and will not be repeated here.

[0124] The functions implemented by each unit in the above-mentioned underwater acoustic target recognition device can be implemented by the same or different processors respectively, which is not limited in the embodiment of the present application.

[0125] It should be understood that the units in the above underwater acoustic target recognition device can be implemented in the form of a processor calling software. For example, the device includes a processor, the processor is connected to a memory, and instructions are stored in the memory. The processor calls the instructions stored in the memory to implement any of the above methods or realize the functions of each unit of the device, wherein the processor can be a general-purpose processor, such as a CPU or a microprocessor, etc., and the memory can be a memory in the device or a memory outside the device. Alternatively, the units in the device can be implemented in the form of hardware circuits, and the functions of some or all units can be realized by designing the hardware circuits. The hardware circuit can be understood as one or more processors; for example, in one implementation, the hardware circuit is an ASIC, and the functions of some or all of the above units are realized by designing the logical relationship of the components in the circuit; for another example, in another implementation, the hardware circuit can be implemented by PLD, taking FPGA as an example, which can include a large number of logic gate circuits, and the connection relationship between the logic gate circuits is configured by the configuration file, so as to realize the functions of some or all of the above units. All units of the above device can be implemented in the form of a processor calling software, or in the form of hardware circuits, or in the form of a processor calling software, and the remaining part can be implemented in the form of hardware circuits.

[0126] In an embodiment of the present application, a processor is a circuit with the ability to process signals. In one implementation, the processor may be a circuit with the ability to read and run instructions, such as a CPU, a microprocessor, a GPU, or a DSP; in another implementation, the processor may implement certain functions through the logical relationship of a hardware circuit, and the logical relationship of the hardware circuit is fixed or reconfigurable, such as a hardware circuit implemented by an ASIC or PLD, such as an FPGA. In a reconfigurable hardware circuit, the process of the processor loading a configuration document to implement the hardware circuit configuration can be understood as the process of the processor loading instructions to implement the functions of some or all of the above units. In addition, it can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as an NPU, TPU, DPU, etc.

[0127] It can be seen that each unit in the above device can be one or more processors (or processing circuits) configured to implement the above method, such as: CPU, GPU, NPU, TPU, DPU, microprocessor, DSP, ASIC, FPGA, or a combination of at least two of these processor forms.

[0128] In addition, all or part of the units in the above device can be integrated together, or can be implemented independently. In one implementation, these units are integrated together and implemented in the form of a SOC. The SOC may include at least one processor for implementing any of the above methods or implementing the functions of each unit of the device. The type of the at least one processor may be different, for example, including a CPU and an FPGA, a CPU and an artificial intelligence processor, a CPU and a GPU, etc.

[0129] An embodiment of the present application further provides a control device, which includes a processor and an interface circuit. The processor in the control device is connected to a data input component via the interface circuit of the control device.

[0130] The data input component specifically refers to a functional component capable of inputting or collecting voice data, such as a microphone, etc.

[0131] The above-mentioned interface circuit can be any interface circuit that can realize the data communication function, for example, it can be a USB interface circuit, a Type-C interface circuit, a serial port circuit, a PCIE circuit, etc.

[0132] The processor in the control device is also a circuit with signal processing capability, which executes the underwater acoustic target recognition method introduced in the above embodiment. The specific implementation of the processor can refer to the above processor implementation, and the embodiment of the present application is not strictly limited.

[0133] Another embodiment of the present application also provides an electronic device, see Figure 6 As shown, the device includes:

[0134] Memory 200 and processor 210;

[0135] The memory 200 is connected to the processor 210 and is used to store programs;

[0136] The processor 210 is used to implement the underwater acoustic target recognition method disclosed in any of the above embodiments by running the program stored in the memory 200.

[0137] Specifically, the electronic device may further include: a bus, a communication interface 220 , an input device 230 and an output device 240 .

[0138] The processor 210, the memory 200, the communication interface 220, the input device 230 and the output device 240 are connected to each other via a bus.

[0139] A bus may include a pathway that transfers information between components of a computer system.

[0140] The processor 210 may be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, etc., or an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of the program of the scheme of the present invention. It may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0141] The processor 210 may include a main processor, and may also include a baseband chip, a modem, and the like.

