CN116647376B - Voiceprint information-based underwater acoustic network node identity authentication method - Google Patents
Voiceprint information-based underwater acoustic network node identity authentication method Download PDFInfo
- Publication number
- CN116647376B CN116647376B CN202310603592.5A CN202310603592A CN116647376B CN 116647376 B CN116647376 B CN 116647376B CN 202310603592 A CN202310603592 A CN 202310603592A CN 116647376 B CN116647376 B CN 116647376B
- Authority
- CN
- China
- Prior art keywords
- network
- identity authentication
- node
- underwater
- model
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000012549 training Methods 0.000 claims abstract description 40
- 238000004891 communication Methods 0.000 claims abstract description 32
- 238000012360 testing method Methods 0.000 claims abstract description 21
- 238000013528 artificial neural network Methods 0.000 claims abstract description 20
- 230000008569 process Effects 0.000 claims abstract description 9
- 238000009432 framing Methods 0.000 claims abstract description 7
- 238000013526 transfer learning Methods 0.000 claims abstract description 5
- 238000002372 labelling Methods 0.000 claims abstract description 4
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 238000013138 pruning Methods 0.000 claims description 18
- 230000006870 function Effects 0.000 claims description 14
- 238000010606 normalization Methods 0.000 claims description 10
- 238000012795 verification Methods 0.000 claims description 10
- 241000854291 Dianthus carthusianorum Species 0.000 claims description 9
- 230000008859 change Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 5
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 4
- 230000008014 freezing Effects 0.000 claims description 3
- 238000007710 freezing Methods 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 abstract description 5
- 238000000605 extraction Methods 0.000 description 6
- 230000005540 biological transmission Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000013145 classification model Methods 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000037433 frameshift Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L63/00—Network architectures or network communication protocols for network security
- H04L63/08—Network architectures or network communication protocols for network security for authentication of entities
- H04L63/0861—Network architectures or network communication protocols for network security for authentication of entities using biometrical features, e.g. fingerprint, retina-scan
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/096—Transfer learning
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B13/00—Transmission systems characterised by the medium used for transmission, not provided for in groups H04B3/00 - H04B11/00
- H04B13/02—Transmission systems in which the medium consists of the earth or a large mass of water thereon, e.g. earth telegraphy
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/16—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L9/00—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols
- H04L9/40—Network security protocols
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Computer Security & Cryptography (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Computer Hardware Design (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses a voiceprint information-based method for authenticating the identity of a node of a voiceprint network, and belongs to the technical field of the security of the voiceprint network. Arranging relevant nodes requiring identity authentication in the underwater acoustic network system to finish signal acquisition and labeling work of all nodes in the underwater acoustic network; collecting collected acoustic signal data of the underwater network node, and constructing a training, verifying and testing data set; preprocessing the collected original acoustic signal data; extracting characteristics of the samples after framing, and taking the extracted characteristics as input of a neural network; training a neural network, and realizing identity mutual recognition through voiceprint information; optimizing a neural network structure, and constructing a lightweight avatar authentication and identification network; and updating and fine-tuning the network model by a transfer learning method when the underwater node is in a missing state or a new member state. The invention realizes the identity authentication of the nodes in the whole process of network communication under the condition that the existing underwater acoustic network communication protocol, communication efficiency and even safety protection mechanism are not affected.
Description
Technical Field
The invention relates to the technical field of underwater acoustic network security, in particular to an underwater acoustic network node identity authentication method based on voiceprint information.
Background
The underwater acoustic network is a communication network formed by using acoustic waves as a transmission carrier by various underwater communication devices, and can be used for realizing tasks such as large-area real-time monitoring of marine environment, long-distance cooperative operation of underwater vehicles and the like. The sound wave has the advantage of long-distance transmission under water, but is also easy to be influenced by complex underwater environment, so that the problems of low data transmission rate, high bit error rate, high transmission delay and the like are caused.
