Underwater target radiation noise classification method, system and related device

Abstract

The invention discloses an underwater target radiation noise classification method and system, and a related device. The method comprises the following steps: obtaining radiation noise signals of the same underwater target under different working conditions and radiation noise signals of different underwater targets under the same working condition; carrying out normalization and sampling processing on the obtained radiation noise signal, and carrying out enhanced image conversion on the radiation noise signal of the underwater target after normalization and sampling processing; taking the converted image data set as the input of the convolutional neural network, and classifying the radiation noise of the underwater target; specifically, on the basis of different enhanced images and a convolutional neural network, loss of effective information is effectively reduced without a noise processing link, frequency domain transformation is not carried out, limitation of frequency domain resolution is avoided, signals in different frequency ranges can be represented more robustly, a nonlinear relation is better captured, and the method is suitable for being popularized and applied. And the method is more sensitive to some nonlinear modes.

Description

Technical Field

[0001] The invention belongs to the field of underwater acoustic signal processing and relates to a method, system and related device for classifying underwater target radiation noise. Background Art

[0002] There are many types of underwater targets, complex working conditions and high environmental noise levels. In addition, the underwater target radiation noise itself has three non-stationary, nonlinear and non-Gaussian characteristics. The coupling and superposition of multiple factors have had a huge impact on the growing underwater target recognition. Traditional underwater target recognition is carried out by using the power spectrum estimation and LOFAR (Low Frequency Analysis Recording) features of the underwater target radiation noise signal, and then using classification models such as support vector machines and decision trees to complete classification and recognition. This method has a very serious impact on noise and even leads to the failure of the recognition process. It has seriously lagged behind the growing data demand. In the prior art, the Chinese patent with publication number CN 118051831 A, based on the underwater acoustic target recognition method of the CNN-Transformer cooperative network model, proposes to convert the audio data of the underwater acoustic signal to be identified into a mel spectrum map through a mel filter group, and input it into the pre-trained CNN-Transformer cooperative network model. The model combines the local feature extraction capability of CNN and the global feature extraction capability of SwinTransformer that integrates the channel attention mechanism. The mel spectrum conversion of the audio data based on the noise is still performed, which cannot avoid the loss of effective information in the noise processing stage, and is insensitive to nonlinear signals, especially in underwater noise environments, where various external interferences and nonlinear signals exist in large quantities. A Chinese patent with publication number CN 116524913 A, a method and device for classifying ship radiated noise, provides: extracting features from ship radiated noise signals, selecting features with high importance for classification and weighting them; passing the weighted features into a neural network classification model for classification, but this method is only sensitive to ship radiated noise classification, and has low universality. However, the underwater environment is complex, and a large number of other noise sources still need to be extracted.

[0003] In view of the above problems, there is an urgent need for a method that has little effect on noise disturbance and has high classification accuracy and efficiency. Summary of the invention

[0004] In order to solve the problems existing in the prior art, the present invention provides a method for classifying underwater target radiation noise, which is specifically based on different enhanced images and convolutional neural networks. It does not go through the noise processing link, effectively reduces the loss of effective information, does not perform frequency domain transformation, is not limited by the frequency domain resolution, and can more robustly perform signals in different frequency ranges, better capture nonlinear relationships, and is more sensitive to some nonlinear modes.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: a method for classifying underwater target radiated noise, comprising the following steps: Obtain the radiation noise signals of the same underwater target under different working conditions and the same working conditions under different underwater targets; The obtained radiation noise signal is normalized and sampled. Enhanced image conversion of the noise signal radiated by underwater targets after normalization and sampling processing; The converted image dataset is used as the input of the convolutional neural network to classify the radiation noise of underwater targets.

[0006] Furthermore, the collected radiation noise signal is normalized and sampled, including: adjusting the data to a uniform scale; converting the continuous signal into a discrete signal, the sampling frequency must be at least twice the bandwidth of the radiation noise signal, converting the radiation noise signal into a state vector in a high-dimensional space to capture the characteristics of the signal, calculating the similarity or transfer characteristics of the radiation noise signal between different states, so as to evaluate the dynamic changes and behavior patterns of the signal and obtain the internal correlation of the radiation noise signal.

[0007] Furthermore, the dimension of the standard normal distribution Z is simplified by piecewise aggregation. By selecting the downscaling factor k and generating a new data sequence, the dimension N is simplified to m:

[0008] This allows a reduction from N dimensions to m dimensions while retaining the approximate trend of the original standard normal distribution Z.

[0009] Further, the enhanced image conversion of the underwater target radiation noise signal after normalization and sampling processing includes matrixing the processed radiation noise signal, and then performing time alignment or position alignment, which specifically includes the following steps: S31, matrixing the signal sequence, converting the one-dimensional sequence into a two-dimensional matrix, dividing the discrete signal into multiple blocks, and arranging each block into rows or columns of the matrix; S32, perform time alignment or position alignment, where the position alignment P(i,j)=g(xi,xj), where i and j represent the states of the sequence, and g(·) is a function that measures the similarity between states; for time alignment, for a time alignment generating matrix M, M(i,j)=f(xi,xj), where i and j are time indices, and f(⋅) is some similarity measure.

[0010] Furthermore, when the discrete signal is divided into a plurality of blocks and each block is arranged into rows or columns of a matrix, the Lagham and corner field methods are used for matrixing through the Lagham matrix, the Markov migration field is used for matrixing based on the Markov transfer matrix, the recursive graph is used for matrixing through the recursive matrix, or the recursive graph is used for matrixing through the relative position matrix.

[0011] Furthermore, the Lagum and angle field method, Markov migration field or relative position matrix are used for position alignment, and the recursive graph is used for time alignment.

[0012] In a second aspect, the present invention can provide an underwater target radiation noise classification system, including a data acquisition module, a data processing module, an enhanced image conversion module and a classification recognition module; The data acquisition module is used to acquire the radiation noise signals of the same underwater target under different working conditions and the radiation noise signals of the same underwater target under different working conditions; The data processing module is used to normalize and sample the obtained radiation noise signal. The enhanced image conversion module is used to enhance the image conversion of the underwater target radiation noise signal after normalization and sampling processing; The classification and recognition module is used to take the converted image dataset as the input of the convolutional neural network to classify the radiation noise of underwater targets.

[0013] Furthermore, the enhanced image conversion module mainly performs matrix processing of digital signals, intercepts data using a sliding window method, establishes a matrix, assigns different color values ​​to different values ​​after achieving time alignment and position alignment, completes image conversion, and establishes features for further image recognition; The classification and recognition module performs classification and recognition of enhanced images, and establishes, trains and uses an image recognition model based on a convolutional neural network. In this process, a basic convolutional neural network model is first established, and the converted enhanced image is used as input for model training. The model parameters are continuously iterated, and the model parameters are optimized with classification and recognition accuracy as the goal. Finally, the optimal convolutional neural network model is obtained for subsequent image classification and recognition. The classification and recognition module is the fully connected layer and result output layer of the convolutional neural network module, which is used to extract the feature map of the convolutional neural network and output it as the recognition result.

[0014] The present invention can also provide a computer device, including a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, the underwater target radiation noise classification method described in the present invention can be implemented.

[0015] A computer-readable storage medium is also provided, in which a computer program is stored. When the computer program is executed by a processor, the underwater target radiation noise classification method of the present invention can be implemented.

[0016] Compared with the prior art, the present invention has at least the following beneficial effects: The underwater target radiation noise classification method based on different enhanced images and convolutional neural networks of the present invention can better retain the global characteristics of the original underwater target radiation noise signal data, reveal new patterns and structural features in the data, and help capture more complex relationships in the signal data; the enhanced image method does not perform frequency domain transformation and is not limited by the frequency domain resolution, and can more robustly perform signals in different frequency ranges; the enhanced image method takes into account the characteristics of the entire signal data, reduces the complexity of parameter selection, and the method is more general; the enhanced image method better captures nonlinear relationships and is more sensitive to some nonlinear patterns. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the underwater target radiation noise classification method of the present invention; Figure 2 This is the specific implementation case result of the present invention; Figure 3 It is a convolutional neural network structure diagram of the present invention; Figure 4 The image classification results converted by the various image enhancement methods of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments 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.

[0019] In the description of the present invention, it should be understood that the terms “include” and “comprises” indicate the presence of described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0020] Traditional underwater target recognition is carried out by using power spectrum estimation of underwater target radiation noise signals and LOFAR (Low Frequency Analysis Recording) and other features, and then using classification models such as support vector machines and decision trees to complete classification and recognition. This method has a very serious impact on noise and may even lead to the failure of the recognition process. It has seriously lagged behind the growing data demand.

[0021] In view of the above problems, there is an urgent need for a method that has little effect on noise disturbance and has high classification accuracy and efficiency. Based on the method of enhanced image and convolutional neural network, the original underwater target radiation noise signal is converted into an image through the enhanced image method without going through the noise processing link, which can effectively reduce the loss of effective information.

[0022] like Figure 1 As shown, a method for classifying underwater target radiation noise based on different enhanced images and convolutional neural networks includes the following steps: Step 1, obtaining the radiation noise of the same underwater target under different working conditions and the radiation noise of different underwater targets under the same working condition; Step 2, processing the radiation noise signal, including normalization and sampling, to enhance stability; Step 3, the processed underwater target radiation noise signal is converted into an enhanced image by using a variety of methods; Step 4: Use the converted image dataset as the input of the convolutional neural network for model training to accurately classify the radiation noise of underwater targets.

[0023] like Figure 2 As shown, there are many methods for converting signals into images based on enhanced images, including but not limited to Lagum and angular field methods, Markov migration field method, recursive graph method, relative position matrix method. The present invention selects the relative position matrix method as a case to elaborate on its implementation method, including the following steps; Step 1: For the original underwater target radiation noise, a standard normal distribution Z is obtained by Z-score standardization: , where μ represents the average value of X, represents the standard deviation of X.

[0024] Step 2: The standard normal distribution Z has a high dimension. The dimension of Z can be reduced to m through piecewise aggregate approximation (PAA). By selecting a suitable downscaling factor k and generating a new data series, the dimension N is reduced to m:

[0025] This allows a reduction from N dimensions to m dimensions while retaining the approximate trend of the original standard normal distribution Z.

[0026] Step 3: Construct an m × m matrix by signal matrixing, align the relative positions between the calculation timestamps, and convert the preprocessed data sequence X into a two-dimensional matrix M:

[0027] The data sequence is represented by a matrix M, where every two timestamps are concatenated with M to get their relative position. Each row and column of M contains information about the entire data sequence with one timestamp as a reference point. Each row of the matrix M shows the data sequence at a different reference point and each column shows the mirror image of the former, which provides an opposite view of the data sequence.

[0028] Step 4, the application of min-max normalization converts M into a gray value matrix, which in turn produces a relative displacement matrix F:

[0029] like Figure 3 As shown in the figure, it is a convolutional neural network structure diagram, including enhanced image conversion module, CNN module, classification and recognition module, which can realize the classification of underwater noise images with high accuracy and determine the source of underwater noise. In the classification and recognition module: for an input image and convolution kernel , the formula for the convolution operation is as follows:

[0030] in are the coordinates of the output feature map, a and b are half of the height and width of the convolution kernel, respectively.

[0031] Using the activation function ReLU (Rectified Linear Unit):

[0032] For the maximum pooling operation, assuming the pooling window size is ,but:

[0033] Fully connected layer: Let the feature vector after convolution and pooling be h, the output of the fully connected layer can be expressed as:

[0034] Where W is the weight matrix and b is the bias term.

[0035] Output layer (Softmax function), the softmax function of the output layer converts the output of the fully connected layer into a probability distribution:

[0036] in is the output corresponding to class j.

[0037] The classification performance is evaluated using the intersection-entropy loss function:

[0038] in is the true label, is the predicted probability.

[0039] The enhanced image conversion module mainly performs matrix processing of digital signals. Different from the traditional sliding window screenshot method, the enhanced image conversion module uses a sliding window method to intercept data and establishes a matrix through a specific mathematical method, including but not limited to the Lagum angular field method, recursive graph method, relative position matrix method, wavelet packet method and Markov migration method. After achieving time alignment and position alignment, different color values ​​are assigned to different values ​​to complete the image conversion, retaining the original signal as much as possible, and establishing features for further image recognition.

[0040] The CNN module is mainly used for classification and recognition of enhanced images. Its core is the establishment, training and use of image recognition models based on convolutional neural networks. In this process, a basic convolutional neural network model is first established, and the converted enhanced image is used as input for model training. During the training process, the model parameters are continuously iterated, with classification and recognition accuracy as the goal, and the model parameters are optimized. Finally, the optimal convolutional neural network model is obtained for subsequent image classification and recognition.

[0041] The classification recognition module is the fully connected layer and result output layer of the CNN module. It is used to extract the feature map of the convolutional neural network and output it as the recognition result. It can also serve as a visualization module for the entire model.

