CN112052880A - Underwater sound target identification method based on weight updating support vector machine - Google Patents

Underwater sound target identification method based on weight updating support vector machine Download PDF

Info

Publication number
CN112052880A
CN112052880A CN202010811948.0A CN202010811948A CN112052880A CN 112052880 A CN112052880 A CN 112052880A CN 202010811948 A CN202010811948 A CN 202010811948A CN 112052880 A CN112052880 A CN 112052880A
Authority
CN
China
Prior art keywords
support vector
weight
underwater sound
vector machine
sound target
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
CN202010811948.0A
Other languages
Chinese (zh)
Inventor
齐滨
梁国龙
郭少祥
张光普
王燕
付进
王逸林
邹男
王晋晋
邱龙皓
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.)
Harbin Engineering University
Original Assignee
Harbin Engineering 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 Harbin Engineering University filed Critical Harbin Engineering University
Priority to CN202010811948.0A priority Critical patent/CN112052880A/en
Publication of CN112052880A publication Critical patent/CN112052880A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
    • G01H3/00—Measuring characteristics of vibrations by using a detector in a fluid
    • 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
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)

Abstract

The invention provides an underwater sound target identification method based on a weight updating support vector machine, which comprises the steps of collecting sample data of two types of underwater sound targets, adding category labels, and establishing an underwater sound target sample library; performing MFCC processing on the underwater sound target sample library, and outputting dynamic characteristics; constructing a support vector machine recognition model with updated weight according to the optimal alpha value and the weight omega, and performing model cross validation and model evaluation on the recognition model by using the remaining 70% of dynamic characteristic samples; counting the identification accuracy of the support vector machine identification model updated by the constructed weight, and evaluating the performance of the support vector machine identification model; and identifying the real data by using the obtained underwater sound target identification model of the support vector machine based on the updated weight. Aiming at the characteristics of the underwater sound target, the invention reasonably provides the optimal Lagrange multiplier and weight by using the dynamic characteristics of the MFCC output underwater sound target, thereby greatly improving the identification accuracy.