[0142] The memory 200 stores a program for executing the technical solution of the present invention, and may also store an operating system and other key services. Specifically, the program may include a program code, and the program code includes a computer operation instruction. More specifically, the memory 200 may include a read-only memory (ROM), other types of static storage devices that can store static information and instructions, a random access memory (RAM), other types of dynamic storage devices that can store information and instructions, a disk storage, a flash, and the like.

[0143] The input device 230 may include a device for receiving data and information input by a user, such as a keyboard, a mouse, a camera, a scanner, a light pen, a voice input device, a touch screen, a pedometer, or a gravity sensor.

[0144] Output device 240 may include devices that allow information to be output to a user, such as a display screen, printer, speaker, etc.

[0145] The communication interface 220 may include any transceiver or the like to communicate with other devices or communication networks, such as Ethernet, Radio Access Network (RAN), Wireless Local Area Network (WLAN), etc.

[0146] The processor 210 executes the program stored in the memory 200 and calls other devices, which can be used to implement each step of any underwater acoustic target recognition method provided in the above embodiments of the present application.

[0147] An embodiment of the present application also proposes a chip, which includes a processor and a data interface. The processor reads and runs a program stored in a memory through the data interface to execute the underwater acoustic target recognition method introduced in any of the above embodiments. The specific processing process and its beneficial effects can be found in the introduction to the embodiment of the above-mentioned underwater acoustic target recognition method.

[0148] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the underwater acoustic target identification method described in any of the above-mentioned embodiments of this specification.

[0149] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user computing device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0150] In addition, an embodiment of the present application may also be a storage medium on which a computer program is stored, and the computer program is executed by a processor to execute the steps of the underwater acoustic target identification method described in any of the above embodiments of this specification.

[0151] For the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0152] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0153] The steps in the methods of each embodiment of the present application can be adjusted in order, combined and deleted according to actual needs, and the technical features recorded in each embodiment can be replaced or combined.

[0154] The modules and sub-modules in the devices and terminals of the various embodiments of the present application can be combined, divided and deleted according to actual needs.

[0155] In the several embodiments provided in the present application, it should be understood that the disclosed terminals, devices and methods can be implemented in other ways. For example, the terminal embodiments described above are only schematic, for example, the division of modules or submodules is only a logical function division, and there may be other division methods in actual implementation, for example, multiple submodules or modules can be combined or integrated into another module, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0156] The modules or submodules described as separate components may or may not be physically separated, and the components of the modules or submodules may or may not be physical modules or submodules, that is, they may be located in one place, or they may be distributed on multiple network modules or submodules. Some or all of the modules or submodules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0157] In addition, each functional module or submodule in each embodiment of the present application may be integrated into one processing module, or each module or submodule may exist physically separately, or two or more modules or submodules may be integrated into one module. The above-mentioned integrated modules or submodules may be implemented in the form of hardware or in the form of software functional modules or submodules.

[0158] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0159] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly by hardware, software units executed by a processor, or a combination of the two. The software units may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0160] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0161] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for underwater acoustic target recognition, characterized in that: include: Acquire local features and global features corresponding to the underwater acoustic spectrum signal to be identified, wherein the local features are obtained by extracting the spectrum signals of different frequency bands in the underwater acoustic spectrum signal using different feature extraction scales; The local features and the global features are combined to obtain multi-scale hydroacoustic features; Based on the multi-scale hydroacoustic features, hydroacoustic target recognition is performed to obtain a hydroacoustic target recognition result.

2. The method according to claim 1, characterized in that Obtain the local features corresponding to the underwater acoustic spectrum signal to be identified, including: Dividing the underwater acoustic spectrum signal into a first frequency band signal and a second frequency band signal, wherein the frequency of the first frequency band signal is greater than the frequency of the second frequency band signal; Extracting coarse-grained local features from the first frequency band signal, and extracting fine-grained local features from the second frequency band signal; The coarse-grained local features and the fine-grained local features are concatenated to obtain the local features corresponding to the underwater acoustic spectrum signal to be identified.

3. The method according to claim 2, characterized in that Extracting fine-grained local features from the second frequency band signal includes: Fine-grained local features are extracted from the sub-signals of each frequency band in the second frequency band signal respectively, and the fine-grained local features corresponding to the sub-signals of each frequency band in the second frequency band signal are concatenated to obtain the fine-grained local features.