The underwater acoustic network is different from the land wireless communication network, and the network environment is relatively public, the communication bandwidth is narrow, and the data throughput rate is low, so that the implementation of information safety protection is difficult. The current underwater acoustic network security research is mainly oriented to a link layer, a network layer and an application layer, and network security protection is realized through mechanisms such as identity authentication, data integrity check and encryption based on cryptography. However, access authentication, message encryption, key management, certificate issuing and other security protocols all need multiple handshakes among nodes to be realized, and because the underwater acoustic network has the characteristics of limited energy sources, low channel quality, changeable network topology and the like, the underwater acoustic network security protection requirements reduce unnecessary service call, do not interfere with the original communication efficiency, do not obviously increase the energy consumption of network nodes and the like.
Aiming at the current situations of low speed, high error code and long delay faced by the current underwater acoustic network communication, in order to realize the most basic identity authentication safety protection capability under the defect network so as to rapidly identify the underwater network node with legal communication authority, the existing identity authentication method based on cryptography needs to be supplemented, and the identity authentication function of the node is provided in the whole process of network communication by extracting and identifying the physical characteristics of the underwater acoustic communication equipment in real time under the condition of completely not affecting the existing underwater acoustic network communication protocol, communication efficiency and even safety protection mechanism.
Disclosure of Invention
In view of the above, the invention provides a method for authenticating the identity of underwater network nodes based on voiceprint information, and sound waves emitted by different underwater network nodes contain equipment fingerprint information and are irrelevant to the type of signals emitted, and legal identities of the underwater network nodes can be distinguished by extracting and identifying voiceprint characteristics of the sound waves, so that an identity authentication function among the nodes is provided for an underwater network system.
The technical scheme of the invention is as follows:
a voiceprint information-based underwater acoustic network node identity authentication method comprises the following steps:
step 1, arranging relevant nodes requiring identity authentication in a hydroacoustic network system under water, setting each node to emit a plurality of groups of acoustic signals, acquiring acoustic signals of a single node by data acquisition equipment to avoid interference among the plurality of nodes, recording the identity numbers of the nodes, sequentially completing signal acquisition and labeling work of all the nodes in the hydroacoustic network, collecting signals of each group for about a plurality of minutes, and meeting acquisition requirements;
step 2, collecting the collected acoustic signal data of the underwater network node, classifying the data according to the labeled node identity number information, and constructing a training, verifying and testing data set;
step 3, preprocessing the collected original acoustic signal data, framing the samples in the step 2 according to 5 seconds as a frame, normalizing the samples to eliminate the difference of data among the samples, wherein each frame of normalized data is one sample, and then each category acquires a plurality of sample data;
step 4, extracting features of the samples after framing, wherein available features comprise MFCC (Mel frequency cepstral coefficients) and GFCC (Gammatone frequency cepstral coefficients) features, extracting first-order differences and second-order differences of the MFCC and the GFCC, and selecting one or more combined features as input of the neural network;
step 5, training a neural network by utilizing all the acoustic signal characteristics obtained in the step 4, training an identity authentication and identification network model according to the network topology structure of the underwater node, forming a group between the cluster head node and the cluster members, wherein the cluster head node and the members in the group can realize identity mutual identification through voiceprint information, and compared with the multiple communication decoding process based on the cryptography identity authentication, the multiple communication decoding process based on the cryptography identity authentication is avoided, and the identity authentication is carried out only through the voiceprint information transmitted by the equipment;
step 6, optimizing a neural network structure, constructing a lightweight identity authentication identification network, guaranteeing real-time processing of an embedded processor end, and further improving timeliness of the underwater node identity authentication;
and 7, transplanting the network model and the feature extraction algorithm to embedded processing equipment of the node, rapidly identifying identity information of the corresponding node by receiving voiceprint information of the corresponding node, and updating and fine-tuning the network model by a transfer learning method to realize updating of the identity authentication model if the underwater node is in a missing state or a new member state and the small-range node is changed.
