CN112329523A - Underwater acoustic signal type identification method, system and equipment - Google Patents

Underwater acoustic signal type identification method, system and equipment Download PDF

Info

Publication number
CN112329523A
CN112329523A CN202011021846.5A CN202011021846A CN112329523A CN 112329523 A CN112329523 A CN 112329523A CN 202011021846 A CN202011021846 A CN 202011021846A CN 112329523 A CN112329523 A CN 112329523A
Authority
CN
China
Prior art keywords
layer
network
neural network
dense connection
data set
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011021846.5A
Other languages
Chinese (zh)
Inventor
王岩
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Taishan University
Original Assignee
Taishan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Taishan University filed Critical Taishan University
Priority to CN202011021846.5A priority Critical patent/CN112329523A/en
Publication of CN112329523A publication Critical patent/CN112329523A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Signal Processing (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)

Abstract

The application discloses a method, a system and equipment for identifying the type of an underwater acoustic signal, which are based on a deep dense connection neural network, wherein the method comprises the following steps: preprocessing various tuning signals transmitted by underwater communication to obtain an input signal type original data set; determining a deep dense connection neural network for processing the input signal kind original data set; and identifying the type of the output underwater sound signal through the deep dense connection neural network. The deep dense connection neural network solves the problem that a cross-layer connection method is easier to fit a deeper model, and in the communication process, analysis in a physical layer signal form is simple, and the deeper model is less involved in use. Through the cross-layer connection, the overfitting phenomenon can be prevented, the generalization capability of the model is improved, and the classification precision of the model is improved.

