Ship radiation signal and natural environment noise separation method based on time domain separation network

Through the method based on the time domain separation network, the water acoustic signals are processed using modules such as convolutional network and timing convolutional network, which solves the problem of low signal-to-noise ratio after signal separation in the prior art, and achieves more accurate ship radiation Separation of signal from natural ambient noise and retention of signal characteristics.

CN119001691APending Publication Date: 2024-11-22NAT INNOVATION INST OF DEFENSE TECH PLA ACAD OF MILITARY SCI
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202410893523.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2024-06-27
Filing Date
2024-07-04
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

In the process of separation of water acoustic signal data, preprocessing steps such as low-pass filtering and molecular space will cause the signal to lose detailed characteristics, resulting in a low signal-to-noise ratio after separation, and it is impossible to effectively separate the ship's radiation signal from the natural environment noise.

Method used

Using a time domain separation network-based method, the target water acoustic signal is input to the trained time domain separation network, and the target convolution network is used to perform higher-order nonlinear feature representation and mask generation through the target convolution network, timing convolution network and transposed convolution network. , realize the separation of ship radiation signals and natural environment noise.

Benefits of technology

The accuracy of separation between ship radiation signals and natural environment noise in water acoustic signals and the signal-to-noise ratio after separation are improved, and the characteristic information of the target signal is retained to the maximum extent.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119001691A_ABST
    Figure CN119001691A_ABST
Patent Text Reader

Abstract

The invention provides a ship radiation signal and natural environment noise separation method based on a time domain separation network, and the method comprises the steps: obtaining a target underwater acoustic signal which comprises a ship radiation signal and natural environment noise; inputting the target underwater acoustic signal into a trained time domain separation network to obtain a separated underwater acoustic signal, and the trained time domain separation network performs ship radiation signal and natural environment noise separation on the high-order nonlinear feature representation corresponding to the target underwater acoustic signal through the obtained target mask corresponding to the target underwater acoustic signal to obtain the separated underwater acoustic signal. According to the method, the accuracy of separating the ship radiation signal and the natural environment noise from the underwater acoustic signal can be improved, and the signal-to-noise ratio of the separated underwater acoustic signal can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of underwater acoustic signal processing, and in particular to a method for separating ship radiation signals from natural environment noise based on a time domain separation network. Background Art

[0002] When conducting ocean environment detection experiments, the specific information of ship targets can be extracted from different angles by analyzing the hydroacoustic signal data. Since the natural ocean environment contains a lot of interference noise, the hydroacoustic signal data also contains a lot of noise information, so it is necessary to separate the ship radiation signal from the natural environment noise in the hydroacoustic signal data.

[0003] The existing noise separation method for underwater acoustic signal data first pre-processes the noisy original underwater acoustic signal by low-pass filtering and subspace division. These methods will cause the original underwater acoustic signal to lose detailed features, resulting in a low signal-to-noise ratio of the separated underwater acoustic signal. Summary of the invention

[0004] The present invention provides a method for separating ship radiation signals and natural environment noise based on a time domain separation network, which is used to solve the defect of missing features of separated underwater acoustic signals in the prior art. By inputting a target underwater acoustic signal into a trained time domain separation network, the trained time domain separation network separates the ship radiation signal from the natural environment noise by obtaining a target mask corresponding to the target underwater acoustic signal to represent the high-order nonlinear features corresponding to the target underwater acoustic signal, and obtains the separated underwater acoustic signal. This can improve the accuracy of separating the ship radiation signal from the natural environment noise from the underwater acoustic signal, as well as the signal-to-noise ratio of the separated underwater acoustic signal.

[0005] The present invention provides a method for separating ship radiation signals from natural environment noise based on a time domain separation network, comprising the following steps: Acquiring a target hydroacoustic signal, wherein the target hydroacoustic signal includes a ship radiation signal and natural environment noise; The target hydroacoustic signal is input into a trained time domain separation network to obtain a separated hydroacoustic signal. The trained time domain separation network separates the ship radiation signal from the natural environment noise by using a target mask corresponding to the target hydroacoustic signal to obtain a high-order nonlinear feature representation corresponding to the target hydroacoustic signal to obtain the separated hydroacoustic signal.

[0006] According to a method for separating ship radiation signals from natural environment noise based on a time domain separation network provided by the present invention, the trained time domain separation network includes a target convolution network, a target timing convolution network and a target transposed convolution network connected in sequence, and the target underwater acoustic signal is input into the trained time domain separation network to obtain a separated underwater acoustic signal, including: The target convolution network is used to convert the time domain waveform features corresponding to the target underwater acoustic signal into high-order nonlinear features, so as to obtain a high-order nonlinear feature representation corresponding to the target underwater acoustic signal; The target masks of different scales generated by the target temporal convolutional network are calculated with the high-order nonlinear feature representation corresponding to the target underwater acoustic signal to obtain a separated feature representation; The separated feature representation is inversely transformed by the target transposed convolutional network to obtain a separated underwater acoustic signal, wherein the separated underwater acoustic signal is represented by a time domain waveform feature.