In step 1, the underwater acoustic network system is a communication network formed by underwater submerged buoy, unmanned underwater vehicle, underwater base station, unmanned surface vessel and other marine equipment equipped with underwater acoustic communication machine, and the nodes exchange data through the underwater acoustic network. The acquired signals are classified according to groups according to the topological structure of the underwater acoustic network, then a plurality of groups of network authentication models are trained, and each node in each group of network can realize mutual authentication. The multiple groups of acoustic signals transmitted by the nodes are used for identifying identity information when the node transmits different acoustic signals by the trained neural network, the acoustic signals of the training set samples can be completely different from the types of the acoustic signals of the verification set and the test set, the acoustic signals transmitted by the nodes are not limited, and can be any type of signals, such as underwater acoustic modulation signals, simulated radiation noise signals and the like, because the fingerprint information of the equipment is only related to the physical characteristics of the equipment, and is irrelevant to the type of the transmitted acoustic signals.
Further, in step 4, the feature extraction method includes a common voiceprint information feature extraction method, such as MFCC, GFCC, etc., where the MFCC and GFCC obtain static features of the signal, and in order to obtain dynamic change conditions of the signal, first-order difference features and second-order difference features of the signal are extracted in a complementary manner.
Further, the step 5 includes the steps of:
and 5.1, constructing an identity authentication model, taking a residual network as a pre-training model, and taking a residual module as a bottleneck residual module, wherein the residual module is sequentially formed by three convolution layers of 1×1, 3×3 and 1×1. Each convolution layer in the residual module is added with a Batchnormalization normalization layer and a LeakyReLU activation function layer. The neural network sequentially comprises a convolution layer, a Batchnormalization normalization layer, a relu activation function layer, a maximum pooling layer, three residual error module layers, an average pooling layer, a Flatten layer, a full connection layer, a Batchnormalization normalization layer and an output layer;
and 5.2, training a network model, wherein the training set and the verification set are used for optimizing main parameters of the network model, the model is input into one or more combination features extracted in the step 4, the output is node identity information, multiple groups of network models can be trained according to different input features, and the model can better understand data due to the fact that the selection of the features is very important to the performance of the classification model, and the accuracy of prediction is improved by reasonably selecting the features.
And 5.3, testing the network model, wherein the test set data, the training set data and the verification set data are acoustic signals of different types, and selecting an identity authentication model according to the accuracy of the identity authentication of the test set data, wherein the model with higher accuracy is used as a final identity authentication model.
In step 6, the optimized network structure performs regularized channel pruning processing on the network, reduces complexity of the model, and performs fine tuning on the pruned model to complete parameter updating, so that better balance between performance and scale of the model is ensured.
Further, step 6 includes the steps of:
step 6.1, training a node identity authentication network model to obtain parameters of each layer;
and 6.2, based on an original network model, pruning some unimportant channels by a regularized channel pruning method, wherein the full connection layer is regarded as a convolution channel with the space size of 1 multiplied by 1. Multiplying the output of each channel by a scaling factor gamma to readjust the model parameters so that the overall performance of the pruned model is less in change, then carrying out sparse regularization on the scaling factor gamma, setting a global pruning threshold value as th, and if the absolute value of the scaling factor is smaller than a preset threshold value, enabling gamma=0 to be pruned by the corresponding channel and setting the corresponding weight parameter as zero so as to realize pruning of unimportant channels;
and 6.3, retraining the network model after pruning of the neural network, wherein the identification performance of the identity authentication network is slightly reduced after pruning of the model, and the retrained network model realizes optimization of the network model by updating parameters.
Further, in step 7, the updating and fine tuning of the network model by the method of transfer learning when the underwater node has a missing or a new member includes the following steps:
step 7.1, collecting a plurality of minutes of acoustic signals of each underwater node in the identity authentication network of the missing or newly added member, and constructing a training set and a testing set after meeting the construction requirement;
step 7.2, because the change of the members in the group is smaller, the characteristics extracted by the model are still effective in the new node network, and parameters after the full-connection layer, including the number of neurons of the full-connection layer and the output layer, can be updated only by utilizing newly collected training set data through freezing all network layer weight parameters before the full-connection layer in the network model so as to realize the optimization of the identity authentication network;
and 7.3, evaluating the performance of the new identity authentication network model through testing the accuracy of the set, if the accuracy meets the requirement, storing an updated model, otherwise, adjusting parameters such as a model loss function, an optimizer and the like to improve the network performance.