Description

Underwater sound target identification method based on weight updating support vector machine
Technical Field
The invention belongs to the technical field of underwater sound target identification, and particularly relates to an underwater sound target identification method based on an updated weight support vector machine.
Background
With the vigorous development of hydroacoustic, hydroacoustic target recognition is widely learned and studied as a single disciplinary system. The underwater acoustic target recognition technology is an important component of underwater acoustic signal processing, is an important technical support for acquiring underwater acoustic information and resisting the underwater acoustic information, and is widely applied in the fields of anti-diving, torpedo defense, submarine topography exploration and the like. The underwater sound target identification is mainly based on the characteristic information of the target. The object characteristic information is information contained or extractable in the object raw data, which can accurately and simply indicate the state and identity of the object. Traditional underwater acoustic targets mainly include characteristic information such as noise, motion, wake, geometry, and the like. With the gradual maturity of speech signal processing technology, the characteristics of the underwater acoustic target become wide, and the typical mel-frequency spectrum coefficient is used to extract the characteristics of the underwater acoustic target. In recent years, the field of machine learning is rapidly developed, and various disciplines are widely applied, wherein a support vector machine which is characterized by small sample amount is very suitable for being applied to the field of underwater sound.
The diversity of marine environments makes it extremely difficult to acquire acoustic information of a target object, the performance of a neural network requiring a large sample size is greatly reduced, and a support vector machine is just a machine learning method for small sample learning. The support vector machine based on the updated weight value can not only overcome the problem of so-called sample size in this respect, but also improve the performance indexes such as the recognition rate of the traditional support vector machine.
Disclosure of Invention
The invention aims to provide an underwater sound target identification method based on a weight updating support vector machine, aiming at the problems of large detection data volume, low automation degree, low identification efficiency and the like during underwater sound target identification. The invention researches the application of machine learning in underwater acoustic target recognition. The network structure principle and the improved version of machine learning are analyzed, and the application situations of the machine learning in the field of underwater sound signal identification and the field of underwater sound image signal identification are explained respectively. The invention mainly combines the characteristics of two fields of voice signal processing and machine learning, thereby achieving the purpose of underwater sound target recognition. Firstly, voice signal processing is used as a main means of feature extraction, a Mel frequency spectrum coefficient (MFCC) can well extract underwater acoustic features, a Lagrange multiplier alpha and a weight omega are initialized, and the Lagrange multiplier alpha and the optimized weight omega are continuously updated by using a double-coordinate descent method, so that the working performance of a support vector machine is improved.
The invention is realized by the following technical scheme, and provides an underwater sound target identification method based on an updated weight support vector machine, which specifically comprises the following steps:
the method comprises the following steps: collecting two types of underwater sound target sample data, adding a category label, and establishing an underwater sound target sample library;
step two: performing MFCC processing on the underwater sound target sample library, and outputting dynamic characteristics;
step three: selecting a Gaussian radial basis kernel function as a kernel function, and giving a Lagrange multiplier alpha and a corresponding weight expression omega ═ Sigmaiαiyiφ(xi) N, n is the number of samples, y is 1,2iFor class labels, phi () represents a feature mapping function, xiIs sample data; using 30% of dynamic characteristics as training samples, and carrying out loop iteration on the training samples by using a double-coordinate descent method to obtain an optimal alpha value and a weight omega; constructing according to the optimal alpha value and weight omegaA support vector machine identification model for updating the weight value;
step four: performing model cross validation and model evaluation on the constructed support vector machine identification model with the weight value updated by using the remaining 70% of dynamic characteristic samples;
step five: counting the identification accuracy of the support vector machine identification model updated by the constructed weight, and evaluating the performance of the support vector machine identification model;
step six: and identifying the real data by using the obtained underwater sound target identification model of the support vector machine based on the updated weight.
Further, in the first step, the collected underwater sound target sample data are classified and marked as "+ 1" and "-1", respectively.
Further, in the second step, the following operations are sequentially performed on the underwater sound target sample data:
step 2.1, pre-emphasis, framing and windowing are carried out on underwater sound target sample data; the pre-emphasis is to improve the attenuation of the high-frequency part at the transmission end and increase the signal-to-noise ratio; the framing processing is to divide the underwater sound target sample into a frame sequence with a fixed time period; the windowing process adopts a Hamming window;
2.2, performing fast Fourier transform on each frame of data;
step 2.3, absolute value detection or square detection;
step 2.4, Mel filtering;
step 2.5, taking logarithm and discrete cosine transform;
and 2.6, outputting the dynamic characteristics.
Further, in step three, assume that the training sample data set is
Figure BDA0002631302370000021
Figure BDA0002631302370000022
n is the number of samples and n is the number of samples,
Figure BDA0002631302370000023
for data space, N is the dimension, and the dynamic feature mapping to the high-dimensional space can be separated by a hyperplane (ω · x) + b ═ 0, where