[0042] For convolutional neural networks, a two-dimensional convolution kernel is selected, and a convolution kernel of a specific size is determined according to the image size to extract the features of the input image; the convolution layer has an adjustable sliding step size to control the step size of the convolution kernel sliding on the input image; the optimizer is an adaptive optimization algorithm used to update the network weights to optimize the loss function, which includes but is not limited to the cross entropy loss function; the batch size is set to facilitate the number of samples used in each iteration; the maximum training rounds are used to control the number of iterations of model training.

[0043] like Figure 4As shown in the figure, the image classification results converted by various enhanced image methods are based on the same set of underwater target radiation noise signals, enhanced image conversion is performed by different methods, and classification is performed based on the same underwater target radiation noise classification model, resulting in accuracy and calculation time.

[0044] Methods 1-4 are wavelet packet method, Markov migration field method, recursive graph, relative position matrix method and Lagum and angular field method respectively. From the results of classification and recognition, it can be seen that after the image is matrixed by different methods, the calculation consumption generated by its classification and recognition is different. The essence is that the kernel calculation method is different and the amount of data generated is different. At the same time, the classification accuracy is also very different. The essence is that the features generated by the different calculation kernels are different. The method with larger difference features has higher classification accuracy. Therefore, the Lagum angular field method has short calculation time and high classification accuracy, but it does not mean that the Lagum angular field is suitable for all underwater target radiation noise classification. This example only illustrates the simulated data. In fact, in the actual underwater target radiation noise classification, multiple methods need to be used in combination. According to different scenarios and different requirements for calculation time and calculation accuracy, specific methods are used in a targeted manner, but the final recognition model still uses the convolutional neural network model.

[0045] The present invention also provides an underwater target radiation noise classification system, comprising a data acquisition module, a data processing module, a conversion module and a classification and recognition module; The data acquisition module is used to acquire the radiation noise signals of the same underwater target under different working conditions and the radiation noise signals of the same underwater target under different working conditions; The data processing module is used to normalize and sample the obtained radiation noise signal. The conversion module is used for enhancing the image conversion of the underwater target radiation noise signal after normalization and sampling processing; The classification and recognition module is used to take the converted image dataset as the input of the convolutional neural network to classify the radiation noise of underwater targets.

[0046] The enhanced image conversion module mainly performs matrix processing of digital signals. Different from the traditional sliding window screenshot method, the enhanced image conversion module uses a sliding window method to intercept data and establishes a matrix through a specific mathematical method, including but not limited to the Lagum angular field method, recursive graph method, relative position matrix method, wavelet packet method and Markov migration method. After achieving time alignment and position alignment, different color values ​​are assigned to different values ​​to complete the image conversion, retaining the original signal as much as possible, and establishing features for further image recognition.

[0047] The CNN module is mainly used for classification and recognition of enhanced images. Its core is the establishment, training and use of image recognition models based on convolutional neural networks. In this process, a basic convolutional neural network model is first established, and the converted enhanced image is used as input for model training. During the training process, the model parameters are continuously iterated, with classification and recognition accuracy as the goal, and the model parameters are optimized. Finally, the optimal convolutional neural network model is obtained for subsequent image classification and recognition.

[0048] The classification recognition module is the fully connected layer and result output layer of the CNN module. It is used to extract the feature map of the convolutional neural network and output it as the recognition result. It can also serve as a visualization module for the entire model.

[0049] On the other hand, the present invention provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the underwater target radiation noise classification method described in the present invention can be implemented.

[0050] The present invention can also provide a computer device, including a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, the underwater target radiation noise classification method described in the present invention can be implemented.

[0051] The computer device may be a laptop computer, a desktop computer or a workstation.

[0052] The processor may be a central processing unit (CPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), or an off-the-shelf field programmable gate array (FPGA).

[0053] The memory of the present invention may be an internal storage unit of a laptop computer, a desktop computer or a workstation, such as a memory or a hard disk; or an external storage unit, such as a mobile hard disk or a flash memory card.

[0054] Computer-readable storage media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer-readable storage media may include: read-only memory (ROM), random access memory (RAM), solid-state drive (SSD) or optical disk, etc. Among them, random access memory may include resistance random access memory (ReRAM) and dynamic random access memory (DRAM).