Description

Underwater acoustic signal type identification method, system and equipment
Technical Field
The application relates to the technical field of signal identification, in particular to a method, a system and equipment for identifying underwater acoustic signal types.
Background
In recent years, a modulation identification technique as a core of a non-cooperative communication system has been intensively studied. As an indispensable part of modern communication systems, modulation and identification show that the system efficiency is improved, and a great deal of practical values such as electronic countermeasure, wireless signal cracking and the like are realized in the military and civil fields. Especially in software radio (including cognitive radio), modulation identification has become a key part in implementing radio system intelligence. In the military field, the method mainly relates to the further identification and processing after the capture of enemy signals in electronic warfare. In order to effectively use interfering hostile signals to prevent hostile communications, higher power signals must be transmitted to exceed hostile signal power in the same frequency band. It is clear that above all the transmitted interfering signals should have the same modulation scheme as the interfering signals detected by the modulation classifier. In the civil field, a flexible LA (Link-Adaptation) system can adaptively select an efficient modulation scheme according to a channel condition obtained by measurement to improve the transmission efficiency of a communication system. However, research on automatic signal modulation type identification has not been widely used in the field of underwater communication.
From a communication point of view, the carrier of underwater information transmission is water itself, which is equivalent to air for land wireless communication. Underwater channels are much more complex than terrestrial wireless channels, because the absorptive nature of underwater electromagnetic waves prevents the waves from propagating long distances in underwater environments, and many classical terrestrial radio communication techniques cannot be used directly in underwater communication processes. The underwater channel is a non-uniform and random medium channel and has the characteristics of time dispersion and slow fading. The energy loss increases not only with distance but also with frequency, so the available bandwidth is narrow (only a few kilohertz), the channel capacity is small, and multipath interference is greatly affected as space-variant during propagation. When the channel bandwidth is limited and the capacity is small, an effective modulation scheme is particularly important to improve the communication efficiency. Modulation and demodulation is the basis of digital communication systems, which is one of the most basic and important links. In an underwater communication system, PSK (Phase Shift Keying) and QAM (Quadrature Amplitude Modulation) are applied. Since the interference is more severe in the underwater environment and the signal is more affected than terrestrial wireless communication, the modulation constellation becomes blurred and the phase shift phenomenon becomes apparent.
Artificial Neural Networks, particularly CNNs (Convolutional Neural Networks), have been vigorously developed since the 1990's. In the field of communications, modulation identification and signal identification are mainly accomplished by using shallow artificial neural networks. Until 2012, AlexNet began to make breakthroughs in image recognition, from ZFNet to VGGNet, *** lenet, and then to ResNet. As shown in FIG. 1, a typical network architecture is VGGNet, where Conv is the convolution layer, 3 × 3 is the kernel size, and 64-512 is the number of image filters. pool is a pooling layer, with 1 to 5 being used to distinguish different pooling layers. fc is the full connection layer and 4096 is the number of hidden units. However, these deep neural networks also generate a high-dimensional nonlinear hyper-parametric space that is difficult to search, resulting in an overfitting phenomenon and poor generalization ability. In this case, the network is deeper and deeper, and the architecture is more and more complex. To improve the generalization capability of the deep network model, many smart solutions are adopted to reduce the problems of overfitting appearance and gradient disappearance in back propagation, such as Dropout, Batch Normalization (BN), and ReLU nonlinear functions, etc. In the field of computer vision, CNN has become the most popular method. Also, the field of communications has begun to use deep learning methods, such as ANN-based machine learning algorithms for dealing with various wireless network problems. Many studies relate to algorithms based on deep learning, such as physical layer communication process analysis, channel decoding by a combination of CNN and BP (Belief Propagation) and using RNN (Recurrent Neural Network) to identify and detect data sequences in a communication system. However, it can be seen that deeper and broader networks can learn more about the functionality of the data set distribution, but they also lead to problems of overfitting, making the model unusable.
Disclosure of Invention
In order to solve the technical problems, the following technical scheme is provided:
in a first aspect, an embodiment of the present application provides an underwater acoustic signal type identification method, which is based on a deep dense connection neural network, and the method includes: preprocessing various tuning signals transmitted by underwater communication to obtain an input signal type original data set; determining a deep dense connection neural network for processing the input signal kind original data set; and identifying the type of the output underwater sound signal through the deep dense connection neural network.
By adopting the implementation mode, the deep dense connection neural network solves the problem that a cross-layer connection method is easier to fit a deeper model, and in the communication process, analysis in a physical layer signal form is simple, and the deeper model is less involved in use. Through the cross-layer connection, the overfitting phenomenon can be prevented, the generalization capability of the model is improved, the model is deepened, and the classification precision is improved.
With reference to the first aspect, in a first possible implementation manner of the first aspect, the determining a deep-dense-connected neural network that processes the input signal category raw data set includes: determining the number of layers and the sequence of each layer of the deep dense connection neural network; the network structure included in each layer of neural network is determined.
With reference to the first possible implementation manner of the first aspect, in a second possible implementation manner of the first aspect, the deep-dense-connection neural network includes, from a network input end to a network output end: the device comprises an input layer, a pretreatment layer, a middle dense connection layer and an output layer.