[0007] According to a method for separating ship radiation signals from natural environment noise based on a time domain separation network provided by the present invention, the target convolution network includes a target convolution kernel, and the size of the target convolution kernel is determined according to the time-frequency characteristics of the target underwater acoustic signal.

[0008] According to a method for separating ship radiation signals from natural environment noise based on a time domain separation network provided by the present invention, the target temporal convolutional network includes a residual path, a skip path, and a plurality of stacked dilated convolutional blocks, the residual path of the previous dilated convolutional block in the plurality of dilated convolutional blocks serves as the input of the next dilated convolutional block, and the skip path vectors of all dilated convolutional blocks serve as the output of the target temporal convolutional network.

[0009] According to a method for separating ship radiation signals from natural environment noise based on a time domain separation network provided by the present invention, the number of moving steps of each layer of expansion factor corresponding to each of the multiple expansion convolution blocks increases exponentially, the input values ​​of each expansion convolution block are filled with zero values, and each expansion convolution block includes an expansion convolution kernel, in which ordinary convolution operations are divided into depth convolution calculation and point convolution calculation.

[0010] According to a method for separating ship radiation signals from natural environment noise based on a time domain separation network provided by the present invention, the separated feature representation includes a plurality of separated feature representation fragments, the separated feature representation is inversely transformed and calculated by the target transposed convolution network to obtain a separated underwater acoustic signal, and the separated underwater acoustic signal is represented by a time domain waveform feature, including: Performing the inverse transformation calculation on the separated multiple feature representation segments through the target transposed convolutional network to obtain the separated multiple signal sub-segments, wherein the network parameters of the transposed convolutional network correspond to the network parameters of the target convolutional network; The separated multiple signal sub-fragments are combined to obtain the separated underwater acoustic signal.

[0011] The present invention also provides a device for separating ship radiation signals and natural environment noise based on a time domain separation network, comprising the following modules: A signal acquisition module, used to acquire a target hydroacoustic signal, wherein the target hydroacoustic signal includes a ship radiation signal and natural environment noise; The signal separation module is used to input the target hydroacoustic signal into a trained time domain separation network to obtain a separated hydroacoustic signal. The trained time domain separation network separates the ship radiation signal from the natural environment noise by using a target mask corresponding to the target hydroacoustic signal to obtain a high-order nonlinear feature representation corresponding to the target hydroacoustic signal to obtain the separated hydroacoustic signal.

[0012] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements any of the above-described methods for separating ship radiation signals from natural environment noise based on a time domain separation network.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for separating ship radiation signals from natural environment noise based on a time domain separation network as described in any of the above is implemented.

[0014] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-mentioned methods for separating ship radiation signals from natural environment noise based on a time domain separation network.

[0015] The method for separating ship radiation signals and natural environment noise based on a time domain separation network provided by the present invention inputs a target hydroacoustic signal into a trained time domain separation network, and the trained time domain separation network separates the ship radiation signal from the natural environment noise by obtaining a target mask corresponding to the target hydroacoustic signal to represent the high-order nonlinear characteristics corresponding to the target hydroacoustic signal, thereby obtaining the separated hydroacoustic signal. This method can improve the accuracy of separating the ship radiation signal from the natural environment noise from the hydroacoustic signal, as well as the signal-to-noise ratio of the separated hydroacoustic signal. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1It is a flow chart of a method for separating ship radiation signals from natural environment noise based on a time domain separation network provided by the present invention.

[0018] Figure 2 It is a flow chart of the method for separating noise from underwater acoustic signals provided by the present invention.

[0019] Figure 3 This is the process intention of the target temporal convolutional network noise separation method provided by the present invention.

[0020] Figure 4 It is a schematic flow chart of the method for obtaining separated hydroacoustic signals provided by the present invention.

[0021] Figure 5 The present invention provides a schematic diagram of the overall process of the method for separating ship radiation signals from natural environment noise based on a time domain separation network.

[0022] Figure 6 It is a structural schematic diagram of a device for separating ship radiation signals and natural environment noise based on a time domain separation network provided by the present invention.

[0023] Figure 7 It is a schematic diagram of the physical structure of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] With the continuous exploration of the natural ocean environment by scientists and technicians, hydroacoustic science and marine science and technology established for the complex and changeable ocean have become important strategic goals for the development of the ocean. When conducting marine environment detection experiments, hydrophones can be used to analyze and process ships, submarines and other underwater vehicles to obtain hydroacoustic signal data. By analyzing these hydroacoustic signal data, specific information of ship targets can be extracted from different angles. However, since the natural ocean environment contains a lot of interference noise, this causes a lot of noise information to be contained in the hydroacoustic target signal data, which may affect the identification, positioning and monitoring of underwater targets. Since the marine environmental noise comes from different sound emission sources, such as ocean currents, airflow radiation sources, marine biological sources, human activities and drilling equipment, it is a challenge to analyze the target signal and the environmental noise signal by shielding a single noise source modeling. In addition, the propagation loss and scattering of sound waves always cause confusion or distortion, which is why the acquired hydroacoustic target signal cannot fully reflect the characteristics of the target.

[0026] At present, deep learning algorithms can analyze target signals and noise signals by learning masks of targets and noise to improve the signal-to-noise ratio of underwater acoustic signals. However, there are still some defects in the feature extraction and analysis of targets. First, in terms of data preprocessing, researchers usually perform manual processing on the original noisy signals. For example, the wavelet decomposition method includes steps such as low-pass filtering and subspace division. These methods will cause the original underwater acoustic signal to lose some detailed features, resulting in a low signal-to-noise ratio of the separated underwater acoustic signal. Therefore, the main research content of the present invention is to use deep learning methods to suppress the interference of natural environmental noise on the target signal, so as to maximize the identifiable features of the target signal.

[0027] In view of this, an embodiment of the present invention provides a method for separating ship radiation signals from natural environmental noise based on a time domain separation network, wherein the target hydroacoustic signal is input into a trained time domain separation network, and the trained time domain separation network separates the ship radiation signal from the natural environmental noise by obtaining a target mask corresponding to the target hydroacoustic signal and representing the high-order nonlinear characteristics corresponding to the target hydroacoustic signal, thereby obtaining the separated hydroacoustic signal. This method can improve the accuracy of separating ship radiation signals from natural environmental noise from hydroacoustic signals, as well as the signal-to-noise ratio of the separated hydroacoustic signal.

[0028] The technical solutions in the embodiments of the present invention will be described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0029] Figure 1The present invention provides a flow chart of a method for separating ship radiation signals from natural environment noise based on a time domain separation network. The method for separating ship radiation signals from natural environment noise based on a time domain separation network can be applied to electronic devices, which can be various types of devices with information processing capabilities during implementation. For example, the electronic device can include a personal computer, a laptop computer, a PDA or a server, etc.; the electronic device can also be a mobile terminal, for example, the mobile terminal can include a mobile phone, a car computer, a tablet computer or a projector, etc. Figure 1 As shown, the method may include the following steps 101 to 102: Step 101: Acquire a target hydroacoustic signal, where the target hydroacoustic signal includes a ship radiation signal and natural environment noise.

[0030] It should be noted that there are many ways to obtain the target water acoustic signal, for example, the target water acoustic signal can be collected, or the target water acoustic signal sent by other devices can be received. The present invention does not limit the way to obtain the target water acoustic signal.

[0031] For example, the problem of separating the underwater acoustic target signal from the natural ocean environment noise can be expressed as: given a mixed signal of target and noise , where the underwater acoustic target signal is defined as , the natural ocean ambient noise signal is defined as Regarding the coupling relationship between the target signal and the natural ocean environment noise signal, the relationship between the two can be defined as shown in formula (1).

[0032] (1).

[0033] At this point, the problem of separating the underwater acoustic target signal from the natural ocean environment noise can be summarized as maximizing the enhancement of the underwater acoustic target signal. , and reduce the natural ocean environment noise signal as much as possible The natural ocean environment noise signal may include ship radiation signals and natural environment noise.

[0034] In some embodiments, obtaining the target water acoustic signal may include obtaining a plurality of water acoustic signal sub-segments. The obtaining of the plurality of water acoustic signal sub-segments may include: dividing the original water acoustic signal, i.e., the input target and noise mixed signal, into sub-segments of length The overlapping signal sub-fragments are represented by the symbol Indicates. Among them, Represents the segment index, Indicates the number of overlapping sub-segments in the input signal.

[0035] Step 102: Input the target underwater acoustic signal into a trained time domain separation network to obtain a separated underwater acoustic signal. The trained time domain separation network separates the ship radiation signal from the natural environment noise by using a target mask corresponding to the target underwater acoustic signal to obtain a separated underwater acoustic signal.

[0036] It should be noted that the trained time domain separation network can obtain the target mask corresponding to the target underwater acoustic signal and the high-order nonlinear feature representation corresponding to the target underwater acoustic signal. The high-order nonlinear feature representation corresponding to the target underwater acoustic signal is separated from the ship radiation signal and the natural environment noise through the target mask to obtain the separated underwater acoustic signal.

[0037] Among them, deep learning noise separation methods are not aimed at modeling a single noise source. They need to integrate the high-order representation and receptive field characteristics of underwater acoustic signals, and fully learn the characteristics of natural environmental noise in a specific sea area for modeling to achieve signal-noise separation.

[0038] It can be understood that the method provided by the present invention for separating the ship radiation signal and the natural environment noise of the target hydroacoustic signal through a trained time domain separation network can improve the accuracy of separating the ship radiation signal and the natural environment noise from the hydroacoustic signal, as well as the signal-to-noise ratio of the hydroacoustic signal after separation.

[0039] In some embodiments, in the existing feature extraction process, deep learning-based technologies mostly rely on acoustic feature extraction methods, such as Mel-scale Frequency Cepstral Coefficients (MFCC) feature extraction and analysis methods and filter group feature Fbank modeling methods. These methods have steps such as filtering and framing, which will change the original temporal structure of the underwater acoustic signal.

[0040] Figure 2 FIG. 1 is a flow chart of a method for separating noise from underwater acoustic signals provided by the present invention. Figure 2 As shown, the trained time domain separation network includes a target convolution network, a target timing convolution network and a target transposed convolution network connected in sequence, and the target underwater acoustic signal is input into the trained time domain separation network to obtain the separated underwater acoustic signal, which may include: Step 201: converting the time domain waveform features corresponding to the target underwater acoustic signal into high-order nonlinear features through the target convolutional network to obtain the high-order nonlinear feature representation corresponding to the target underwater acoustic signal.

[0041] It should be noted that the encoder module of the network, namely the target convolutional network, can be constructed using a one-dimensional convolutional network. By pre-defining the convolution kernel size, a convolution operation is performed on the underwater acoustic signal sequence to convert the waveform characteristics of the time domain signal into a high-order nonlinear representation.

[0042] Step 202: Calculate the target masks of different scales generated by the target temporal convolutional network and the high-order nonlinear feature representations corresponding to the target underwater acoustic signal to obtain separated feature representations.

[0043] It should be noted that the signal separation module, namely the target temporal convolutional network, can be constructed using the temporal convolutional network and the dilated convolutional kernel. The receptive field of the convolutional kernel can be expanded to extract the mask information of deeper targets and noise in the underwater acoustic signal without increasing the computational workload of the hardware equipment and the memory occupancy of the model. The back propagation learning method can be used to estimate the separation structure of the target and noise.

[0044] Step 203: performing inverse transformation calculation on the separated feature representation through the target transposed convolutional network to obtain a separated underwater acoustic signal, wherein the separated underwater acoustic signal is represented by a time domain waveform feature.

[0045] It should be noted that the transposed convolutional network can be used to construct a decoder module, namely a target transposed convolutional network, and the decoder module can be used to perform inverse transformation calculations on the filtered features to reconstruct the waveform feature representation method of the underwater acoustic signal.

[0046] It is understandable that the present invention uses a mask network and a full convolution module to build an end-to-end framework. Using an overall framework to build an end-to-end model cascades each module into a network, reduces engineering complexity and GPU computing space, and can optimize the model based on an objective function, thereby avoiding mutual interference between multiple modules. Among them, the input format is an audio file of the underwater acoustic signal, and the output is the separated underwater acoustic signal audio file.

[0047] In some embodiments, the target convolution network in step 201 includes a target convolution kernel, and the size of the target convolution kernel is determined according to the time-frequency characteristics of the target underwater acoustic signal. In the face of natural ocean environments of different time and space, hyperparameters such as the window function size of the framing process, the number of encoders, and the number of short-time Fourier transforms can be adjusted to adaptively extract signal feature vectors of different sizes and lengths.

[0048] Exemplarily, the encoder process is to convert the signal Input into a one-dimensional convolutional neural network (1D-Conv) to obtain the feature representation of the signal segment .Signal The calculation after the encoder is shown in (2), where the Rectified Linear Unit (ReLu) ensures that the representation is non-negative. Indicates that first Perform a one-dimensional convolution operation , and then pass the result to the ReLU activation function .

[0049] (2).

[0050] It is understandable that the present invention constructs an encoding method for underwater acoustic signals based on a one-dimensional convolutional network model to replace the traditional feature extraction method. Among them, the encoder can adaptively learn the primary expression vector of the signal according to the time-frequency characteristics of the underwater acoustic signals in different sea areas, thereby improving the efficiency of feature extraction.

[0051] In some embodiments, the target temporal convolutional network includes a residual path, a skip path, and a plurality of stacked dilated convolutional blocks, wherein the residual path of a previous dilated convolutional block in the plurality of dilated convolutional blocks serves as the input of the next dilated convolutional block, and the skip path vectors of all dilated convolutional blocks serve as the output of the target temporal convolutional network.

[0052] For example, the signal is separated by applying a set of weighting functions (masks) to the output of the encoder, and masks of different scales are generated by a target temporal convolutional time domain separation network. The target convolutional time domain separation network is composed of stacked one-dimensional dilated convolution extension blocks, which can be used to analyze the difference characteristics between the target signal and the natural environment noise signal.

[0053] Figure 3 This is the process intention of the target temporal convolutional network noise separation method provided by the present invention, such as Figure 3 As shown in Figure 2, the specific structure of the temporal convolutional network is as follows: Figure 3As shown in Figure 1, the entire architecture contains a residual path and a skip path. For example, a 1 × 1 filter maps a complete feature fragment of an underwater acoustic signal to different subspaces, so that the fragment will lose some of the original time-frequency information in the local space, so a skip path is introduced to fuse the signal information in different spatial domains. By taking the residual path of a block as the input of the next block, the entire temporal convolutional network can be continuously expanded, and the skip path vectors of all blocks are added as the output of the temporal convolutional network. Since the essence of underwater acoustic signals is a time-dependent information sequence, in order to capture linear dependencies in a longer time period, it is necessary to stack multiple network modules at different levels, which will increase the number of parameters. In order to increase the receptive field of the network model without increasing the amount of computation of the network layer, downsampling operations are usually used. Although downsampling can reduce the overfitting phenomenon of the model during training, it reduces the spatial resolution of the signal feature map. Therefore, in order to avoid the loss of resolution while still expanding the receptive field, we use dilated convolution kernels to construct temporal convolution networks. In addition, in the dilated convolution kernel, the ordinary convolution operation is divided into more precise sub-processes, and the convolution operator is decomposed into two continuous functions, one for depth convolution calculation and the other for point convolution operation.

[0054] In some embodiments, the prior art needs to build deeper network layers in the noise separation process to obtain sufficient feature information to pursue the performance of the model, but the deep stacked network results will introduce many parameters, which makes the underwater acoustic signal detection system require better hardware support and computing space.

[0055] In an embodiment of the present invention, the number of moving steps of each layer of expansion factor corresponding to each of the multiple expansion convolution blocks in step 202 increases exponentially, the input values ​​of each expansion convolution block are filled with zero values, and each expansion convolution block includes an expansion convolution kernel, in which the ordinary convolution operation is divided into depth convolution calculation and point convolution calculation.

[0056] It should be noted that the present invention uses a temporal convolutional network (TCN) to construct a module for separating underwater acoustic target signals from environmental noise signals. The module consists of stacked dilated convolution blocks. In the back-propagation calculation process, the gradient of the temporal convolutional network will not be accumulated along the time dimension. The forward transmission process of the input information of all time steps is carried out simultaneously in each stage. The temporal convolutional network structure is used as the basic architecture for constructing the signal separation module to avoid inherent gradient dispersion. The number of moving steps of each layer of expansion factor increases exponentially to ensure that sufficient information can be captured in all forward transmission processes. Therefore, the temporal convolutional network can capture the long-term dependency details of the time series. In the signal separation module, the expansion factor is The dilated convolutional block has The number of repetitions is The input value of each dilated convolution block is padded with zero values ​​to ensure that the output sequence length is the same as the input sequence length.

[0057] Exemplarily, the method for separating each frame of the underwater acoustic signal is to estimate the target mask Multiply by the encoder module output value For calculation, the constraint condition of the target mask is defined as , and by applying and the corresponding mask To calculate the separated target signal , as shown in formula (3).

[0058] (3).

[0059] In some embodiments, when the target water acoustic signal includes multiple water acoustic signal sub-segments, the separated feature representation also includes multiple separated feature representation segments.

[0060] Figure 4 FIG. 1 is a flow chart of a method for obtaining a separated hydroacoustic signal provided by the present invention. Figure 4 As shown, the separated feature representation includes a plurality of separated feature representation fragments, and the separated feature representation is inversely transformed and calculated by the target transposed convolutional network to obtain a separated underwater acoustic signal, and the separated underwater acoustic signal is represented by a time domain waveform feature, which may include: Step 301: performing the inverse transform calculation on the separated multiple feature representation fragments through the target transposed convolutional network to obtain the separated multiple signal sub-fragments, and the network parameters of the transposed convolutional network correspond to the network parameters of the target convolutional network.

[0061] It should be noted that the decoder module uses a transposed convolutional neural network, whose parameters such as moving step size and kernel size are the same as those of the encoder, and the input of the decoder is the target mask Mixed signal segment characteristics The elements between them are multiplied, and then the sub-segment of the target signal is estimated through the transposed convolutional neural network. .

[0062] Step 302: merging the multiple separated signal sub-segments to obtain the separated underwater acoustic signal.

[0063] Exemplarily, as shown in formula (4), the reconstructed signal sub-segments are added to generate a filtered underwater acoustic signal.

[0064] (4).

[0065] in, Represents the convolution network through the transpose After processing, the filtered underwater acoustic signal is obtained.

[0066] It can be understood that the present invention constructs the encoding and decoding methods of the underwater acoustic signal based on the one-dimensional convolutional network model to replace the traditional feature extraction method. Among them, the encoder can adaptively learn the primary expression vector of the signal according to the time-frequency characteristics of the underwater acoustic signal in different sea areas, and the decoder uses the inversion operation to convert the high-order expression vector of the signal into the time domain representation of the underwater acoustic signal, which can improve the accuracy of separating the ship radiation signal and the natural environment noise from the underwater acoustic signal, as well as the signal-to-noise ratio of the separated underwater acoustic signal.