[0055] The above contents are only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A method for classifying underwater target radiated noise, characterized in that: The following steps are involved: Obtain the radiation noise signals of the same underwater target under different working conditions and the same working conditions under different underwater targets; The obtained radiation noise signal is normalized and sampled. Enhanced image conversion of the noise signal radiated by underwater targets after normalization and sampling processing; The converted image dataset is used as the input of the convolutional neural network to classify the radiation noise of underwater targets.

2. The underwater target radiated noise classification method according to claim 1, characterized in that: The normalization and sampling processing of the collected radiation noise signal includes: adjusting the data to a uniform scale; converting the continuous signal into a discrete signal, the sampling frequency must be at least twice the bandwidth of the radiation noise signal, converting the radiation noise signal into a state vector in a high-dimensional space to capture the characteristics of the signal, and calculating the similarity or transfer characteristics of the radiation noise signal between different states to evaluate the dynamic changes and behavior patterns of the signal and obtain the internal correlation of the radiation noise signal.

3. The underwater target radiated noise classification method according to claim 1, characterized in that: The dimension of the standard normal distribution Z is simplified by piecewise aggregation. By selecting the downscaling factor k and generating a new data sequence, the dimension N is reduced to m: This allows a reduction from N dimensions to m dimensions while retaining the approximate trend of the original standard normal distribution Z.

4. The underwater target radiated noise classification method according to claim 1, characterized in that: The enhanced image conversion of the underwater target radiation noise signal after normalization and sampling processing includes: matrixing the processed radiation noise signal, and then performing time alignment or position alignment, which specifically includes the following steps: S31, matrixing the signal sequence, converting the one-dimensional sequence into a two-dimensional matrix, dividing the discrete signal into multiple blocks, and arranging each block into rows or columns of the matrix; S32, perform time alignment or position alignment, where the position alignment P(i,j)=g(xi,xj), where i and j represent the states of the sequence, where g(·) is a function that measures the similarity between states; for time alignment, for a time alignment generating matrix M, M(i,j)=f(xi,xj), where i and j are time indices, where f(⋅) is some similarity measure.

5. The underwater target radiation noise classification method according to claim 4, characterized in that: When a discrete signal is divided into blocks and each block is arranged into rows or columns of a matrix, Lagham and corner field methods are used for matrixing through Lagham matrices, Markov migration fields are used for matrixing based on Markov transfer matrices, recurrence graphs are used for matrixing through recurrence matrices, or through relative position matrices.

6. The underwater target radiation noise classification method according to claim 4, characterized in that: Lagum and angle field methods, Markov migration fields or relative position matrices are used for position alignment, and recursive graphs are used for time alignment.

7. A classification system for underwater target radiated noise, characterized in that: It includes a data acquisition module, a data processing module, an enhanced image conversion module and a classification and recognition module; The data acquisition module is used to acquire the radiation noise signals of the same underwater target under different working conditions and the radiation noise signals of the same underwater target under different working conditions; The data processing module is used to normalize and sample the obtained radiation noise signal. The enhanced image conversion module is used to enhance the image conversion of the underwater target radiation noise signal after normalization and sampling processing; The classification and recognition module is used to take the converted image dataset as the input of the convolutional neural network to classify the radiation noise of underwater targets.

8. The underwater target radiated noise classification system according to claim 7, characterized in that: The enhanced image conversion module mainly performs matrix processing of digital signals, intercepts data using a sliding window method, establishes a matrix, assigns different color values ​​to different values ​​after achieving time alignment and position alignment, completes image conversion, and establishes features for further image recognition; The classification and recognition module performs classification and recognition of enhanced images, and establishes, trains and uses an image recognition model based on a convolutional neural network. In this process, a basic convolutional neural network model is first established, and the converted enhanced image is used as input for model training. The model parameters are continuously iterated, and the model parameters are optimized with classification and recognition accuracy as the goal. Finally, the optimal convolutional neural network model is obtained for subsequent image classification and recognition. The classification and recognition module is the fully connected layer and result output layer of the convolutional neural network module, which is used to extract the feature map of the convolutional neural network and output it as the recognition result.

9. A computer device, characterized in that: The invention comprises a processor and a memory, wherein the memory is used to store a computer executable program, and the processor reads part or all of the computer executable program from the memory and executes the program. When the processor executes part or all of the computer executable program, the underwater target radiation noise classification method according to any one of claims 1 to 7 can be implemented.

10. A computer-readable storage medium, characterized in that: A computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the underwater target radiation noise classification method according to any one of claims 1 to 7 can be implemented.

Images