With reference to the second possible implementation manner of the first aspect, in a third possible implementation manner of the first aspect, the input layer converts the input signal type raw data set into an input data set for further learning by an input network through preset data adjustment.
With reference to the second possible implementation manner of the first aspect, in a fourth possible implementation manner of the first aspect, the preprocessing layer includes a first Conv layer and a first BN layer, the first Conv layer is configured to extract data features of an input data set, and the first BN layer is configured to normalize data set distribution during each forward propagation.
With reference to the second possible implementation manner of the first aspect, in a fifth possible implementation manner of the first aspect, the intermediate dense connection layer includes a second BN layer, a ReLU layer, and a second Conv layer, where the ReLU layer is a nonlinear activation function layer, and the ReLU function form is:
Figure BDA0002700883800000041
after the ReLU layer is activated nonlinearly, the closer to x, the more relevant the characteristics are; the closer to 0, the smaller the correlation; in feature extraction, irrelevant data is directly discarded to reduce the amount of data.
With reference to the second possible implementation manner of the first aspect, in a sixth possible implementation manner of the first aspect, the output layer is a softmax activation functional layer:
Figure BDA0002700883800000042
where i represents one of the various categories and Q represents the total number of all categories.
With reference to the first aspect, in a seventh possible implementation manner of the first aspect, in the training process of the deep-dense-connection neural network, each layer in the network extracts only part of data features, and discarding part of network layers randomly does not affect convergence of the network.
In a second aspect, an embodiment of the present application provides an underwater acoustic signal class identification system, which is based on a deep dense connection neural network, and the system includes: the preprocessing module is used for preprocessing data of various tuning signals transmitted by underwater communication to obtain an input signal type original data set; the training module is used for determining a deep dense connection neural network for processing the input signal type original data set; and the identification output module is used for identifying the type of the output underwater sound signal through the deep dense connection neural network.
In a third aspect, an embodiment of the present application provides an apparatus, including: a processor; a memory for storing computer executable instructions; when the processor executes the computer-executable instructions, the processor executes the underwater acoustic signal type identification method described in the first aspect or any one of the possible implementation manners of the first aspect to perform type identification on the underwater acoustic signal.
Drawings
Fig. 1 is a schematic structural diagram of a VGGNet provided in an embodiment of the present application;
fig. 2 is a schematic flowchart of a method for identifying a type of an underwater acoustic signal according to an embodiment of the present application;
fig. 3 is a schematic diagram of a ResNet structure provided in an embodiment of the present application;
fig. 4 is a schematic structural diagram of a basic unit of a short-connection fast residual error network according to an embodiment of the present application;
fig. 5 is a schematic diagram of a dense connection network structure provided in an embodiment of the present application;
FIG. 6 is a schematic diagram of an underwater communication process model provided by an embodiment of the present application;
fig. 7 is a schematic diagram illustrating an identification effect of a dense connection network provided in an embodiment of the present application under different loss functions;
fig. 8 is a schematic diagram illustrating an identification effect of a dense connection network provided in an embodiment of the present application under different optimizers;
fig. 9 is a schematic view illustrating an identification effect of a dense connection network provided in an embodiment of the present application in different layers;
fig. 10 is a schematic diagram of an underwater acoustic signal type identification system according to an embodiment of the present application;
fig. 11 is a schematic diagram of an apparatus provided in an embodiment of the present application.
Detailed Description
The present invention will be described with reference to the accompanying drawings and embodiments.
Fig. 2 is a schematic flowchart of a method for identifying a type of an underwater acoustic signal according to an embodiment of the present application, with reference to fig. 2, where the method includes:
s101, preprocessing data of various tuning signals transmitted by underwater communication to obtain an input signal type original data set.
S102, determining a deep dense connection neural network for processing the input signal type original data set.
The deep learning network can effectively learn the modulation profile characteristics of the received signal data set. Given that deeper networks are prone to overfitting problems, the network-learned feature parameters can mitigate this phenomenon by using a cross-layer connection sharing approach. In addition, cross-layer connections can improve the transmission efficiency of parameters in the network, thereby reducing computational complexity. The designed network can fully learn more characteristics of the signal data set, so that the robustness to Doppler frequency shift and multipath interference is better.
The deep learning network architecture can effectively improve the classification effect by deepening the network structure. However, deep network architectures can easily lead to model degradation and gradient vanishing. In deeper networks, both of these cases are more likely to occur and become more severe. As the number of layers of the network increases, the input information and the gradient information generated after each layer will gradually decrease and eventually may be washed away. To solve the degradation and gradient disappearance problems caused by the deepening of the network architecture, the solution is to minimize the connection between the front and back layers. This effectively passes the learned gradient information between the layers.
In order to alleviate the deep learning network training problem brought by the two problems, a ResNet network structure form is designed. A typical ResNet network structure is shown in fig. 3. These network architectures have in common that: a shortcut is established between each layer and subsequent layers. With the advent of the ResNet model, deeper CNN models can be trained in this way to achieve higher accuracy. With the deepening of a network architecture, the probability distribution of the data set can be more fully learned, and therefore higher identification accuracy can be achieved. The premise for practical use of deep networks is to overcome the problem of overfitting. The core of the ResNet model is to establish short-cut connections between the front and back layers, which helps to propagate the gradient information backwards during the training process to train deeper CNN networks. Deeper network structures can learn more data set representations, thereby achieving more effective training effects and higher recognition accuracy.