Figure BDA0002631302370000024
b belongs to R, R is a real number, the positive case samples and the negative case samples are distributed on two sides of the hyperplane, a vector nearest to the hyperplane is called a support vector, and a hyperplane is searched to maximize the distance between the positive case support vector and the negative case support vector, so that the problem is converted into a convex quadratic programming problem:
Figure BDA0002631302370000031
the optimization problem comprises a convex quadratic optimization object and can be linearly divided, in summary, the quadratic optimization of an optimal hyperplane needs to be found, and a lagrange multiplier is solved in matlab by a quadprog function;
formula (1) translates to the following dual problem:
Figure BDA0002631302370000032
in the formula (2), alpha is a Lagrange multiplier, C is a constant, and the fitting degree of the support vector machine identification model is controlled; κ is a kernel function; f () is a decision function;
by solving the equations (1) and (2), the weight expression can be obtained as follows:
Figure BDA0002631302370000033
the predicted result expression is as follows:
sgn(ωTφ(x)) (4)
where x is the input vector.
Further, the two-coordinate descent method specifically includes:
step 3.1, give lagrange multiplier alpha and its phaseCorresponding weight expression ω ═ Σiαiyiφ(xi)；
Step 3.2, calculating the squared value of the Euclidean length under the characteristic mapping phi, and recording the squared value as Qii：
Figure BDA0002631302370000034
3.3, selecting an alpha value, and circulating for n times;
step 3.4, calculate gradient G ═ yiωTφ(xi) 1, the gradient tends to zero to obtain an optimal alpha value and a weight omega, otherwise, step 3.5 and step 3.6 are executed to update the alpha value and the weight omega;
and 3.5, updating the alpha value:
Figure BDA0002631302370000035
αi←min(max(αi-G/Qii,0),C)，
Figure BDA0002631302370000036
is an estimated value;
step 3.6, updating the weight omega:
Figure BDA0002631302370000037
further, in the fourth step, 30% of dynamic feature sample data is selected as cross validation evaluation model parameters, and model cross validation is carried out on the support vector machine identification model with the built weight value updated; and selecting 40% of dynamic characteristic sample data as an evaluation model parameter, and carrying out model evaluation on the constructed support vector machine identification model with the updated weight value.
The invention has the beneficial effects that:
(1) compared with the traditional machine learning method, the method disclosed by the invention greatly improves the working efficiency, the running time, the recognition accuracy and the like, and can effectively classify the underwater sound target sample data.
(2) Compared with the traditional underwater sound target recognition method, the method combines the advantages of speech signal processing and machine learning and has better recognition effect.
(3) Aiming at the characteristics of the underwater sound target, the invention reasonably provides the optimal Lagrange multiplier and weight by using the dynamic characteristics of the MFCC output underwater sound target, thereby greatly improving the identification accuracy.
Drawings
Fig. 1 is a flow chart of mel-frequency spectral coefficient operation.
FIG. 2 is a schematic diagram of a support vector machine.
FIG. 3 is a flow chart of the underwater acoustic target identification method of the support vector machine for updating the weight.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
With reference to fig. 1 to fig. 3, the present invention provides an underwater acoustic target identification method based on an updated weight support vector machine, which specifically includes the following steps:
the method comprises the following steps: collecting two types of underwater sound target sample data, adding a category label, and establishing an underwater sound target sample library; in the first step, the collected underwater sound target sample data are classified and marked as "+ 1" and "-1" respectively.
Step two: performing MFCC processing on the underwater sound target sample library, and outputting dynamic characteristics;
in the second step, the following operations are sequentially carried out on the underwater sound target sample data:
step 2.1, pre-emphasis, framing and windowing are carried out on underwater sound target sample data; the pre-emphasis is to improve the attenuation of the high-frequency part at the transmission end and increase the signal-to-noise ratio; the framing processing is to divide the underwater sound target sample into a frame sequence with a fixed time period; the windowing process adopts a Hamming window;
2.2, performing fast Fourier transform on each frame of data;
step 2.3, absolute value detection or square detection;
step 2.4, Mel filtering;
step 2.5, taking logarithm and Discrete Cosine (DCT) transformation;
and 2.6, outputting the dynamic characteristics.
Step three: selecting a Gaussian radial basis kernel function as a kernel function, and giving a Lagrange multiplier alpha and a corresponding weight expression omega ═ Sigmaiαiyiφ(xi) N, n is the number of samples, y is 1,2iFor class labels, phi () represents a feature mapping function, xiIs sample data; using 30% of dynamic characteristics as training samples, and carrying out loop iteration on the training samples by using a double-coordinate descent method to obtain an optimal alpha value and a weight omega; constructing a support vector machine identification model for updating the weight according to the optimal alpha value and the weight omega;
in step three, assume that the training sample data set is
Figure BDA0002631302370000051
Figure BDA0002631302370000052
n is the number of samples and n is the number of samples,
Figure BDA0002631302370000053
for data space, N is the dimension, and the dynamic feature mapping to the high-dimensional space can be separated by a hyperplane (ω · x) + b ═ 0, where
Figure BDA0002631302370000054
b belongs to R, R is a real number, the positive case samples and the negative case samples are distributed on two sides of the hyperplane, a vector nearest to the hyperplane is called a support vector, and a hyperplane is searched to maximize the distance between the positive case support vector and the negative case support vector, so that the problem is converted into a convex quadratic programming problem:
Figure BDA0002631302370000055
the optimization problem comprises a convex quadratic optimization object and can be linearly divided, in summary, the quadratic optimization of an optimal hyperplane needs to be found, and a lagrange multiplier is solved in matlab by a quadprog function;
