Underwater acoustic signal identification method based on cross attention feature fusion

Abstract

The invention provides an underwater acoustic signal identification method based on cross attention feature fusion, and belongs to the technical field of underwater acoustic signal processing. In the noise reduction module, a double-branch structure is used for processing amplitude and phase information respectively, an information interaction mode is adopted, information of the other path can be used as reference in the amplitude or phase processing process, and therefore the feature representation capacity is improved; the amplitude and phase information obtained by the noise reduction module is dynamically fused by using the feature fusion module, so that the problem that only the amplitude information is used and the phase feature is ignored is solved, and the robustness of the system is enhanced while the information loss is remarkably reduced; in the recognition module, a feature extraction module is designed, the time dimension and the frequency dimension are processed through TF-transform, and the time-frequency distribution characteristics of the voice are modeled by capturing the global dependency relation of the time dimension and the frequency dimension.

Description

Technical Field

[0001] The present invention belongs to the technical field of underwater acoustic signal processing, and in particular relates to an underwater acoustic signal recognition method based on cross-attention feature fusion. Background Art

[0002] Passive sonar systems have the functions of remote monitoring, positioning, tracking and identification of underwater targets such as ships. In particular, passive sonar can achieve target identification by receiving the radiation noise signal emitted by the target ship itself. Hydroacoustic target recognition technology identifies individual targets by analyzing the characteristics of hydroacoustic signals. It is one of the key technologies for realizing the intelligence of other underwater equipment and an important research direction in the field of hydroacoustic engineering. In the actual marine environment, when taking underwater target signals, the hydrophone will inevitably be interfered by underwater environmental noise, such as wind, rain, and sound waves emitted by marine organisms, resulting in a relatively low signal-to-noise ratio when the signal reaches the detection system. This poses a huge challenge to the fine analysis and processing of target signals, resulting in a significant decrease in the performance of hydroacoustic signal recognition.

[0003] In recent years, researchers have optimized the accuracy of underwater acoustic signal recognition from multiple perspectives, including ways to improve the signal-to-noise ratio and optimize the network model. A common strategy is to integrate the audio denoising front end into the target recognition system, aiming to improve the quality of the target audio by preprocessing the audio signal with a low signal-to-noise ratio and improving the signal-to-noise ratio, thereby improving the accuracy of target recognition. The denoising front end algorithms usually include Wiener filtering, spectral subtraction, and deep learning-based denoising algorithms, which process the signal in the time domain and time-frequency domain in different ways to reduce the impact of noise. However, the application of the denoising front end does not always improve the recognition performance. In fact, the underwater acoustic audio signal after denoising may be distorted, which is usually caused by excessive suppression of the noise of the target signal or artifacts introduced during the denoising process. Especially in low signal-to-noise ratio environments, the denoising algorithm often causes the partial loss of the spectral characteristics of the target signal, which in turn affects the accuracy of the subsequent recognition system.

[0004] In order to reduce the distortion caused by the noise reduction model, researchers usually pre-process the noise reduction front-end data first, and then let the back-end recognition model directly process the noise-reduced audio. However, this method is highly dependent on the performance of the noise reduction model. To address the distortion problem caused by the noise reduction front-end, researchers have developed a joint training framework. The noise reduction front-end and the recognition back-end are jointly trained to improve the robustness and accuracy of target recognition by optimizing the synergy of feature extraction and recognition processes.

[0005] This approach ensures that the front-end and back-end models can be optimized together, reducing error accumulation and improving feature discrimination, thereby providing better performance in a noisy environment. However, the above method only considers amplitude spectrum denoising and recognition, without considering phase information, which affects the accuracy and robustness of recognition.

[0006] Therefore, we consider both amplitude and phase information in the noise reduction front end to achieve accurate signal reconstruction and optimized underwater acoustic target signal feature extraction, thereby better improving the signal-to-noise ratio. Self-attention and cross-attention can fully capture global and local information, which is conducive to the fusion of noise amplitude information with the amplitude and phase information output by the noise reduction front end. Since TF-transformer has excellent performance in speech recognition, we consider using it in feature extraction of back-end recognition to better capture audio features. Therefore, inspired by the above methods, we propose a joint training framework based on amplitude and phase feature fusion to jointly train noise reduction and recognition tasks to improve the accuracy and robustness of underwater target recognition. Summary of the invention

[0007] In view of this, the purpose of the present invention is to provide an underwater acoustic signal recognition method based on cross-attention feature fusion, which can optimize the underwater acoustic audio distortion problem caused by the noise reduction front end.