The basic structure of a typical short-connection, shortcut connection network element is shown in fig. 4. Consider the input and output relationships in a network as:
Figure BDA0002700883800000071
by gradient method, directly find
Figure BDA0002700883800000072
Degradation problems are encountered. The degradation problem is that a small number of hidden units in each layer change their activation values for different inputs and that most of the hidden units respond differently to the same input. Eventually, the direction of the gradient is inaccurate, resulting in a trained model being unusable. This is because the learning curve slows down the velocity substantially in the direction of the parameter space degradation, thereby reducing the effective size of the model. Training the network model is equivalent to having the model look for parameters that fit the data set. In practice, the degree of freedom with which the model can be fitted efficiently is reduced. If a short-cut network-like structure is used, the optimization of the variable parameters is no longer aimed at
Figure BDA0002700883800000073
While
Figure BDA0002700883800000074
To indicate the part to be optimized
Figure BDA0002700883800000075
Figure BDA0002700883800000076
Where z-m equals the observed value in the network unit hypothesis.
Figure BDA0002700883800000077
The residuals corresponding to short connections are called residual networks. The purpose of the network is to learn the residuals
Figure BDA0002700883800000078
Than direct learning
Figure BDA0002700883800000079
And is simpler. Now only the difference between the input and the output needs to be learned, the absolute number becomes the relative number.
Figure BDA00027008838000000710
Is the amount of change in output relative to input and is much simpler to optimize.
This type of network is mainly formed by using the original input information directly in the next layer or using the result of processing the input information in the previous layer. This maximizes the flow of information between layers in the network. In the back propagation process, the input gradient information contains the result of a loss function that directly derives the input information. This is more favorable to the gradient propagation between the network layers, thus effectively solving the problem of model degradation.
The specific short connection residual error quick connection structure is as follows:
Figure BDA0002700883800000081
wherein n isjIs the output of the current network layer j, which is the output zj-1And network input nj-1Then by activating the function
Figure BDA0002700883800000082
When the input z is in its most general form
z＝vn+η (5)
Where n is the input, v is the weight of the network, and η is the deviation. Equation (4) can be written as
Figure BDA0002700883800000083
By adding a direct connection from input to output, the problem of network performance degradation can be solved and efficient training of the network can also be achieved. Assuming that from layer j to layer j-1, from the fully-connected layer, conventionally
Figure BDA0002700883800000084
Reference (6), short-connection residual shortcut structure delivers the inverse derivative. The reverse conversion process is
Figure BDA0002700883800000085
The training speed can be improved and the network performance can be effectively restrained by the formula (8). As the network becomes very deep, training on the network can become very slow. With the normalization of the weighting values, the weighting at this time becomes very small. With the addition of a multi-stage stack, the gradient in the layers is small. If 1 is added in this process, the model can increase the gradient and make the network easier to train. This form thus avoids degradation of the model and the best matching size can be found in the correct gradient direction.
In view of this, the present application uses a deep dense connected neural network that determines the processing of the raw data set of the input signal class as shown in fig. 5. Specifically, the number of layers and the order of each layer of the deep-dense connection neural network and the network structure included in each layer of the neural network are determined.
In this embodiment, the deep-dense connection neural network includes, from a network input end to a network output end: the device comprises an input layer, a pretreatment layer, a middle dense connection layer and an output layer.
The input layer represents underwater communication data including various modulation forms. After being adjusted by basic data, the data is converted into a data format which can be input into a network for further learning.
The pretreatment layer consists of a first Conv layer and a first BN layer. The first Conv layer and the first BN layer indicate that the layer has both of these functions. The first Conv layer is used to extract data features of the input data set, and the main purpose of the first BN layer is to normalize the data set distribution during each forward propagation. The values in the training data are all in the same order, which makes the values obtained more stable.
The intermediate dense connection layer includes a second BN layer, a ReLU layer, and a second Conv layer. Although deeper networks lead to better classification results, it also results in a large number of network parameters. The large number of network parameters consumes more hardware resources and the actual utilization of the network is low. Some layers cannot effectively learn the features of the data set and can therefore be selectively deleted. In the random deep network, each layer in the training process is found to randomly discard some layers, which can significantly improve the expression capability of ResNet. This means that the network does not have to be a progressive hierarchy. That is, a certain layer in the network does not have to obtain the features of the data set through the immediately upper layer, and the features of the data set may also be learned from the features provided by the upper layer. During the training process, many layers that are randomly dropped do not affect the convergence of the network. This indicates that ResNet still has significant redundancy, indicating that each layer in the network extracts only partial data features. The trained ResNet randomly discards several layers, and has little influence on the prediction result. This means that the learned features at each level are too small to improve network efficiency by reducing redundancy.
The unit module layers in the intermediate dense connection layer are specifically explained as follows:
ReLU (rectified Linear Unit) represents a layer of nonlinear activation function, the ReLU function being of the form
Figure BDA0002700883800000101
The ReLU layer is non-linearly activated. The closer to x, the more relevant the feature. The closer to 0, the smaller the correlation. In feature extraction, irrelevant data is directly discarded to reduce the amount of data.