The effective benefits of the invention are as follows:
1. in order to improve the recognition rate of a neural network model and adapt to dynamically-changed underwater node identity authentication, when an underwater node is in a missing state or a new member, the network model is updated and fine-tuned by a transfer learning method, and the parameters of a feature extraction layer in the network are fixed, so that the network does not participate in subsequent model training, and only the parameters of a final full-connection layer and an output layer are updated. The identity authentication and identification network model is trained according to the network topology structure of the underwater node, for example, a group is formed between the cluster head node and the cluster member, and the cluster head node and the members in the group can realize mutual identity authentication through voiceprint information.
2. Under the condition that the existing underwater acoustic network communication protocol, communication efficiency and even safety protection mechanism are not affected, the invention realizes the function of providing identity authentication of the node in the whole process of network communication by extracting and identifying the physical characteristics of the underwater acoustic communication equipment in real time.
Drawings
FIG. 1 is a schematic flow chart diagram of an implementation of the underwater acoustic network node identity authentication method of the present invention;
FIG. 2 is a schematic diagram of an underwater acoustic network of the method for authentication of an underwater acoustic network node according to the present invention;
fig. 3 is a schematic diagram of a training packet of a network model of the underwater acoustic network node identity authentication method of the present invention.
Detailed Description
For the purpose of making the objects and technical solutions of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings and examples, it being understood that the specific examples described herein are for the purpose of illustration only and are not intended to limit the present invention.
In this embodiment, the method for authenticating the identity of the underwater acoustic network node based on the voiceprint information, as shown in fig. 1, includes the following steps:
step 1, arranging relevant nodes requiring identity authentication in a hydroacoustic network system under water, setting each node to transmit a plurality of groups of acoustic signals, acquiring acoustic signals of a single node by data acquisition equipment to avoid interference among the plurality of nodes, recording the identity numbers of the nodes, and sequentially completing signal acquisition and labeling work of all the nodes in the hydroacoustic network, wherein each group of signals is collected for about 10 minutes.
In this embodiment, the underwater acoustic network system described in step 1 is a communication network composed of an underwater submerged buoy, a buoy, an unmanned underwater vehicle, a submarine base station, an unmanned surface vessel and a land data center, as shown in fig. 2, and the underwater nodes exchange data through the underwater network. The sampling rate of the acoustic signal transmitted by the node is set to 22050, the type of signal is not limited, and the signal can be any form of signal, such as a hydroacoustic modulation signal, an analog radiation noise signal, and the like, because the fingerprint information of the device is only related to the physical characteristics of the device itself, and is not related to what acoustic signal is transmitted.
And 2, collecting the collected acoustic signal data of the underwater network node, classifying the data according to the marked node identity number information, and constructing a training, verifying and testing data set, wherein the acoustic signal types of the training set sample, the verification set and the testing set can be completely different.
The collected signals are classified according to the underwater acoustic network topology structure and are used for training a plurality of groups of network authentication models, white nodes in the elliptical areas represent cluster head nodes, gray nodes represent members in the groups, the cluster head nodes in each area can carry out identity authentication with the members in the groups, and the identity authentication among the areas can be authenticated through the models trained by the cluster head nodes in each area.
And 3, preprocessing the acquired original acoustic signal data, framing the samples in the step 2 according to 5 seconds as one frame, normalizing the samples to eliminate the difference of data among the samples, wherein each frame of normalized data is one sample, and then each category acquires a plurality of sample data.
And 4, extracting features of the samples after framing, wherein available features comprise MFCC (Mel frequency cepstral coefficients) and GFCC (Gammatone frequency cepstral coefficients) features, extracting first-order differences and second-order differences of the MFCC and the GFCC, and selecting one or more combined features as input of the neural network.
When the MFCC, GFCC and the first-order difference and second-order difference characteristics thereof are calculated, the FFT window size is set to 2048, the frame shift is 512, the MFCC and GFCC return number is set to 40, and the signal characteristic size is obtained (40, 216).
And 5, training the neural network by utilizing all the acoustic signal characteristics of the category obtained in the step 4, realizing the identification of each node identity, avoiding the repeated communication decoding process of the password identity verification compared with the password-based identity verification, and carrying out the identity verification only by the equipment voiceprint information.