formula (1) translates to the following dual problem:
Figure BDA0002631302370000056
in the formula (2), alpha is a Lagrange multiplier, C is a constant, and the fitting degree of the support vector machine identification model is controlled; κ is a kernel function; f () is a decision function;
by solving the equations (1) and (2), the weight expression can be obtained as follows:
the predicted result expression is as follows:
sgn(ωTφ(x)) (4)
where x is the input vector.
From the result of the formula (4), it is important to obtain an accurate weight, and the model is adjusted according to the optimization of the weight as a starting point.
The two-coordinate descent method specifically comprises the following steps:
step 3.1, the lagrange multiplier alpha and the corresponding weight value expression omega ═ Σ are giveniαiyiφ(xi)；
Step 3.2, calculating the squared value of the Euclidean length under the characteristic mapping phi, and recording the squared value as Qii：
Figure BDA0002631302370000061
3.3, selecting an alpha value, and circulating for n times;
step 3.4, calculate gradient G ═ yiωTφ(xi) 1, the gradient tends to zero to obtain an optimal alpha value and a weight omega, otherwise, step 3.5 and step 3.6 are executed to update the alpha value and the weight omega;
and 3.5, updating the alpha value:
Figure BDA0002631302370000062
αi←min(max(αi-G/Qii,0),C)，
Figure BDA0002631302370000063
is an estimated value;
step 3.6, updating the weight omega:
Figure BDA0002631302370000064
step four: performing model cross validation and model evaluation on the constructed support vector machine identification model with the weight value updated by using the remaining 70% of dynamic characteristic samples;
in the fourth step, 30% of dynamic characteristic sample data is selected as cross validation evaluation model parameters, and model cross validation is carried out on the support vector machine identification model with the built weight value updated; and selecting 40% of dynamic characteristic sample data as an evaluation model parameter, and carrying out model evaluation on the constructed support vector machine identification model with the updated weight value. The time consumption of parameter selection by a grid search method is avoided, and the working efficiency is greatly improved.
Step five: counting the identification accuracy of the support vector machine identification model updated by the constructed weight, and evaluating the performance of the support vector machine identification model;
step six: and identifying the real data by using the obtained underwater sound target identification model of the support vector machine based on the updated weight.
Examples
The method comprises the following steps: for a second type of underwater sound target identification problem, acoustic information of whales collected at different time intervals is marked as positive samples (+1), and acoustic information samples of whales at different time intervals are marked as negative samples (-1). Respectively labeling the data and establishing a marine organism acoustic information sample library;
step two: extracting dynamic characteristics of data according to the marine organism acoustic information base obtained in the last step, processing and outputting the acoustic information dynamic characteristics of marine organisms by adopting MFCC (Mel frequency cepstrum coefficient), and establishing a marine organism acoustic information dynamic characteristic sample base;
the specific MFCC processing process comprises the following steps:
pre-emphasis: the attenuation of the high frequency part at the transmission end is mainly improved for increasing the signal-to-noise ratio. The specific implementation method is that a first-order finite excitation response high-pass filter is controlled:
Figure BDA0002631302370000071
in equation (5), x is an input speech signal, y is an emphasized signal, a and b are filter coefficients, and k is N, which represents the dimension.
Mel-filtering: based on the principle of human auditory perception. The formula for converting the common frequency to Mel frequency is as follows:
mel(f)＝2595*log10(1+f/700) (6)
where f is the actual frequency and mel (f) is the converted mel frequency.
Discrete Cosine Transform (DCT): the method is applied to signal processing and image processing, and the inverse transformation process in cepstrum analysis is mainly realized here. The one-dimensional DCT transform formula is as follows:
Figure BDA0002631302370000072
where f (i) is a coefficient of the cosine term, here set to a constant of 1; n is the point number of the original signal, c (u) is a compensation coefficient, which can make the DCT transformation matrix be an orthogonal matrix; f (u) is the entire DCT transformed coefficient.
The output result after discrete cosine transform is set as d (t):
D(t)＝F(u)·log(Y(t)) (8)
wherein Y (t) is the filtered output signal; and finally, obtaining the dynamic characteristics of each frame through a cepstrum lifting device.
The output results thus far can be described by a series of cepstral vectors, each vector corresponding to the MFCC features of each frame.
Step three: selecting 30% of samples in the dynamic characteristic sample library as training samples, updating and iterating to obtain optimal weight and alpha value, and constructing a support vector machine recognition model based on weight updating.
Step four: and selecting 30% of samples in the dynamic characteristic sample library as cross-validation samples. And taking the rest 40% as a test sample, judging that the model identification success rate is very high if the model identification accuracy rate of the test result is more than 90%, and returning to the third step to carry out weight optimization. The recognition rate of the model is verified to reach more than 90% through experiments, and the recognition rate is far higher than that of a traditional support vector machine.
Step five: according to the obtained underwater sound target recognition model of the support vector machine based on the updated weight, other data are distinguished, tiger whales and whales at the head can be well recognized, and the recognition rate is 92%.
The underwater acoustic target identification method based on the updated weight support vector machine provided by the invention is introduced in detail, a specific example is applied in the text to explain the principle and the implementation mode of the invention, and the description of the embodiment is only used for helping to understand the method and the core idea of the invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.