The output layer outputs the final classification, which is implemented by the softmax activation function layer. Which represents the final decision output of the result.
Figure BDA0002700883800000102
Where i represents one of the various categories and Q represents the total number of all categories.
Each layer takes as input the output of the previous layer, assuming the network has a total D connection. For the designed network, there is one layer D3 +4(D ≧ 2). The latter layers can directly use the original input information, which maximizes the information flow between the layers in the network. Meanwhile, in the process of back propagation, the condition is more favorable for gradient propagation. In addition, the connection of a plurality of short connection quick residual modes has a regularization effect and has a good inhibition effect on overfitting. Moreover, the network width is narrower, the parameter quantity can be controlled in a reasonable range, and the underwater acoustic communication system can be embedded more easily.
And S103, identifying the type of the output underwater sound signal through the deep dense connection neural network.
A general underwater communication system is shown in fig. 6. The receiver receives the modulated signal in the form of
Figure BDA0002700883800000103
Where g (t) is the original signal to be transmitted, consisting of (0, 1), and s (t) is the modulated signal. After the signal is sent through the underwater channel, the transmitted signal is not only affected by the underwater transmission medium itself, but also interfered by external additional noise. Can represent the sum of the channel characteristics {. and the external noise n (t). The symbol {. here denotes a transformation, the input signal reacting to the output modulation signal s' (t) via the channel characteristic δ (ε, R, t, B). The channel δ (ε, R, t, B) is the impulse response function, ε is the pulse duration, R is the data rate, t is time, and B is the finite bandwidth. After demodulation by the receiving hydrophone, the last signal received at the receiver is g' (t).
Shallow sea channel model parameters are used in the simulation experiments because the shallow sea environment is more complex and better reflects the recognition performance in the model. The simulation uses channel parameters generated by the actual environment. Meanwhile, the simulation based on the radio platform defined by software can better reflect the real underwater acoustic communication environment. The depth of the transmitting end and the receiving end is 10m, the distance between the transmitting end and the receiving end is 5000m, the carrier wave is 10kHz, and the wind speed is 20 knots. The SNR ranges from-20 dB to 20 dB. The noise is assumed to be additive white gaussian noise with standard deviation and zero mean.
In a first step, in order to obtain better model performance, model parameters suitable for the underwater communication data set need to be selected.
These parameters include mainly the loss function, the optimizer and the number of model layers. Fig. 7 to 9 show the model training effect under different loss functions, optimizers, model depths and filters.
FIG. 7 illustrates the effect of using different loss functions (lossfunction) on network model training. Cross entropy (cross entropy) has a distinct advantage over mse. At higher SNR (SNR > 0), the difference is smaller. This is mainly because at high SNR the interference is small and the constellation of the signal modulation can be better distinguished. The advantage of using cross entropy is even more pronounced when the loss function is at low SNR (SNR < 0). Mainly comes from serious interference under low SNR, and the judgment standard of the model can be better evaluated by selecting a proper loss function, and the recognition rate is improved to a certain extent.
FIG. 8 also illustrates the impact of selecting different optimizers (optimizers) on training performance. The optimizer mainly comprises Adam, RMSProp and SGD (SGD is set to 0.001). There are more local minima in the non-convex signal dataset. Compared with the other two optimizers, Adam can jump out the local minimum more effectively, find the global minimum and finally obtain a better modulation identification result.
Fig. 9 can obtain different recognition effects by setting different layer numbers (layers). When the SNR is greater than 0, the recognition effect of each layer is basically the same. Although more layers may better learn the distribution characteristics of the data set, it can be seen that the effect does not continue to improve when the number of layers reaches a certain number. Comparing 166 layers with 199 layers, the increased number of layers does not result in significant improvement in recognition, and excessive number of layers with higher complexity also results in a reduction in training speed. When the SNR is less than 0, the layer is 166, so that better recognition effect can be obtained.
Secondly, in the underwater communication process, the Doppler effect and the multipath effect are main problems influencing the communication effect. Thus, the validity of the designed network can be verified under these two different conditions.
The effect of doppler effect (doppler) recognition in two typical underwater acoustic communication parameters. It can be seen that the designed network has good adaptability to the doppler effect. Under two typical Doppler effects of underwater acoustic communication, the difference of the recognition rates is small, and the fact that the network has good robustness to the Doppler effect is proved.
In the case of SNR > -10dB, the network identification performance is almost the same in the case of various multipath interferences. In the case of low SNR (SNR < -10dB), 18 paths are identified slightly better than the other two paths, but with little difference. The result shows that the designed network has stronger capability of resisting multipath interference in underwater communication modulation identification.
And thirdly, comparing a common neural network method, the reasonability of the innovative network architecture of the application can be proved. The main contrast networks include Deep Neural Networks (DNN), Multi-Layer perceptrons (MLP), Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM), and the networks designed herein are denoted by MuCoModel. The various neural network approaches do not differ much when the SNR is < -18 dB. Starting with SNR > -15dB, MuCoModel has significant recognition advantages over other neural network approaches.
It can be known from the foregoing embodiment that, the embodiment provides an underwater acoustic signal type identification method based on a deep dense connection neural network, and the deep dense connection neural network overcomes the problem that a cross-layer connection method is easier to fit a deeper model. Through the cross-layer connection, the overfitting phenomenon can be prevented, the generalization capability of the model is improved, the model is deepened, and the classification precision is improved.