[0067] The following describes an exemplary application of an embodiment of the present invention in a practical application scenario.

[0068] Figure 5 The overall flow chart of the method for separating ship radiation signals from natural environment noise based on the time domain separation network provided by the present invention is as follows. Figure 5The figure shows the basic architecture of the Fully Convolutional TimeDomain Separation Network (FCTDSN). First, the encoder module of the network is constructed using a one-dimensional convolutional network. The convolution operation is performed on the underwater acoustic signal sequence by pre-defining the convolution kernel size to convert the waveform characteristics of the time domain signal into a high-order nonlinear representation. When facing natural ocean environments of different time and space, the hyperparameters, such as the window function size of the framing process, the number of encoders, and the number of short-time Fourier transforms, can be adjusted to adaptively extract signal feature vectors of different sizes and lengths. Second, the signal separation module is constructed using a temporal convolutional network and an expanded convolution kernel. Without increasing the computational workload of the hardware device and the memory usage of the model, the receptive field range of the convolution kernel is expanded to extract the mask information of deeper targets and noise in the underwater acoustic signal, and the back-propagation learning method is used to estimate the separation structure of the target and noise. Third, the decoder is constructed using a transposed convolutional network. The decoder module performs inverse transformation calculations on the separated features to reconstruct the waveform feature representation method of the underwater acoustic signal.

[0069] In order to test the denoising effect of the fully convolutional time-domain separation network, we use the underwater ship sound recording database ShipsEarShipsEar for testing. The recordings of the ShipsEar dataset are made by hydrophones placed on the dock to record the ship speed and cavitation noise associated with docking or unhooking operations. These are real ship sounds recorded in a natural environment. Therefore, background noise from humans and nature, as well as vocalizations of marine mammals, are present. The dataset consists of 90 .wav file recordings, which are divided into five main categories. Each main category has one or more subcategories (for example, category A consists of dredgers, fish boats, trawlers, mussel boats, and tugboats), and the length of each audio clip is 15 seconds to 10 minutes. In order to better verify the denoising performance of the model. All signals are segmented at a fixed time of 5 seconds, resulting in 1956 labeled sound samples. Samples without noise categories are randomly extracted. The noise category is fused with these extracted samples to make the signal-to-noise ratio of the sample 0dB, and is divided into training, validation, and test sets in a ratio of 8:1:1. The results of the verification are as follows: 1. Under the conditions of small data volume and few target priors, when separating ship radiation signals from natural environmental noise for different target signals, its noise separation effect is more excellent. 2. In the process of processing underwater acoustic signals of different fragments, its network model is more stable, which can improve the signal-to-noise ratio of underwater acoustic signals.

[0070] The above embodiment has the following advantages: 1. The encoding and decoding methods of the underwater acoustic signal are constructed based on the one-dimensional convolutional network model to replace the traditional feature extraction method. Among them, the encoder can adaptively learn the primary expression vector of the signal according to the time-frequency characteristics of the underwater acoustic signal in different sea areas, and the decoder uses the inversion operation to convert the high-order expression vector of the signal into the time domain representation of the underwater acoustic signal. 2. The signal is separated by applying a set of weighting functions (masks) to the output of the encoder, and masks of different scales are generated by the time-series convolution time domain separation network. Among them, the convolution time domain separation network is composed of stacked one-dimensional dilated convolution extension blocks, which can be used to analyze the differential characteristics of the target signal and the natural environment noise signal. 3. The end-to-end framework is constructed using the mask network and the full convolution module. Using an overall framework to build an end-to-end model cascades each module into a network, reduces the engineering complexity and GPU computing space, and optimizes the model based on an objective function, thereby avoiding the mutual interference of multiple modules. Among them, the input format is an audio file of the underwater acoustic signal, and the output is the separated underwater acoustic signal audio file.

[0071] Based on the foregoing embodiments, an embodiment of the present invention provides a device for separating ship radiation signals and natural environment noise based on a time domain separation network. The modules included in the device and the units included in each module can be implemented by a processor; of course, they can also be implemented by a specific logic circuit; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP) or a field programmable gate array (FPGA), etc.

[0072] The following is a description of the device for separating ship radiation signals and natural environment noise based on a time domain separation network provided by the present invention. The device for separating ship radiation signals and natural environment noise based on a time domain separation network described below and the method for separating ship radiation signals and natural environment noise based on a time domain separation network described above can be referred to each other.

[0073] Figure 6 Schematic diagram of the structure of the device for separating ship radiation signals and natural environment noise based on the time domain separation network provided by the present invention. Figure 6 As shown, the device 400 includes a signal acquisition module 401 and a signal separation module 402, wherein: The signal acquisition module 401 is used to acquire a target hydroacoustic signal, wherein the target hydroacoustic signal includes a ship radiation signal and natural environment noise; The signal separation module 402 is used to input the target hydroacoustic signal into a trained time domain separation network to obtain a separated hydroacoustic signal. The trained time domain separation network separates the ship radiation signal from the natural environment noise by using a target mask corresponding to the target hydroacoustic signal to obtain a high-order nonlinear feature representation corresponding to the target hydroacoustic signal to obtain the separated hydroacoustic signal.

[0074] In some embodiments, the trained time domain separation network includes a target convolutional network, a target temporal convolutional network and a target transposed convolutional network connected in sequence, and the signal separation module 402 includes a signal conversion unit, a signal separation unit and a signal restoration unit, wherein:

[0075] The signal conversion unit is used to convert the time domain waveform characteristics corresponding to the target underwater acoustic signal into high-order nonlinear characteristics through the target convolution network, so as to obtain the high-order nonlinear characteristic representation corresponding to the target underwater acoustic signal; The signal separation unit is used to calculate the target masks of different scales generated by the target temporal convolutional network and the high-order nonlinear feature representation corresponding to the target underwater acoustic signal to obtain a separated feature representation; The signal restoration unit is used to perform inverse transformation calculation on the separated feature representation through the target transposed convolutional network to obtain a separated underwater acoustic signal, and the separated underwater acoustic signal is represented by a time domain waveform feature.

[0076] In some embodiments, the target convolution network includes a target convolution kernel, and the size of the target convolution kernel is determined according to the time-frequency characteristics of the target underwater acoustic signal.

[0077] In some embodiments, the target temporal convolutional network includes a residual path, a skip path, and a plurality of stacked dilated convolutional blocks, wherein the residual path of a previous dilated convolutional block in the plurality of dilated convolutional blocks serves as the input of the next dilated convolutional block, and the skip path vectors of all dilated convolutional blocks serve as the output of the target temporal convolutional network.

[0078] In some embodiments, the number of moving steps of each layer of expansion factor corresponding to each of the multiple expansion convolution blocks increases exponentially, the input values ​​of each expansion convolution block are filled with zero values, and each expansion convolution block includes an expansion convolution kernel, in which ordinary convolution operations are divided into depth convolution calculation and point convolution calculation.

[0079] In some embodiments, the separated feature representation includes multiple separated feature representation fragments, and the signal restoration unit is specifically used to: perform the inverse transform calculation on the multiple separated feature representation fragments through the target transposed convolutional network to obtain multiple separated signal sub-fragments, and the network parameters of the transposed convolutional network correspond to the network parameters of the target convolutional network; merge the multiple separated signal sub-fragments to obtain the separated underwater acoustic signal.

[0080] In the embodiment of the present invention, the device can improve the accuracy of separating the ship radiation signal and the natural environment noise from the hydroacoustic signal, as well as the signal-to-noise ratio of the hydroacoustic signal after separation.

[0081] Figure 7 is a schematic diagram of the physical structure of the electronic device provided by the present invention, such as Figure 7 As shown, the electronic device may include: a processor 501, a communication interface 502, a memory 503 and a communication bus 504, wherein the processor 501, the communication interface 502 and the memory 503 communicate with each other through the communication bus 504. The processor 501 may call the logic instructions in the memory 503 to execute a method for separating ship radiation signals from natural environment noise based on a time domain separation network, the method comprising: obtaining a target hydroacoustic signal, the target hydroacoustic signal including a ship radiation signal and natural environment noise; inputting the target hydroacoustic signal into a trained time domain separation network to obtain a separated hydroacoustic signal, the trained time domain separation network performing separation of the ship radiation signal from the natural environment noise on the high-order nonlinear feature representation corresponding to the target hydroacoustic signal through the target mask corresponding to the acquired target hydroacoustic signal, to obtain the separated hydroacoustic signal.

[0082] In addition, the logic instructions in the above-mentioned memory 503 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0083] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the ship radiation signal and natural environment noise separation method based on the time domain separation network provided by the above methods. The method includes: obtaining a target water acoustic signal, wherein the target water acoustic signal includes a ship radiation signal and natural environment noise; inputting the target water acoustic signal into a trained time domain separation network to obtain a separated water acoustic signal, and the trained time domain separation network separates the ship radiation signal from the natural environment noise by using a target mask corresponding to the target water acoustic signal to represent the high-order nonlinear characteristics of the target water acoustic signal to obtain the separated water acoustic signal.

[0084] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that includes one or more available media integrated therein. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0085] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the ship radiation signal and natural environment noise separation method based on a time domain separation network provided by the above-mentioned methods, the method comprising: obtaining a target water acoustic signal, the target water acoustic signal comprising a ship radiation signal and natural environment noise; inputting the target water acoustic signal into a trained time domain separation network to obtain a separated water acoustic signal, the trained time domain separation network performing separation of the ship radiation signal and the natural environment noise on the high-order nonlinear feature representation corresponding to the target water acoustic signal through the acquired target mask corresponding to the target water acoustic signal, to obtain the separated water acoustic signal.