Claims (6)

1. An underwater sound target identification method based on an updated weight support vector machine is characterized in that: the method specifically comprises the following steps:
the method comprises the following steps: collecting two types of underwater sound target sample data, adding a category label, and establishing an underwater sound target sample library;
step two: performing MFCC processing on the underwater sound target sample library, and outputting dynamic characteristics;
step three: choosing Gaussian radial basisThe kernel function is a kernel function, and a Lagrange multiplier alpha and a corresponding weight value expression omega are giveniαiyiφ(xi) N, n is the number of samples, y is 1,2iFor class labels, phi () represents a feature mapping function, xiIs sample data; using 30% of dynamic characteristics as training samples, and carrying out loop iteration on the training samples by using a double-coordinate descent method to obtain an optimal alpha value and a weight omega; constructing a support vector machine identification model for updating the weight according to the optimal alpha value and the weight omega;
step four: performing model cross validation and model evaluation on the constructed support vector machine identification model with the weight value updated by using the remaining 70% of dynamic characteristic samples;
step five: counting the identification accuracy of the support vector machine identification model updated by the constructed weight, and evaluating the performance of the support vector machine identification model;
step six: and identifying the real data by using the obtained underwater sound target identification model of the support vector machine based on the updated weight.
2. The method of claim 1, wherein: in the first step, the collected underwater sound target sample data are classified and marked as "+ 1" and "-1" respectively.
3. The method of claim 2, wherein: in the second step, the following operations are sequentially carried out on the underwater sound target sample data:
step 2.1, pre-emphasis, framing and windowing are carried out on underwater sound target sample data; the pre-emphasis is to improve the attenuation of the high-frequency part at the transmission end and increase the signal-to-noise ratio; the framing processing is to divide the underwater sound target sample into a frame sequence with a fixed time period; the windowing process adopts a Hamming window;
2.2, performing fast Fourier transform on each frame of data;
step 2.3, absolute value detection or square detection;
step 2.4, Mel filtering;
step 2.5, taking logarithm and discrete cosine transform;
and 2.6, outputting the dynamic characteristics.
4. The method of claim 3, wherein: in step three, assume that the training sample data set is
Figure FDA0002631302360000011
n is the number of samples and n is the number of samples,
Figure FDA0002631302360000012
for data space, N is the dimension, and the dynamic feature mapping to the high-dimensional space can be separated by a hyperplane (ω · x) + b ═ 0, where
Figure FDA0002631302360000021
b belongs to R, R is a real number, the positive case samples and the negative case samples are distributed on two sides of the hyperplane, a vector nearest to the hyperplane is called a support vector, and a hyperplane is searched to maximize the distance between the positive case support vector and the negative case support vector, so that the problem is converted into a convex quadratic programming problem:
Figure FDA0002631302360000022
the optimization problem comprises a convex quadratic optimization object and can be linearly divided, in summary, the quadratic optimization of an optimal hyperplane needs to be found, and a lagrange multiplier is solved in matlab by a quadprog function;
formula (1) translates to the following dual problem:
Figure FDA0002631302360000023
in the formula (2), alpha is a Lagrange multiplier, C is a constant, and the fitting degree of the support vector machine identification model is controlled; κ is a kernel function; f () is a decision function;
by solving the equations (1) and (2), the weight expression can be obtained as follows:
Figure FDA0002631302360000024
the predicted result expression is as follows:
sgn(ωTφ(x)) (4)
where x is the input vector.
5. The method of claim 4, wherein: the two-coordinate descent method specifically comprises the following steps:
step 3.1, the lagrange multiplier alpha and the corresponding weight value expression omega ═ Σ are giveniαiyiφ(xi)；
Step 3.2, calculating the squared value of the Euclidean length under the characteristic mapping phi, and recording the squared value as Qii：
Figure FDA0002631302360000025
3.3, selecting an alpha value, and circulating for n times;
step 3.4, calculate gradient G ═ yiωTφ(xi) 1, the gradient tends to zero to obtain an optimal alpha value and a weight omega, otherwise, step 3.5 and step 3.6 are executed to update the alpha value and the weight omega;
and 3.5, updating the alpha value:
Figure FDA0002631302360000026
αi←min(max(αi-G/Qii,0),C)，
Figure FDA0002631302360000027
is an estimated value;
step 3.6, updating the weight omega:
Figure FDA0002631302360000028
6. the method of claim 5, wherein: in the fourth step, 30% of dynamic characteristic sample data is selected as cross validation evaluation model parameters, and model cross validation is carried out on the support vector machine identification model with the built weight value updated; and selecting 40% of dynamic characteristic sample data as an evaluation model parameter, and carrying out model evaluation on the constructed support vector machine identification model with the updated weight value.
CN202010811948.0A 2020-08-13 2020-08-13 Underwater sound target identification method based on weight updating support vector machine Pending CN112052880A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010811948.0A CN112052880A (en) 2020-08-13 2020-08-13 Underwater sound target identification method based on weight updating support vector machine