In correspondence with the method for identifying the type of an underwater acoustic signal provided in the foregoing embodiment, the present application also provides an embodiment of an underwater acoustic signal type identification system, for example, and referring to fig. 10, the system for identifying the type of an underwater acoustic signal 20 includes: a preprocessing module 201, a training module 202 and a recognition output module 203.
The preprocessing module 201 is configured to perform data preprocessing on multiple tuning signals transmitted by underwater communication to obtain an input signal type original data set.
The training module 202 is configured to determine a deep dense connection neural network for processing the input signal type raw data set. Specifically, the number of layers and the order of each layer of the deep-dense connection neural network and the network structure included in each layer of the neural network are determined.
The deep dense connection neural network comprises, from a network input to a network output: the device comprises an input layer, a pretreatment layer, a middle dense connection layer and an output layer. And the input layer converts the input signal type original data set into an input data set which is input into a network for further learning through preset data adjustment. The preprocessing layer includes a first Conv layer for extracting data features of an input data set and a first BN layer for normalizing data set distribution during each forward propagation. The intermediate dense connection layer comprises a second BN layer, a ReLU layer and a second Conv layer, the ReLU layer is a nonlinear activation function layer, and the ReLU function form is as follows:
Figure BDA0002700883800000131
after the ReLU layer is activated nonlinearly, the closer to x, the more relevant the characteristics are; the closer to 0, the smaller the correlation; in feature extraction, irrelevant data is directly discarded to reduce the amount of data. The output layer is a softmax activation function layer:
Figure BDA0002700883800000141
where i represents one of the various categories and Q represents the total number of all categories.
In this embodiment, in the training process of the deep dense connection neural network, each layer in the network extracts only part of the data features, and discarding part of the network layers randomly does not affect convergence of the network.
The identification output module 203 is configured to identify the type of the output underwater acoustic signal through the deep dense connection neural network.
The present application also provides an apparatus, see fig. 11, the apparatus 30 comprising: a processor 301, a memory 302, and a communication interface 303.
In fig. 11, a processor 301, a memory 302, and a communication interface 303 may be connected to each other by a bus; the bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown in FIG. 11, but this is not intended to represent only one bus or type of bus.
The processor 301 generally controls the overall functions of the device 30, for example, starting the device 30, and performing data preprocessing on various tuning signals transmitted by underwater communication after the device is started to obtain an original data set of the input signal type; determining a deep dense connection neural network for processing the input signal kind original data set; and identifying the type of the output underwater sound signal through the deep dense connection neural network.
Further, the processor 301 may be a general-purpose processor, such as a Central Processing Unit (CPU), a Network Processor (NP), or a combination of a CPU and an NP. The processor may also be a Microprocessor (MCU). The processor may also include a hardware chip. The hardware chips may be Application Specific Integrated Circuits (ASICs), Programmable Logic Devices (PLDs), or a combination thereof. The PLD may be a Complex Programmable Logic Device (CPLD), a Field Programmable Gate Array (FPGA), or the like.
Memory 302 is configured to store computer-executable instructions to support the operation of device 30 data. The memory 301 may be implemented by any type or combination of volatile or non-volatile memory devices, such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disks.
After the device 30 is started, the processor 301 and the memory 302 are powered on, and the processor 301 reads and executes the computer executable instructions stored in the memory 302 to complete all or part of the steps in the above-mentioned underwater acoustic signal type identification method embodiment based on the deep dense connection neural network.
The communication interface 303 is used for the device 30 to transfer data, for example, to enable data communication with a user. The communication interface 303 includes a wired communication interface, and may also include a wireless communication interface. The wired communication interface comprises a USB interface, a Micro USB interface and an Ethernet interface. The wireless communication interface may be a WLAN interface, a cellular network communication interface, a combination thereof, or the like.
In an exemplary embodiment, the device 30 provided by embodiments of the present application further includes a power supply component that provides power to the various components of the device 30. The power components may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power for device 30.
A communications component configured to facilitate communications between device 30 and other devices in a wired or wireless manner. The device 30 may access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. The communication component receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel. The communication component also includes a Near Field Communication (NFC) module to facilitate short-range communications. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID) technology, infrared data association (IrDA) technology, Ultra Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.
The same and similar parts among the various embodiments in the specification of the present application may be referred to each other. In particular, for the system and apparatus embodiments, since the method therein is substantially similar to the method embodiments, the description is relatively simple, and reference may be made to the description in the method embodiments for relevant points.
It is noted that, in this document, relational terms such as "first" and "second," and the like, may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
Of course, the above description is not limited to the above examples, and technical features that are not described in this application may be implemented by or using the prior art, and are not described herein again; the above embodiments and drawings are only for illustrating the technical solutions of the present application and not for limiting the present application, and the present application is only described in detail with reference to the preferred embodiments instead, it should be understood by those skilled in the art that changes, modifications, additions or substitutions within the spirit and scope of the present application may be made by those skilled in the art without departing from the spirit of the present application, and the scope of the claims of the present application should also be covered.