[0008] A method for underwater acoustic signal recognition based on cross-attention feature fusion, comprising: Convert the underwater acoustic signal into spectrum data D; The spectrum data D is input into a noise reduction front-end module; the noise reduction front-end module includes an encoder part, an information exchange part, and a decoder part; each part is divided into an amplitude branch and a phase branch; The spectrum data D is processed in the amplitude branch of the encoder part to obtain the amplitude mask prediction data A1; the spectrum data D is processed in the phase branch of the encoder part to obtain the phase mask prediction data P1; In the information exchange part, the amplitude mask prediction data A1 and the phase mask prediction data P1 are shared, and each branch receives the key information of the other party, thereby obtaining the corresponding feature A4 in the amplitude branch and the feature P4 in the phase branch; In the decoder part, the amplitude branch decodes feature A4 to obtain feature A7; the phase branch decodes feature P4 to obtain feature P5; The feature A7 and the feature P5 are sent to a feature fusion module based on a cross attention mechanism for processing to obtain a fused feature C5; The fused feature C5 is sent to the recognition backend module for processing to identify the category of the target.

[0009] Preferably, when converting the water acoustic signal into the spectrum data D, the water acoustic signal is divided into multiple time periods; the divided water acoustic signal is processed using Fourier transform to obtain the spectrum data D. ; Among them, the variable T represents the time step, F represents the number of frequency bands, that is, the number of frequency bands or frequency points divided by the spectrum graph along the frequency axis, and each frequency band represents the energy or amplitude information of the signal within a specific frequency range; variable 2 represents the real and imaginary parts of the signal.

[0010] Preferably, in the encoder, the amplitude branch consists of two convolutional layers and two activation functions, the spectrum graph data D1 is first processed by the first convolutional layer and the first ReLu activation function, and then processed by the second convolutional layer and the second ReLu activation function to obtain feature A1; the phase branch consists of two convolutional layers, and the spectrum graph data D2 is processed by two convolutional layers in turn to obtain feature P1.

[0011] Preferably, in the amplitude branch of the information exchange module, feature A1 is processed by three convolutional layers in sequence, wherein each convolutional layer is followed by a batch normalization layer BN and a ReLu activation function, and finally feature A2 is obtained; in the phase branch, feature P1 is first processed by linear normalization, and then processed by two convolutional layers in sequence to obtain feature P2; Feature A2 is processed through a convolution layer and a Tanh activation function to obtain feature A3, which is then multiplied with feature P2 to obtain feature P4; feature P2 is processed through a convolution layer and a Tanh activation function to obtain feature P3, which is then multiplied with feature A2 to obtain feature A4.

[0012] Preferably, in the decoder part, feature A4 passes through a convolution layer, a shaping layer and three fully connected layers FC in sequence to obtain feature A5; wherein the activation function connected after the convolution layer is a Sigmoid activation function, the number of units of the first two fully connected layers FC is 600, the activation function connected after each fully connected layer FC is a ReLu activation function, the number of units of the last fully connected layer FC is the number of frequency bands, and the activation function is Sigmoid; feature P4 passes through a convolution layer and amplitude regularization in sequence to obtain feature P5, and phase mask prediction is completed; feature A5 is dot-multiplied with the input spectrum graph data D to obtain feature A6, and feature A6 is dot-multiplied with feature P5 to obtain feature A7 to complete amplitude mask prediction.

[0013] Preferably, the feature fusion module adopts a dual-branch structure to independently process the amplitude and phase features, specifically including: Feature A7 and feature P5 are transformed into feature A8 and feature P6 respectively after Mel spectrum conversion and self-attention mechanism. Feature A8 and feature P6 are transformed into feature A9 and feature P7 respectively after two fully connected layers FC. They are added with feature A8 and feature P6 through skip connection to obtain feature A10 and feature P8 respectively. Feature A10 is used as the Q parameter input of cross attention, and feature P8 is used as the K parameter and V parameter input of cross attention. After cross attention, feature C1 is obtained. Feature C1 is added with feature A10 through skip connection to obtain feature C2. Feature C3 is obtained after a convolution layer and the subsequent activation function ReLu. Feature C3 is successively transformed into feature C4 after a fully connected layer FC and its activation function Sigmoid, convolution layer and the subsequent activation function ReLu. Feature C4 is added with feature C3 through skip connection to obtain feature C5, thus completing feature fusion.