Said step 5 comprises the steps of:
and 5.1, constructing an identity authentication model, taking a residual network as a pre-training model, and taking a residual module as a bottleneck residual module, wherein the residual module is sequentially formed by three convolution layers of 1×1, 3×3 and 1×1. Each convolution layer in the residual module is added with a Batchnormalization normalization layer and a LeakyReLU activation function layer. The neural network sequentially comprises a convolution layer, a Batchnormalization normalization layer, a relu activation function layer, a maximum pooling layer, three residual error module layers, an average pooling layer, a Flatten layer, a full connection layer, a Batchnormalization normalization layer and an output layer;
step 5.2, training a network model, wherein the training set and the verification set are used for optimizing main parameters of the network model, the model is input into one or more combination features extracted in the step 4, the output is node identity information, multiple groups of network models can be trained according to different input features, and the model can better understand data due to the fact that the selection of the features is very important to the performance of the classification model, and the accuracy of prediction is improved due to the fact that the reasonable selection of the features;
and 5.3, testing the network model, and selecting an identity authentication model according to the accuracy of the data identity authentication of the test set, wherein the model with higher accuracy is used as a final identity authentication model.
In this embodiment, in order to find the optimal neural network input, a neural network model is trained with multiple combinations of MFCCs, GFCCs and their first and second order differential features, including individual MFCC features, individual GFCC features, MFCCs and their first order differential features, MFCCs and their first and second order differential features, GFCCs and their first and second order differential features, and hybrid features of MFCCs and GFCCs. A neural network model with MFCC and its first and second order differential features as inputs is selected according to test set accuracy, the model input size is (40, 216,3), and the output is node identity number.
And 6, optimizing a neural network structure and constructing a lightweight avatar authentication and identification network.
And 6.1, training the identity authentication network model in the step 5, and obtaining parameters of each layer of the model.
Step 6.2, based on the original network model, pruning some unimportant channels by a regularized channel pruning method, wherein the full-connection layer is regarded as a convolution channel with a space size of 1×1, each channel outputs a scaling factor gamma to readjust model parameters so as to enable the overall performance of the pruned model to change less, then sparse regularized is carried out on the scaling factor gamma, a global pruning threshold value is set as th, if the absolute value of the scaling factor is smaller than a preset threshold value, gamma=0, namely the corresponding channel is pruned, and the corresponding weight parameter is set as zero, so that pruning of the unimportant channel is realized.
The BN layer is utilized in batch normalization to effectively identify and prune non-important channels in the network, and the BN layer on each miniband performs the following transformations during training:
wherein μ and σ are average and standard deviation values input on miniband, ε is an extremely small constant for avoiding the case that denominator is 0, γ and β are trainable superparameters, and channel sparsity is performed by regularization, i.e., the γ value is thinned by using L1 norm, and the loss of network training is:
where (x, y) represents the training input and the label, w represents the trainable weight, λ is the penalty factor, λ=0.001 in this embodiment,corresponding to the training loss of the network model, the embodiment enablesA cross entropy loss function is used.
And 6.3, retraining the network model after pruning of the neural network, wherein the identification performance of the identity authentication network is slightly reduced after pruning of the model, and the retrained network model realizes optimization of the network model by updating parameters.
Step 7, the network model and the feature extraction algorithm are transplanted to the embedded processing equipment of the node, the identity information of the corresponding node is rapidly identified by receiving the voiceprint information of the corresponding node, and if the underwater node is in a missing state or a new member state, the network model is updated and fine-tuned by a migration learning method so as to realize the updating of the identity authentication model, and the method specifically comprises the following steps:
step 7.1, collecting 2-minute acoustic signals of each underwater node in the identity authentication network of the missing or newly added member, and constructing a training set and a testing set according to a ratio of 7:3;
step 7.2, because the change of the members in the group is small, the characteristics extracted by the model are still effective in the new node network, and parameters after the full-connection layer, including the number of neurons of the full-connection layer and the output layer, can be updated only by using newly collected training set data through freezing all network layer weight parameters before the full-connection layer of the network model so as to realize the optimization of the identity authentication network;
and 7.3, evaluating the performance of the new identity authentication network model through testing the accuracy of the set, if the accuracy meets the requirement, storing an updated model, otherwise, adjusting parameters such as a model loss function, an optimizer and the like to improve the network performance.