[0086] The above-mentioned computer-readable storage medium can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, - but not limited to - an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM) or flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0087] Computer-readable signal media may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0088] The program code embodied on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, radio frequency (RF), etc., or any suitable combination of the foregoing.

[0089] Computer program code for performing the operations of the present specification may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0090] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0091] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for separating ship radiation signals from natural environment noise based on a time domain separation network, characterized in that: include: Acquiring a target hydroacoustic signal, wherein the target hydroacoustic signal includes a ship radiation signal and natural environment noise; The target hydroacoustic signal is input into a trained time domain separation network to obtain a separated hydroacoustic signal. The trained time domain separation network separates the ship radiation signal from the natural environment noise by using a target mask corresponding to the target hydroacoustic signal to obtain a high-order nonlinear feature representation corresponding to the target hydroacoustic signal to obtain the separated hydroacoustic signal.

2. According to the method for separating ship radiation signals and natural environment noise based on time domain separation network in claim 1, it is characterized in that: The trained time domain separation network includes a target convolution network, a target timing convolution network and a target transposed convolution network connected in sequence, and the target underwater acoustic signal is input into the trained time domain separation network to obtain a separated underwater acoustic signal, including: The target convolution network is used to convert the time domain waveform features corresponding to the target underwater acoustic signal into high-order nonlinear features, so as to obtain a high-order nonlinear feature representation corresponding to the target underwater acoustic signal; The target masks of different scales generated by the target temporal convolutional network are calculated with the high-order nonlinear feature representation corresponding to the target underwater acoustic signal to obtain a separated feature representation; The separated feature representation is inversely transformed by the target transposed convolutional network to obtain a separated underwater acoustic signal, wherein the separated underwater acoustic signal is represented by a time domain waveform feature.

3. The method for separating ship radiation signals from natural environment noise based on a time domain separation network according to claim 2 is characterized in that: The target convolution network includes a target convolution kernel, and the size of the target convolution kernel is determined according to the time-frequency characteristics of the target underwater acoustic signal.

4. The method for separating ship radiation signals and natural environment noise based on a time domain separation network according to claim 2 is characterized in that: The target temporal convolutional network includes a residual path, a skip path and a plurality of stacked dilated convolutional blocks, wherein the residual path of a previous dilated convolutional block in the plurality of dilated convolutional blocks serves as the input of the next dilated convolutional block, and the skip path vectors of all dilated convolutional blocks serve as the output of the target temporal convolutional network.

5. The method for separating ship radiation signals from natural environment noise based on a time domain separation network according to claim 4 is characterized in that: The number of moving steps of each layer of expansion factor corresponding to each of the multiple expansion convolution blocks increases exponentially, the input values ​​of each expansion convolution block are filled with zero values, and each expansion convolution block includes an expansion convolution kernel. In the expansion convolution kernel, ordinary convolution operations are divided into depth convolution calculation and point convolution calculation.

6. The method for separating ship radiation signals from natural environment noise based on a time domain separation network according to claim 2 is characterized in that: The separated feature representation includes a plurality of separated feature representation fragments, and the separated feature representation is inversely transformed and calculated by the target transposed convolution network to obtain a separated underwater acoustic signal, wherein the separated underwater acoustic signal is represented by a time domain waveform feature, including: Performing the inverse transformation calculation on the separated multiple feature representation segments through the target transposed convolutional network to obtain the separated multiple signal sub-segments, wherein the network parameters of the transposed convolutional network correspond to the network parameters of the target convolutional network; The separated multiple signal sub-fragments are combined to obtain the separated underwater acoustic signal.

7. A device for separating ship radiation signals and natural environment noise based on a time domain separation network, characterized in that: include: A signal acquisition module, used to acquire a target hydroacoustic signal, wherein the target hydroacoustic signal includes a ship radiation signal and natural environment noise; The signal separation module is used to input the target hydroacoustic signal into a trained time domain separation network to obtain a separated hydroacoustic signal. The trained time domain separation network separates the ship radiation signal from the natural environment noise by using a target mask corresponding to the target hydroacoustic signal to obtain a high-order nonlinear feature representation corresponding to the target hydroacoustic signal to obtain the separated hydroacoustic signal.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for separating ship radiation signals and natural environment noise based on a time domain separation network as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for separating ship radiation signals from natural environment noise based on a time domain separation network as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for separating ship radiation signals from natural environment noise based on a time domain separation network as described in any one of claims 1 to 6 is implemented.

Citation Information

Cited By

  • Direct wave signal suppression method of continuous wave active sonar detection system

    CN120722331A

  • Passive coding equipment positioning and tracking method and device

    CN121385791A

  • Passive coded device positioning and tracking method and apparatus

    CN121385791B