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010811948.0A CN112052880A (en) 2020-08-13 2020-08-13 Underwater sound target identification method based on weight updating support vector machine

Publications (1)

Publication Number Publication Date
CN112052880A true CN112052880A (en) 2020-12-08

Family

ID=73601754

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010811948.0A Pending CN112052880A (en) 2020-08-13 2020-08-13 Underwater sound target identification method based on weight updating support vector machine

Country Status (1)

Country Link
CN (1) CN112052880A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113252161A (en) * 2021-04-26 2021-08-13 西北工业大学 Deep forest based small sample underwater target identification method
CN113345443A (en) * 2021-04-22 2021-09-03 西北工业大学 Marine mammal vocalization detection and identification method based on mel-frequency cepstrum coefficient

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106096557A (en) * 2016-06-15 2016-11-09 浙江大学 A kind of semi-supervised learning facial expression recognizing method based on fuzzy training sample
CN106203485A (en) * 2016-07-01 2016-12-07 北京邮电大学 A kind of parallel training method and device of support vector machine
CN107368840A (en) * 2016-05-11 2017-11-21 中国科学院声学研究所 A kind of underwater manoeuvre Small object recognition methods based on MFCC and SVMs
CN107682786A (en) * 2017-10-31 2018-02-09 广东小天才科技有限公司 A kind of microphone apparatus anti-interference method and microphone apparatus