In summary, in the embodiment of the invention, the lightweight voiceprint information-based identity authentication network model is generated through the steps 1-7, and the network model and the feature extraction algorithm are transplanted to the embedded equipment of the underwater node, so that the identity of the relevant underwater node is authenticated in real time through the voiceprint information. The identity authentication method provided by the invention provides the identity authentication function of the node in the whole process of network communication by extracting and identifying physical characteristics of the underwater acoustic communication equipment in real time under the condition that the existing underwater acoustic network communication protocol, communication efficiency and even a safety protection mechanism are not affected.
Claims (4)
1. The underwater acoustic network node identity authentication method based on the voiceprint information is characterized by comprising the following steps of:
step 1, arranging relevant nodes requiring identity authentication in a hydroacoustic network system under water, setting each node to transmit a plurality of groups of acoustic signals, acquiring acoustic signals of a single node by data acquisition equipment to avoid interference among the plurality of nodes, recording the identity numbers of the nodes, and sequentially completing signal acquisition and labeling work of all the nodes in the hydroacoustic network, wherein each group of signals is collected for a plurality of minutes until the acquisition requirement is met;
step 2, collecting the collected acoustic signal data of the underwater network node, classifying the data according to the labeled node identity number information, and constructing a training, verifying and testing data set;
step 3, preprocessing the collected original acoustic signal data, framing the samples in the step 2 according to 5 seconds as a frame, normalizing the samples to eliminate the difference of data among the samples, wherein each frame of normalized data is one sample, and then each category acquires a plurality of sample data;
step 4, extracting features of the samples after framing, wherein the available features comprise MFCC and GFCC features, extracting first-order difference and second-order difference of the MFCC and GFCC, and selecting one or more combined features as input of a neural network;
step 5, training a neural network by utilizing all the acoustic signal characteristics obtained in the step 4, training an identity authentication and identification network model according to the network topology structure of the underwater node, forming a group between the cluster head node and the cluster members, wherein the cluster head node and the members in the group can realize identity mutual identification through voiceprint information, and compared with the multiple communication decoding process based on the cryptography identity authentication, the multiple communication decoding process based on the cryptography identity authentication is avoided, and the identity authentication is carried out only through the voiceprint information transmitted by the equipment;
step 5.1, constructing an identity authentication recognition network model, taking a residual network as a pre-training model, and taking a residual module as a bottleneck residual module which is sequentially formed by three convolution layers of 1 multiplied by 1, 3 multiplied by 3 and 1 multiplied by 1; adding a Batchnormalization normalization layer and a LeakyReLU activation function layer after each convolution layer in the residual module; the neural network sequentially comprises a convolution layer, a Batchnormalization normalization layer, a relu activation function layer, a maximum pooling layer, three residual error module layers, an average pooling layer, a Flatten layer, a full connection layer, a Batchnormalization normalization layer and an output layer;
step 5.2, training the identity authentication and identification network model
The training set and the verification set are used for optimizing main parameters of the identity authentication and identification network model, the model is input into one or more combination features extracted in the step 4, the output is node identity information, and a plurality of groups of network models are trained according to different input features;
step 5.3, identity authentication and identification network model test
The test set data, the training set and the verification set data are acoustic signals of different types, and the identity authentication recognition network model is selected according to the accuracy of the identity authentication of the test set data, and the model with higher accuracy is used as the final identity authentication recognition network model.