[0014] Preferably, in the recognition backend module, feature C5 is first processed by two convolutional layers to obtain feature C6; feature C6 is further processed by three feature extraction modules in turn to obtain feature C7; Each of the feature extraction modules is composed of a TF-transformer module, three convolutional layers and a TF-transformer module connected in sequence, and is used to extract the features of underwater acoustic signals; wherein the three convolutional layers are used to extract the local correlation of features, and the sizes are 5×5, 1×25, and 5×5 respectively, and each convolutional layer is followed by a normalization layer, a ReLu activation function and a Dropout layer; TF-transformer is responsible for capturing the time and frequency dependencies in the signal; the TF-transformer module consists of two parts, the F-transformer module and the T-transformer module, which act on the signal frequency dimension and time dimension respectively; in the F-transformer module, feature C6 is processed by the multi-head attention mechanism to obtain feature F1, feature F2 is obtained by skip connection, feature F3 is obtained by the fully connected layer FC, feature F4 is obtained by LSTM and the fully connected layer FC, feature F5 is obtained by skip connection, feature F6 is obtained by the fully connected layer FC, and feature F7 is obtained by skip connection, and feature F7 is obtained by the T-transformer module to obtain feature T7; feature C7 is averaged and pooled to aggregate the information of the feature map in the time dimension and frequency dimension to obtain feature C8 and feature C9 respectively, feature C8 and feature C9 are obtained by addition operation to obtain feature C10, which is processed by the first fully connected layer FC and attached to the subsequent ReLu activation function, and finally processed by the second fully connected layer FC to obtain the final target category.

[0015] Preferably, the method further includes training the noise reduction front-end module; wherein the loss function used for training the noise reduction front-end module is the loss of amplitude prediction. and the loss of phase prediction Weighted summation, specifically:

[0016]

[0017] in, Characterizes the amplitude branch output of the decoder of the noise reduction front-end module, Express Modulo operation; represents the raw spectrogram data, Indicates first Modulo operation and then power law compression spectrum operation symbol Represents the dot product operation.

[0018] Preferably, the method further includes training of the recognition backend module, wherein the loss function used for training the recognition backend module is as follows:

[0019] in, is the number of samples, For the Sample No. The true value of each category, To identify the backend module Sample No. The predicted probability of each category, C represents the number of target categories.

[0020] Preferably, the denoising front-end module, the feature fusion module and the recognition back-end module are jointly trained, and the loss function of the joint training is the weighted sum of the loss functions of the denoising front-end module and the recognition back-end module.

[0021] The present invention has the following beneficial effects: The present invention provides an underwater acoustic signal recognition method based on cross-attention feature fusion. In the noise reduction module, the present invention utilizes a dual-branch structure to process amplitude and phase information respectively, and adopts an information interaction mode so that the information of another channel can be used as a reference during amplitude or phase processing, thereby improving the representation ability of the feature; the amplitude and phase information obtained by the noise reduction module are dynamically fused using a feature fusion module, which solves the problem of only using amplitude information and ignoring phase features, and significantly reduces information loss while enhancing the robustness of the system; in the recognition module, the present invention designs a feature extraction module, utilizes TF-transformer to process time and frequency dimensions respectively, and models the time-frequency distribution characteristics of speech by capturing the global dependency between the two. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a joint training framework based on cross-attention fusion; Figure 2 It is a noise reduction front-end module; Figure 3 It is a feature fusion module based on cross attention; Figure 4 To identify the backend module; Figure 5 Extract structure for features; Figure 6 It is a TF-transform structure; Figure 7 is the feature distribution before feature fusion; Figure 8 is the feature distribution after feature fusion; Fig. 9 Comparison of recognition rate, precision, recall rate and F1 score between different models under different data sets. DETAILED DESCRIPTION

[0023] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0024] like Figure 1 As shown in the figure, the proposed joint training model consists of a noise reduction front end, feature fusion, and recognition back end. First, the audio data of the underwater acoustic signal is converted into a spectrum graph data by STFT and input into the noise reduction front end module: , where T represents the time step, which refers to how many time periods or frames the signal is divided into when converting the sound signal into a spectrogram. Each time step usually corresponds to a specific short time window in the signal. Within this window, the signal is processed using Fourier transform (or similar transform methods) to analyze the frequency components in this time segment. The variable F represents the number of frequency bands, that is, the number of frequency bands or frequency points along the frequency axis of the spectrogram, and each frequency band represents the energy or amplitude information of the signal within a specific frequency range. Variable 2 represents the real and imaginary parts of the signal.

[0025] like Figure 2 As shown in the figure, the noise reduction front-end module includes an encoder, an information exchange module, and a decoder. Each part is divided into two branches, the amplitude branch (upper layer) and the phase branch (lower layer), which respectively complete the prediction of the amplitude mask and the phase mask; the spectrum data is also divided into two branches, defined as spectrum data D1 and spectrum data D2. In the encoder, the amplitude branch consists of two convolutional layers and two activation functions. The spectrum data D1 first passes through the first convolutional layer Conv(7×1) and the ReLu activation function, and then passes through the second convolutional layer Conv(1×7) and the ReLu activation function to obtain the feature A1. The two convolutional layers do not change the length of the time domain and the frequency domain, but only change the number of channels. The phase branch consists of two convolutional layers. The spectrum data D2 passes through the convolutional layers Conv(5×3) and Conv(5×3) in turn to obtain the feature P1. Like the amplitude coding layer, the two convolutional layers do not change the length of the time domain and the frequency domain, but only change the number of channels.

[0026] In the information exchange module, the amplitude branch first models the energy distribution of the underwater audio signal to help the model understand which parts are the main energy concentration areas of the signal, which is conducive to recovering the high-frequency components in the low-frequency power. A1 passes through three convolutional layers in sequence. Each convolutional layer has BN (Batch Normalization) and ReLu activation functions to obtain A2. Using three convolutional layers can process the correlation of frequency domain features. The phase branch part P1 passes through two convolutional layers in sequence to obtain P2. Before the phase information is transmitted into the convolutional layer, Layer Normalization (LN) must be performed first. Unlike BN, LN normalizes all features of a single data sample. LN can reduce the internal covariate shift when training deep neural networks, thereby accelerating the training process and making model training more stable. In order to promote information sharing between the amplitude branch and the phase branch, we added a Conv and activation function Tanh. The specific form is that A2 passes through Conv and activation function Tanh to obtain A3, and then performs a dot multiplication with P2 to obtain P4. P2 passes through Conv and activation function Tanh to obtain P3, and then performs a dot multiplication with A2 to obtain A4. Through this design, each branch can receive key information from the other branch, thereby obtaining a more comprehensive and detailed representation of the input features. This complementary information exchange mechanism not only enhances the network's feature learning ability, but also improves the overall signal processing efficiency and accuracy.

[0027] In the decoding layer, A4 is sequentially passed through the convolution layer, the reshape layer, and three fully connected layers (FC) to obtain A5. The activation function of the convolution layer is Sigmoid, the number of units of the first two FCs is 600, the activation function is ReLu, and the number of units of the last FC is F, and the activation function is Sigmoid. P4 is sequentially passed through the convolution layer and amplitude regularization to obtain P5, completing the phase mask prediction. A5 is dot-multiplied with D to obtain A6, and A6 is dot-multiplied with P5 to obtain A7 to complete the amplitude mask prediction.

[0028] like Figure 4As shown in the figure, the present invention designs a feature fusion module. The module adopts a dual-branch structure to independently process the amplitude and phase features. Subsequently, feature fusion is achieved through a cross-attention mechanism. The cross-attention mechanism can extract complementary information from amplitude features and phase features, significantly enhancing key features. At the same time, the fused Mel spectrogram can retain the inherent details of the signal and show high feature cross-correlation and excellent nonlinear feature extraction capabilities. Therefore, feature A7 and feature P5 are successively transformed through Mel spectrum conversion and self-attention mechanism to obtain feature A8 and feature P6 respectively. Feature A8 and feature P6 are respectively obtained through two FCs to obtain feature A9 and feature P7, which are added with feature A8 and feature P6 through skip connection to obtain feature A10 and feature P8 respectively. We use feature A10 as the Q parameter input of cross attention, and feature P8 as the K and V parameter input of cross attention. After cross attention, we get C1, which is added with feature A10 through skip connection to obtain feature C2, and passes through a convolutional layer and its activation function Relu to obtain feature C3. Feature C3 passes through FC and its activation function sigmoid, convolutional layer and its activation function ReLu in turn to obtain feature C4. Feature C4 is added with feature C3 through skip connection to obtain feature C5, completing feature fusion.

[0029] like Figure 5 As shown in the figure, in the recognition backend, C5 first performs preliminary processing on the input features through two convolutional layers to obtain C6. The kernel size of each convolutional layer is 3*3, and ReLu activation function and Dropout layer are attached to enhance the feature expression ability and prevent overfitting. After completing the preliminary processing, C6 is further processed through three feature extraction modules to obtain C7. Each feature extraction module consists of TF-transformer, three convolutional layers and TF-transformer, which are used to extract the features of underwater acoustic signals. Among them, the three convolutional layers can extract the local correlation of features, and the sizes are 5*5, 1*25, and 5*5 respectively, and each convolutional layer is attached with a normalization layer, ReLu activation function and Dropout layer; TF-transformer is responsible for capturing the complex time and frequency dependencies in the signal, while the convolutional layer is used to extract the local correlation of features. The structure of TF-transformer is shown as follows Figure 6As shown, it includes two parts, F-transformer and T-transformer, which act on the frequency dimension and time dimension respectively. In F-transformer, C6 obtains F1 through the multi-head attention mechanism, obtains F2 through skip connection, obtains F3 through FC, obtains F4 through LSTM and FC, obtains F5 through skip connection, obtains F6 through FC, and obtains F7 through skip connection. Similarly, F7 obtains T7 through T-transformer. C7 output by the feature extraction module is average pooled to aggregate the information of the feature map in the time dimension and frequency dimension to obtain C8 and C9 respectively. Through the addition operation C10, it passes through the first FC (size is 1024) and adds the ReLu activation function. The size of the second FC is 8, which is equal to the number of target categories, to achieve the final classification task.

[0030] The loss function of the denoising front-end model is MSE, and the loss value is mainly composed of two parts: and , refers to the magnitude of the predicted loss, is the loss of phase prediction.

[0031]

[0032]

[0033]

[0034] in, Characterizes the amplitude branch output of the decoder of the noise reduction front-end module, It means that the feature A7 is first modulo-operated and then the power law compression spectrum is performed; represents the raw spectrogram data, Indicates first Modulo operation is then performed to compress the spectrum using power law. By applying power law compression to change the dynamic range of the spectrum graph, the weak part of the signal is more obvious than the strong part, which helps to analyze and highlight the subtle changes of the signal.

[0035] In the recognition backend, we chose cross entropy loss as the loss function. Cross entropy loss is widely used in classification tasks due to its intuitiveness and effectiveness. This loss function guides the model to optimize parameters by quantifying the difference between the model's predicted distribution and the true distribution. Specifically, cross entropy loss encourages the model to assign a higher probability to the correct category while effectively suppressing the probability of the wrong category, thereby improving classification performance.

[0036]

[0037] in, represents the number of samples, Indicates Sample No. The true value of each category, Indicates Sample No. The predicted probability of each category, C represents the number of category labels.

[0038] The denoising front-end, feature fusion and recognition back-end are integrated into one, and the denoising and recognition targets are optimized simultaneously using the joint training method. The joint training optimization method can not only improve the model performance, but also enhance its robustness and generalization ability. For the loss function of joint training :

[0039] in, are the relative weight parameters for these loss functions.

[0040] The recognition method of the present invention can significantly improve the feature expression ability of the model by introducing the feature fusion module, effectively alleviating the feature mismatch problem between the noise reduction front end and the recognition back end. Feature fusion enhances the robustness of the model to noise by integrating amplitude and phase information, while improving the overall recognition performance. Figure 7 and Figure 8 .

[0041] Although the method of fusing the amplitude features of noisy audio with the amplitude features of noise-reduced audio improves the noise robustness of underwater target recognition, it loses phase information, limiting recognition performance. Therefore, we fully exploit the potential of phase information by fusing the amplitude features of noise-reduced audio with the estimated phase features. The model is tested on noisy and clean datasets. The average recognition rate, accuracy, recall rate, and F1 score of this model are improved by 6.99%, 6.61%, 7.04%, and 6.91%, respectively. The specific results are as follows: Fig. 9 shown.

[0042] The present invention adds a feature extraction module containing a TF-transformer in recognition, wherein the TF-transformer is responsible for capturing the complex time and frequency dependencies in the signal, and the convolutional layer is used to extract the local correlation of the features.

[0043] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for underwater acoustic signal recognition based on cross-attention feature fusion, characterized in that: include: Convert the underwater acoustic signal into spectrum data D; Inputting the spectrum data D into the noise reduction front-end module; The noise reduction front-end module includes an encoder part, an information exchange part, and a decoder part; each part is divided into an amplitude branch and a phase branch; The spectrum data D is processed in the amplitude branch of the encoder part to obtain the amplitude mask prediction data A1; the spectrum data D is processed in the phase branch of the encoder part to obtain the phase mask prediction data P1; In the information exchange part, the amplitude mask prediction data A1 and the phase mask prediction data P1 are shared, and each branch receives the key information of the other party, thereby obtaining the corresponding feature A4 in the amplitude branch and the feature P4 in the phase branch; In the decoder part, the amplitude branch decodes feature A4 to obtain feature A7; the phase branch decodes feature P4 to obtain feature P5; The feature A7 and the feature P5 are sent to a feature fusion module based on a cross attention mechanism for processing to obtain a fused feature C5; The fused feature C5 is sent to the recognition backend module for processing to identify the category of the target.

2. The method for underwater acoustic signal recognition based on cross-attention feature fusion as claimed in claim 1, characterized in that: When converting the underwater acoustic signal into the spectrum data D, the underwater acoustic signal is divided into multiple time periods; the segmented underwater acoustic signal is processed using Fourier transform to obtain the spectrum data D. ; Wherein, variable T represents the time step, F represents the number of frequency bands, that is, the number of frequency bands or frequency points into which the spectrum graph is divided along the frequency axis, and each frequency band represents the energy or amplitude information of the signal within the set frequency range; variable 2 represents the real and imaginary parts of the signal.

3. A method for underwater acoustic signal recognition based on cross-attention feature fusion as described in claim 1 or 2, characterized in that: In the encoder, the amplitude branch consists of two convolutional layers and two activation functions. The spectrum graph data D1 is first processed by the first convolutional layer and the first ReLu activation function, and then processed by the second convolutional layer and the second ReLu activation function to obtain feature A1; the phase branch consists of two convolutional layers, and the spectrum graph data D2 is processed by two convolutional layers in turn to obtain feature P1.

4. The method for underwater acoustic signal recognition based on cross-attention feature fusion as claimed in claim 3, characterized in that: In the amplitude branch of the information exchange part, feature A1 is processed by three convolutional layers in sequence, each of which is followed by a batch normalization layer BN and a ReLu activation function, and finally feature A2 is obtained; in the phase branch, feature P1 is first processed by linear normalization, and then processed by two convolutional layers in sequence to obtain feature P2; Feature A2 is processed through a convolution layer and Tanh activation function to obtain feature A3, which is then multiplied with feature P2 to obtain feature P4. After feature P2 is processed by a convolution layer and Tanh activation function to obtain feature P3, it is then dot-multiplied with feature A2 to obtain feature A4.

5. The method for underwater acoustic signal recognition based on cross-attention feature fusion as claimed in claim 4, characterized in that: In the decoder part, feature A4 passes through the convolution layer, the shaping layer and three fully connected layers FC in sequence to obtain feature A5; the activation function connected after the convolution layer is the Sigmoid activation function, the number of units of the first two fully connected layers FC is 600, the activation function connected after each fully connected layer FC is the ReLu activation function, the number of units of the last fully connected layer FC is the number of frequency bands, and the activation function is Sigmoid; feature P4 passes through the convolution layer and amplitude regularization in sequence to obtain feature P5, and completes the phase mask prediction; feature A5 is dot-multiplied with the input spectrum graph data D to obtain feature A6, and feature A6 is dot-multiplied with feature P5 to obtain feature A7 to complete the amplitude mask prediction.

6. The method for underwater acoustic signal recognition based on cross-attention feature fusion as claimed in claim 5, characterized in that: The feature fusion module adopts a dual-branch structure to independently process the amplitude and phase features, specifically including: Feature A7 and feature P5 are transformed into feature A8 and feature P6 respectively after Mel spectrum conversion and self-attention mechanism. Feature A8 and feature P6 are transformed into feature A9 and feature P7 respectively after two fully connected layers FC. They are added with feature A8 and feature P6 through skip connection to obtain feature A10 and feature P8 respectively. Feature A10 is used as the Q parameter input of cross attention, and feature P8 is used as the K parameter and V parameter input of cross attention. After cross attention, feature C1 is obtained. Feature C1 is added with feature A10 through skip connection to obtain feature C2. Feature C3 is obtained after a convolution layer and the subsequent activation function ReLu. Feature C3 is successively transformed into feature C4 after a fully connected layer FC and its activation function Sigmoid, convolution layer and the subsequent activation function ReLu. Feature C4 is added with feature C3 through skip connection to obtain feature C5, thus completing feature fusion.

7. The method for underwater acoustic signal recognition based on cross-attention feature fusion as claimed in claim 6, characterized in that: In the recognition backend module, feature C5 is first processed by two convolutional layers to obtain feature C6; feature C6 is further processed by three feature extraction modules in turn to obtain feature C7; Each of the feature extraction modules is composed of a TF-transformer module, three convolutional layers and a TF-transformer module connected in sequence, and is used to extract the features of underwater acoustic signals; wherein the three convolutional layers are used to extract the local correlation of features, and the sizes are 5×5, 1×25, and 5×5 respectively, and each convolutional layer is followed by a normalization layer, a ReLu activation function and a Dropout layer; The TF-transformer module is responsible for capturing the time and frequency dependencies in the signal; the TF-transformer module consists of two parts, the F-transformer module and the T-transformer module, which act on the signal frequency dimension and time dimension respectively; in the F-transformer module, feature C6 is processed by the multi-head attention mechanism to obtain feature F1, feature F2 is obtained by skip connection, feature F3 is obtained by the fully connected layer FC, feature F4 is obtained by LSTM and the fully connected layer FC, feature F5 is obtained by skip connection, feature F6 is obtained by the fully connected layer FC, and feature F7 is obtained by skip connection, and feature F7 is obtained by the T-transformer module to obtain feature T7; feature C7 is averaged and pooled to aggregate the information of the feature map in the time dimension and frequency dimension to obtain feature C8 and feature C9 respectively, feature C8 and feature C9 are obtained by addition operation to obtain feature C10, which is processed by the first fully connected layer FC and attached to the subsequent ReLu activation function, and finally processed by the second fully connected layer FC to obtain the final target category.

8. The method for underwater acoustic signal recognition based on cross-attention feature fusion as claimed in claim 7, characterized in that: It also includes training of the noise reduction front-end module; wherein the loss function used for training the noise reduction front-end module is the loss of amplitude prediction and the loss of phase prediction Weighted summation, specifically: in, Characterizes the amplitude branch output of the decoder of the noise reduction front-end module, Express Modulo operation; represents the raw spectrogram data, Indicates first Modulo operation and then power law compression spectrum operation symbol Represents the dot product operation.

9. The method for underwater acoustic signal recognition based on cross-attention feature fusion as claimed in claim 8, characterized in that: It also includes the training of the recognition backend module, where the loss function used to train the recognition backend module is as follows: in, is the number of samples, For the The sample The true value of each category, To identify the backend module The sample The predicted probability of each category, C represents the number of target categories.

10. The underwater acoustic signal recognition method based on cross-attention feature fusion according to claim 9, characterized in that: The denoising front-end module, the feature fusion module and the recognition back-end module are jointly trained, and the loss function of the joint training is the weighted sum of the loss functions of the denoising front-end module and the recognition back-end module.

Images