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107368840A (en) * 2016-05-11 2017-11-21 中国科学院声学研究所 A kind of underwater manoeuvre Small object recognition methods based on MFCC and SVMs
CN106096557A (en) * 2016-06-15 2016-11-09 浙江大学 A kind of semi-supervised learning facial expression recognizing method based on fuzzy training sample
CN106203485A (en) * 2016-07-01 2016-12-07 北京邮电大学 A kind of parallel training method and device of support vector machine
CN107682786A (en) * 2017-10-31 2018-02-09 广东小天才科技有限公司 A kind of microphone apparatus anti-interference method and microphone apparatus

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
CHO-JUI HSIEH 等: "A Dual Coordinate Descent Method for Large-scale Linear SVM", 《PROCEEDINGS OF THE 25TH INTERNATIONAL CONFERENCE ON MACHINE LEARNING》, 30 July 2008 (2008-07-30), pages 408 - 415, XP058106342, DOI: 10.1145/1390156.1390208 *
阿泽: "【机器学习】支持向量机 SVM（非常详细）", 《HTTPS://ZHUANLAN.ZHIHU.COM/P/77750026》, 13 June 2020 (2020-06-13), pages 1 - 8 *
黄华娟 等: "基于模糊核聚类粒化的粒度支持向量机", 《智能系统学报》, vol. 14, no. 6, 5 September 2019 (2019-09-05), pages 1271 - 1277 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113345443A (en) * 2021-04-22 2021-09-03 西北工业大学 Marine mammal vocalization detection and identification method based on mel-frequency cepstrum coefficient
CN113252161A (en) * 2021-04-26 2021-08-13 西北工业大学 Deep forest based small sample underwater target identification method

Similar Documents

Publication Publication Date Title
CN107657964B (en) Depression auxiliary detection method and classifier based on acoustic features and sparse mathematics
CN107564513B (en) Voice recognition method and device
CN112364779A (en) Underwater sound target identification method based on signal processing and deep-shallow network multi-model fusion
CN111724770B (en) Audio keyword identification method for generating confrontation network based on deep convolution
CN113361636B (en) Image classification method, system, medium and electronic device
CN110853630B (en) Lightweight speech recognition method facing edge calculation
CN112750442B (en) Crested mill population ecological system monitoring system with wavelet transformation and method thereof
JP6923089B2 (en) Information processing equipment, methods and programs
Wei et al. A method of underwater acoustic signal classification based on deep neural network
CN112052880A (en) Underwater sound target identification method based on weight updating support vector machine
CN103456302A (en) Emotion speaker recognition method based on emotion GMM model weight synthesis
CN111564179A (en) Species biology classification method and system based on triple neural network
CN105931646A (en) Speaker identification method base on simple direct tolerance learning algorithm
CN113111786B (en) Underwater target identification method based on small sample training diagram convolutional network
CN117310668A (en) Underwater sound target identification method integrating attention mechanism and depth residual error shrinkage network
CN113488069B (en) Speech high-dimensional characteristic rapid extraction method and device based on generation type countermeasure network
CN114818789A (en) Ship radiation noise identification method based on data enhancement
CN111837185A (en) Sound classification method, device and medium based on constrained semi-nonnegative matrix factorization
CN112735442B (en) Wetland ecology monitoring system with audio separation voiceprint recognition function and audio separation method thereof
CN112735477A (en) Voice emotion analysis method and device
CN112348072A (en) Health state assessment method based on slow feature analysis and hidden Markov
Therese et al. A linear visual assessment tendency based clustering with power normalized cepstral coefficients for audio signal recognition system
CN115910099B (en) Automatic musical instrument identification method based on depth probability map neural network
CN117116290B (en) Method and related equipment for positioning defects of numerical control machine tool parts based on multidimensional characteristics
CN116052725B (en) Fine granularity borborygmus recognition method and device based on deep neural network

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