2. The underwater sound network node identity authentication method based on the voiceprint information according to claim 1, further comprising the steps of 6, optimizing a neural network structure, and constructing a lightweight avatar authentication and identification network; the method specifically comprises the following steps:
step 6.1, training a node identity authentication network model to obtain parameters of each layer;
step 6.2, based on the original identity authentication network model, pruning some unimportant channels by a regularized channel pruning method, wherein the full-connection layer is regarded as a convolution channel with the space size of 1 multiplied by 1; multiplying the output of each channel by a scaling factor gamma to readjust the model parameters so that the overall performance of the pruned model is less in change, then carrying out sparse regularization on the scaling factor gamma, setting a global pruning threshold value as th, and if the absolute value of the scaling factor is smaller than a preset threshold value, enabling gamma=0 to be pruned by the corresponding channel and setting the corresponding weight parameter as zero so as to realize pruning of unimportant channels;
and 6.3, retraining the identity authentication recognition network model after pruning of the neural network, and optimizing the identity authentication recognition network model by updating parameters through the trained identity authentication recognition network model.
3. The method for authenticating the identity of the underwater acoustic network node based on the voiceprint information according to claim 2, further comprising the step 7 of updating and fine-tuning a network model by a transfer learning method when a missing or new member occurs in the underwater node, specifically comprising the following steps:
step 7.1, collecting a plurality of clock signals of each underwater node in the identity authentication network of the missing or newly added member until the building requirement is met, and building a training set and a testing set;
step 7.2, because the change of the members in the group is smaller, the characteristics extracted by the model are still effective in the new node network, and parameters after the full-connection layer, including the number of neurons of the full-connection layer and the output layer, can be updated only by utilizing newly collected training set data through freezing all network layer weight parameters before the full-connection layer in the network model so as to realize the optimization of the identity authentication network;
and 7.3, evaluating the performance of the new identity authentication network model through testing the accuracy of the set, if the accuracy meets the requirement, storing an updated model, otherwise, adjusting the model loss function and the optimizer parameters to improve the network performance.
4. A method for authenticating the identity of a node of a underwater acoustic network based on voiceprint information according to any one of claims 1 to 3, wherein in the step 1, the underwater acoustic network system is a communication network composed of underwater submerged buoy, unmanned underwater vehicle, underwater base station, and marine equipment for configuring underwater acoustic communication machine by unmanned surface vessel, and data are exchanged between nodes through the underwater acoustic network; collecting signals according to the underwater acoustic network topology structure, classifying the signals according to groups, then training a plurality of groups of identity authentication recognition network models, wherein each node in each group of network can realize mutual authentication; the acoustic signal transmitted by the node is not limited.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310603592.5A CN116647376B (en) | 2023-05-25 | 2023-05-25 | Voiceprint information-based underwater acoustic network node identity authentication method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310603592.5A CN116647376B (en) | 2023-05-25 | 2023-05-25 | Voiceprint information-based underwater acoustic network node identity authentication method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116647376A CN116647376A (en) | 2023-08-25 |
| CN116647376B true CN116647376B (en) | 2024-01-26 |
Family
ID=87639306
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310603592.5A Active CN116647376B (en) | 2023-05-25 | 2023-05-25 | Voiceprint information-based underwater acoustic network node identity authentication method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116647376B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110245608A (en) * | 2019-06-14 | 2019-09-17 | 西北工业大学 | A kind of Underwater targets recognition based on semi-tensor product neural network |
| CN111723679A (en) * | 2020-05-27 | 2020-09-29 | 上海五零盛同信息科技有限公司 | Face and voiceprint authentication system and method based on deep migration learning |