4. The method according to claim 3, characterized in that The feature extraction scales corresponding to the sub-signals of each frequency band in the second frequency band signal are the same or different.

5. The method according to claim 1, characterized in that The method of acquiring local features and global features corresponding to the hydroacoustic spectrum signal to be identified, fusing the local features with the global features to obtain multi-scale hydroacoustic features, and performing hydroacoustic target identification based on the multi-scale hydroacoustic features to obtain a hydroacoustic target identification result includes: The hydroacoustic spectrum signal to be identified is input into a hydroacoustic target recognition model, so as to obtain local features and global features corresponding to the hydroacoustic spectrum signal to be identified through the hydroacoustic target recognition model, and the local features and the global features are fused to obtain multi-scale hydroacoustic features. Hydroacoustic target recognition is performed based on the multi-scale hydroacoustic features to obtain a hydroacoustic target recognition result.

6. The method according to claim 5, characterized in that The underwater acoustic target recognition model includes a frequency band division unit, a coarse-grained feature extraction unit, a fine-grained feature extraction unit and a splicing unit; Wherein, obtaining the local features corresponding to the underwater acoustic spectrum signal to be identified through the underwater acoustic target recognition model includes: The underwater acoustic spectrum signal is divided into a first frequency band signal and a second frequency band signal by the frequency band division unit, wherein the frequency of the first frequency band signal is greater than the frequency of the second frequency band signal; Extracting coarse-grained local features from the first frequency band signal by the coarse-grained feature extraction unit, and extracting fine-grained local features from the second frequency band signal by the fine-grained feature extraction unit; The coarse-grained local features and the fine-grained local features are spliced ​​by the splicing unit to obtain the local features corresponding to the underwater acoustic spectrum signal to be identified.

7. The method according to claim 6, characterized in that The fine-grained feature extraction unit includes various fine-grained feature extraction units; The step of extracting fine-grained local features from the second frequency band signal by a fine-grained feature extraction unit includes: The fine-grained local features are extracted from the sub-signals of each frequency band in the second frequency band signal by respective fine-grained feature extraction units, and the fine-grained local features corresponding to the sub-signals of each frequency band in the second frequency band signal are spliced ​​to obtain the fine-grained local features.

8. The method according to any one of claims 5 to 7, characterized in that: The underwater acoustic target recognition model is trained by the following steps: Obtaining local feature samples and global feature samples corresponding to the underwater acoustic spectrum signal samples through the underwater acoustic target recognition model to be trained, wherein the local feature samples are obtained by extracting features from spectrum signal samples of different frequency bands in the underwater acoustic spectrum signal samples using different feature extraction scales; The local feature samples and the global feature samples are fused to obtain a multi-scale hydroacoustic feature sample, and a hydroacoustic target is predicted based on the multi-scale hydroacoustic feature sample to obtain a predicted hydroacoustic target category; Based on the difference between the predicted hydroacoustic target category and the marked hydroacoustic target category, the hydroacoustic target prediction loss is determined, and based on the hydroacoustic target prediction loss, the hydroacoustic target recognition model to be trained is converged to obtain the hydroacoustic target recognition model.

9. An underwater acoustic target recognition device, characterized in that: include: An acquisition unit, used for acquiring local features and global features corresponding to the underwater acoustic spectrum signal to be identified, wherein the local features are obtained by extracting the spectrum signals of different frequency bands in the underwater acoustic spectrum signal using different feature extraction scales; A fusion unit, used for fusing the local features with the global features to obtain multi-scale hydroacoustic features; The recognition unit is used to perform hydroacoustic target recognition based on the multi-scale hydroacoustic features to obtain a hydroacoustic target recognition result.

10. An electronic device, characterized in that: including memory and processor; The memory is connected to the processor and is used to store programs; The processor is configured to implement the method according to any one of claims 1 to 8 by running the program in the memory.

11. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the method according to any one of claims 1 to 8 is implemented.

12. A computer program product, characterized in that The method comprises computer program instructions, which, when executed by a processor, cause the processor to implement the method as claimed in any one of claims 1 to 8.

Images