Claims (10)

1. A method for identifying the type of an underwater acoustic signal is based on a deep dense connection neural network, and is characterized by comprising the following steps:
preprocessing various tuning signals transmitted by underwater communication to obtain an input signal type original data set;
determining a deep dense connection neural network for processing the input signal kind original data set;
and identifying the type of the output underwater sound signal through the deep dense connection neural network.
2. The method according to claim 1, wherein the determining a deep dense connection neural network for processing the raw data set of the input signal class comprises:
determining the number of layers and the sequence of each layer of the deep dense connection neural network;
the network structure included in each layer of neural network is determined.
3. The underwater acoustic signal kind identification method according to claim 2, wherein the deep-dense connection neural network comprises, from a network input to a network output: the device comprises an input layer, a pretreatment layer, a middle dense connection layer and an output layer.
4. The method as claimed in claim 3, wherein the input layer converts the raw data set of the input signal category into an input data set for further learning by the input network through a preset data adjustment.
5. The underwater acoustic signal kind identification method according to claim 3, wherein the preprocessing layer includes a first Conv layer for extracting data characteristics of the input data set and a first BN layer for normalizing data set distribution during each forward propagation.
6. The method according to claim 3, wherein the intermediate dense connection layer comprises a second BN layer, a ReLU layer and a second Conv layer, the ReLU layer is a nonlinear activation function layer, and the ReLU function is of the form:
Figure FDA0002700883790000021
after the ReLU layer is activated nonlinearly, the closer to x, the more relevant the characteristics are; the closer to 0, the smaller the correlation; in feature extraction, irrelevant data is directly discarded to reduce the amount of data.
7. The underwater acoustic signal kind identification method according to claim 3, wherein the output layer is a softmax activation functional layer:
Figure FDA0002700883790000022
where i represents one of the various categories and Q represents the total number of all categories.
8. The method for identifying the type of the underwater acoustic signal according to claim 1, wherein in the training process of the deep dense connection neural network, each layer in the network only extracts partial data features, and randomly discarding partial network layers does not affect convergence of the network.
9. An underwater acoustic signal type identification system based on a deep dense connection neural network, the system comprising:
the preprocessing module is used for preprocessing data of various tuning signals transmitted by underwater communication to obtain an input signal type original data set;
the training module is used for determining a deep dense connection neural network for processing the input signal type original data set;
and the identification output module is used for identifying the type of the output underwater sound signal through the deep dense connection neural network.
10. An apparatus, comprising:
a processor;
a memory for storing computer executable instructions;
when the processor executes the computer-executable instructions, the processor executes the underwater sound signal class identification method according to any one of claims 1 to 8 to identify the class of the underwater sound signal.
CN202011021846.5A 2020-09-25 2020-09-25 Underwater acoustic signal type identification method, system and equipment Pending CN112329523A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011021846.5A CN112329523A (en) 2020-09-25 2020-09-25 Underwater acoustic signal type identification method, system and equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011021846.5A CN112329523A (en) 2020-09-25 2020-09-25 Underwater acoustic signal type identification method, system and equipment

Publications (1)

Publication Number Publication Date
CN112329523A true CN112329523A (en) 2021-02-05

Family

ID=74304031

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011021846.5A Pending CN112329523A (en) 2020-09-25 2020-09-25 Underwater acoustic signal type identification method, system and equipment

Country Status (1)

Country Link
CN (1) CN112329523A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112836674A (en) * 2021-02-28 2021-05-25 西北工业大学 Underwater target identification method based on micro Doppler characteristics
CN117131416A (en) * 2023-08-21 2023-11-28 四川轻化工大学 Small sample modulation identification method, system, electronic equipment and storage medium

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6718316B1 (en) * 2000-10-04 2004-04-06 The United States Of America As Represented By The Secretary Of The Navy Neural network noise anomaly recognition system and method
CN109299697A (en) * 2018-09-30 2019-02-01 泰山学院 Deep neural network system and method based on underwater sound communication Modulation Mode Recognition
CN109462564A (en) * 2018-11-16 2019-03-12 泰山学院 Subsurface communication Modulation Mode Recognition method and system based on deep neural network
CN111431825A (en) * 2020-02-25 2020-07-17 泰山学院 Signal automatic classification and identification method based on deep multi-flow neural network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6718316B1 (en) * 2000-10-04 2004-04-06 The United States Of America As Represented By The Secretary Of The Navy Neural network noise anomaly recognition system and method
CN109299697A (en) * 2018-09-30 2019-02-01 泰山学院 Deep neural network system and method based on underwater sound communication Modulation Mode Recognition
CN109462564A (en) * 2018-11-16 2019-03-12 泰山学院 Subsurface communication Modulation Mode Recognition method and system based on deep neural network
CN111431825A (en) * 2020-02-25 2020-07-17 泰山学院 Signal automatic classification and identification method based on deep multi-flow neural network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
GAO HUANG等: "Densely Connected Convolutional Networks", 《ARXIV》 *
叶凯: "基于机器学习的水声信号识别技术研究", 《中国优秀硕士学位论文全文数据库 基础科学辑》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112836674A (en) * 2021-02-28 2021-05-25 西北工业大学 Underwater target identification method based on micro Doppler characteristics
CN112836674B (en) * 2021-02-28 2024-03-26 西北工业大学 Underwater target identification method based on micro Doppler characteristics
CN117131416A (en) * 2023-08-21 2023-11-28 四川轻化工大学 Small sample modulation identification method, system, electronic equipment and storage medium
CN117131416B (en) * 2023-08-21 2024-06-04 四川轻化工大学 Small sample modulation identification method, system, electronic equipment and storage medium

Similar Documents

Publication Publication Date Title
Savitha et al. Projection-based fast learning fully complex-valued relaxation neural network
CN109361635B (en) Underwater communication modulation mode identification method and system based on depth residual error network
US20210027161A1 (en) Learning in communication systems
CN112887239B (en) Method for rapidly and accurately identifying underwater sound signal modulation mode based on deep hybrid neural network
CN112329523A (en) Underwater acoustic signal type identification method, system and equipment
CN112836569B (en) Underwater acoustic communication signal identification method, system and equipment based on sequence convolution network
Mossad et al. Deep convolutional neural network with multi-task learning scheme for modulations recognition
Wang et al. Adoption of hybrid time series neural network in the underwater acoustic signal modulation identification
CN109462564B (en) Underwater communication modulation mode identification method and system based on deep neural network
CN114143040A (en) Confrontation signal detection method based on multi-channel feature reconstruction
CN109547374B (en) Depth residual error network and system for underwater communication modulation recognition
Huang et al. Adaptive modulation and coding in underwater acoustic communications: a machine learning perspective
Peyvandi et al. SONAR systems and underwater signal processing: classic and modern approaches
Onasami et al. Underwater acoustic communication channel modeling using reservoir computing
Su et al. Deep non-cooperative spectrum sensing over Rayleigh fading channel
Zhang et al. Deep learning based underwater acoustic OFDM receiver with joint channel estimation and signal detection
CN114595729A (en) Communication signal modulation identification method based on residual error neural network and meta-learning fusion
Cai et al. An integrated incremental self-organizing map and hierarchical neural network approach for cognitive radio learning
CN112329524A (en) Signal classification and identification method, system and equipment based on deep time sequence neural network
CN112348165A (en) Underwater acoustic communication signal classification and identification method and system based on hybrid cycle network
CN112257648A (en) Signal classification and identification method based on improved recurrent neural network
KR102262392B1 (en) Method and apparatus of massive mimo detection based on deep neural network
CN116132235B (en) Continuous phase modulation signal demodulation method based on deep learning
Gao et al. Supervised Contrastive Learning‐Based Modulation Classification of Underwater Acoustic Communication
Zolotukhin et al. On Assessing Vulnerabilities of the 5G Networks to Adversarial Examples

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20210205

RJ01 Rejection of invention patent application after publication