| CN112257521A (en) * | 2020-09-30 | 2021-01-22 | 中国人民解放军军事科学院国防科技创新研究院 | CNN underwater acoustic signal target identification method based on data enhancement and time-frequency separation |
| CN113129897A (en) * | 2021-04-08 | 2021-07-16 | 杭州电子科技大学 | Voiceprint recognition method based on attention mechanism recurrent neural network |
| CN113191178A (en) * | 2020-12-04 | 2021-07-30 | 中国船舶重工集团公司第七一五研究所 | Underwater sound target identification method based on auditory perception feature deep learning |
| CN113435276A (en) * | 2021-06-16 | 2021-09-24 | 中国电子科技集团公司第五十四研究所 | Underwater sound target identification method based on antagonistic residual error network |
| WO2022252272A1 (en) * | 2021-06-03 | 2022-12-08 | 江苏大学 | Transfer learning-based method for improved vgg16 network pig identity recognition |
-
2023
- 2023-05-25 CN CN202310603592.5A patent/CN116647376B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110245608A (en) * | 2019-06-14 | 2019-09-17 | 西北工业大学 | A kind of Underwater targets recognition based on semi-tensor product neural network |
| CN111723679A (en) * | 2020-05-27 | 2020-09-29 | 上海五零盛同信息科技有限公司 | Face and voiceprint authentication system and method based on deep migration learning |
| CN112257521A (en) * | 2020-09-30 | 2021-01-22 | 中国人民解放军军事科学院国防科技创新研究院 | CNN underwater acoustic signal target identification method based on data enhancement and time-frequency separation |
| CN113191178A (en) * | 2020-12-04 | 2021-07-30 | 中国船舶重工集团公司第七一五研究所 | Underwater sound target identification method based on auditory perception feature deep learning |
| CN113129897A (en) * | 2021-04-08 | 2021-07-16 | 杭州电子科技大学 | Voiceprint recognition method based on attention mechanism recurrent neural network |
| WO2022252272A1 (en) * | 2021-06-03 | 2022-12-08 | 江苏大学 | Transfer learning-based method for improved vgg16 network pig identity recognition |
| CN113435276A (en) * | 2021-06-16 | 2021-09-24 | 中国电子科技集团公司第五十四研究所 | Underwater sound target identification method based on antagonistic residual error network |
Non-Patent Citations (1)
| Title |
|---|
| 深度迁移模型下的小样本声纹识别方法;孙存威;文畅;谢凯;贺建飚;;计算机工程与设计(12);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116647376A (en) | 2023-08-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110365612B (en) | Deep learning beam domain channel estimation method based on approximate message transfer algorithm | |
| CN109639739B (en) | Abnormal flow detection method based on automatic encoder network | |
| CN111783558A (en) | Satellite navigation interference signal type intelligent identification method and system | |
| CN110113288B (en) | Design and demodulation method of OFDM demodulator based on machine learning | |
| Zhang et al. | Modulation recognition of underwater acoustic signals using deep hybrid neural networks | |
| CN113259288B (en) | Underwater sound modulation mode identification method based on feature fusion and lightweight hybrid model | |
| CN109581339B (en) | Sonar identification method based on automatic adjustment self-coding network of brainstorming storm | |
| CN113381828B (en) | Sparse code multiple access random channel modeling method based on condition generation countermeasure network | |
| WO2021103206A1 (en) | Machine learning algorithm-based wireless radio frequency device identity recognition method and system | |
| CN111050315B (en) | Wireless transmitter identification method based on multi-core two-way network | |
| CN115146670A (en) | Radio frequency fingerprint identification method and system based on data enhancement and comparison learning | |
| Ding et al. | Data-and-knowledge dual-driven automatic modulation recognition for wireless communication networks | |
| CN115238749A (en) | Feature fusion modulation identification method based on Transformer | |
| CN114757224A (en) | Specific radiation source identification method based on continuous learning and combined feature extraction | |
| CN116647376B (en) | Voiceprint information-based underwater acoustic network node identity authentication method | |
| Hu et al. | Feature fusion convolution-aided transformer for automatic modulation recognition | |
| CN117560104A (en) | Construction method of interpretable machine learning-assisted channel model in mixed traffic | |
| Gai et al. | Spectrum sensing method based on residual cellular network | |
| CN112434716B (en) | Underwater target data amplification method and system based on condition countermeasure neural network | |
| CN114118151A (en) | Intelligent spectrum sensing method with environment adaptive capacity | |
| CN117692026B (en) | Link sensing method and device for power line communication | |
| CN117939467A (en) | Underwater identity mutual authentication method for AUV cluster | |
| Motwani et al. | Analysis of various neural network architectures for automatic modulation techniques | |
| Shi et al. | GAF-MAE: A Self-supervised Automatic Modulation Classification Method Based on Gramian Angular Field and Masked Autoencoder | |
| CN112529035B (en) | Intelligent identification method for identifying individual types of different radio stations |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |