CN114462457A - Ship underwater noise deep learning identification method based on intrinsic probability density function - Google Patents

Ship underwater noise deep learning identification method based on intrinsic probability density function Download PDF

Info

Publication number
CN114462457A
CN114462457A CN202210370774.8A CN202210370774A CN114462457A CN 114462457 A CN114462457 A CN 114462457A CN 202210370774 A CN202210370774 A CN 202210370774A CN 114462457 A CN114462457 A CN 114462457A
Authority
CN
China
Prior art keywords
probability density
ship
signal
noise
function
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.)
Granted
Application number
CN202210370774.8A
Other languages
Chinese (zh)
Other versions
CN114462457B (en
Inventor
姜莹
刘宗伟
杨春梅
吕连港
段德鑫
张远凌
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.)
First Institute of Oceanography MNR
Original Assignee
First Institute of Oceanography MNR
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 First Institute of Oceanography MNR filed Critical First Institute of Oceanography MNR
Priority to CN202210370774.8A priority Critical patent/CN114462457B/en
Publication of CN114462457A publication Critical patent/CN114462457A/en
Application granted granted Critical
Publication of CN114462457B publication Critical patent/CN114462457B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • 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
    • 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
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Signal Processing (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

The invention relates to a deep learning and identifying method for underwater noise of ships based on an intrinsic probability density function, which belongs to the technical field of ocean information. The method is more stable in complex and variable marine environments, compared with the prior art, the identification accuracy is improved from 90.8% to 99.6%, the operation speed is improved, and the online processing requirement can be met.

Description

Ship underwater noise deep learning identification method based on intrinsic probability density function
Technical Field
The invention belongs to the technical field of ocean information, and relates to a deep learning and identification method for ship underwater noise based on an intrinsic probability density function.
Background
Accurate identification of the marine vessel target has important practical significance on marine vessel monitoring, air route planning, marine equity maintenance, marine military capacity improvement and the like. Currently, underwater radiated noise is one of the commonly used signals in ship identification.
Noise signals generated in the sailing process of ships can be transmitted to the periphery in the form of sound waves under water. The sound wave propagation can be influenced by the boundaries of the marine water body environment and the sea surface and the seabed, and is a complex physical process which is expressed in two aspects: one is that the sound channel can produce modulation effect on the signal when in transmission, and has the characteristic of nonlinearity; the other is that physical ocean dynamic phenomena such as ocean, internal wave and frontal surface can introduce non-steady time-varying characteristics. The non-stationary and non-linear characteristics of the underwater radiation noise signal bring difficulty to the identification of the ship.
At present, ship identification is mainly based on the characteristics of underwater radiation noise time domain (probability density method, zero-crossing rate method and the like) and frequency domain (frequency spectrum, high-order statistics, time-frequency analysis and the like). The methods are mainly based on the assumption of stability or near stability, and the nature and stable characteristics of the corresponding target are difficult to obtain, so that the identification effect is poor and the error is large. The commonly used modulation spectrum method at present is based on the modulation mechanism of propeller rotation on cavitation noise, obtains the characteristics corresponding to the physical essence of a target in a signal, and can further obtain information such as the number of propeller blades, the rotating speed and the like. However, in the modulation spectrum method, a near-stationary approximation is adopted, so that the modulation spectrum method is difficult to obtain stable characteristics, is seriously influenced by time variation of underwater ship signals, and has poor identification effect.
Disclosure of Invention
The invention aims to solve the technical problem of providing a deep learning and identifying method for ship underwater noise based on an intrinsic probability density function. The method obtains an intrinsic mode function of a signal through mode decomposition, solves corresponding probability density, and adopts a deep learning classifier to realize automatic identification of the type of the ship.
The invention is realized by the following technical scheme:
a ship underwater noise deep learning identification method based on an intrinsic probability density function comprises the following steps:
firstly, extracting a section of noise signal s with the total length T to obtain a normalized noise signal st(ii) a The step of normalizing is to unify the sampling rates of the different data samples to FSstdIn which FSstdGreater than 20 kHz; the dc component of the signal is then removed:
Figure 470244DEST_PATH_IMAGE001
where mean () represents the averaging operation; finally, normalizing the signal power:
Figure 977449DEST_PATH_IMAGE002
wherein std () represents a take standard deviation operation;
secondly, calculating an intrinsic probability density function of the signal; first, for the signal stPerforming modal decomposition, and expressing as follows:
Figure 65621DEST_PATH_IMAGE003
,
Figure 771409DEST_PATH_IMAGE004
（1）
wherein, ci(t) isiAn Intrinsic Mode Function (IMF) component;
then, for each intrinsic mode function, the probability density function estimated by the kernel function is obtained
Figure 919625DEST_PATH_IMAGE005
；
Figure 710863DEST_PATH_IMAGE006
（2）
Wherein, K () is a kernel function, and h is a smoothing parameter;
the combination of the probability density functions of all the obtained intrinsic mode functions is the intrinsic probability density function of the signal;
thirdly, solving the deviation of the signal intrinsic probability density function and the Gaussian function, and combining to obtain a deviation profile;
fourthly, inputting the deviation contour map as the characteristic of the classification model, and training the classification model; wherein the classification model adopts a neural network model based on deep learning, specifically a Convolutional Neural Network (CNN) model; dividing the amplitude modulation holographic spectrogram into a training set sample and a verification set sample, training the CNN model by using the training set, and performing parameter adjustment on the network model by using the verification set to obtain a trained CNN model;
fifthly, classifying and identifying the ship noise by using the trained classification model; and processing the ship noise according to the first step to the third step to obtain a deviation profile, inputting the deviation profile into the trained CNN model, and giving a corresponding ship classification result by the CNN model.
Further, for ship noise, N in equation (1) is greater than 8.
Further, if the gaussian curve corresponding to the normal distribution is selected as the kernel function in the formula (2), then
Figure 602727DEST_PATH_IMAGE007
（3）
Compared with the prior art, the invention has the beneficial effects that:
the method uses the deviation profile as the characteristic of the ship noise, the deviation profile represents the condition of deviation from normal distribution in the signal, which often corresponds to the external force action of the physical structure (mainly driven by a propeller) of a ship on a water body, and the interference degree of environmental noise on the ship characteristic can be weakened, so that the characteristic is more stable in a complex and changeable marine environment, and compared with the prior art, the identification accuracy is improved to 99.6% from 90.8%. The CNN model has proven to have a powerful capability in picture recognition. The deviation profile map is combined with the CNN model, and finally accurate and stable identification of the ship noise can be realized. The processing speed of the method depends on the EMD decomposition speed, and the required CNN model has a simple structure and can meet the online processing requirement.
Drawings
FIG. 1 is a flow chart of the process steps of the method of the present invention;
FIG. 2 is a time domain waveform and a spectrogram of a marine vessel signal;
FIG. 3 is an eigenmode function graph obtained by decomposing A ship signals;
FIG. 4 is an eigenmode function graph obtained by decomposing a B ship signal;
FIG. 5 is a graph of a probability density function versus a normal distribution for each eigenmode function of two vessel signals; a is a comparison graph of the ship A, and B is a comparison graph of the ship B;
FIG. 6 is a graph of the deviation profile of two vessel signals; a is the deviation profile of the ship A, and B is the deviation profile of the ship B.
Detailed Description
There are two general processing modes for the collected ship underwater radiation noise, one is to analyze and process the return data on the shore; the other is local online processing at reception. The latter of which places high demands on processing speed and computational resources.
Examples
Here, the underwater noise signals of 5 ships are taken as an example, and the ship names, the lengths and the sampling rate information are shown in the following table. The signals are framed with a length of 10 seconds each, and the frames overlap by 9 seconds, so that for these 5 vessels we have 123, 170, 215, 402 and 675 frames, respectively. For these data, the processing steps of the method of the invention (see FIG. 1) are as follows:
Figure 100705DEST_PATH_IMAGE008
and (5) standardizing operation in the first step. Performing normalization operation on a frame of noise signal s with the total length of 10 seconds to obtain a normalized noise signal st. Since the sampling rates are already consistent, uniform sampling rate operation is no longer required. Removing the direct current component of the signal:
Figure 934668DEST_PATH_IMAGE001
where mean () represents the averaging operation; power normalization of the signal:
Figure 695427DEST_PATH_IMAGE002
where std () represents a take standard deviation operation. After normalization, the time domain waveform of the a-ship signal is shown as a in fig. 2, and B in fig. 2 is a spectrum diagram of the B-ship signal.
And secondly, calculating a probability density function of the intrinsic mode functions of the signals. First, for the signal stCarrying out modal decomposition, wherein the modal decomposition method comprises the following steps: empirical Mode Decomposition (EMD), Ensemble Empirical Mode Decomposition (EEMD), or adaptive noise-complete ensemble empirical mode decomposition (CEEM-EMD). Here, EEMD is selected to decompose these ship signals. After modal decomposition, the signal can be represented as:
Figure 577932DEST_PATH_IMAGE003
wherein c isi(t) is the Intrinsic Mode Function (IMF) component. As shown in FIG. 3 and FIG. 4, the results of the decomposition of the noise signals of two ships are shown, wherein Data is the original signal, IMFiIs as followsiAnd each eigenmode function, namely Residual is decomposed, and the right-side numerical value is the frequency corresponding to each eigenmode function and the Residual.
Then, for each intrinsic mode function, the probability density function is obtained:
Figure 992733DEST_PATH_IMAGE006
where K () is a kernel function and h is a smoothing parameter. Commonly used kernel functions are: rectangles, triangles, Epanechnikov curves, gaussian curves, etc. For the noise signal of the ship, a gaussian curve corresponding to a normal distribution is preferably used as a kernel function,
Figure 748330DEST_PATH_IMAGE007
，h=100。
thirdly, solving the deviation between the probability density function and the gaussian function of each intrinsic mode function (as shown in fig. 5, the black dotted line is the gaussian function, and the gray solid line is the probability density function of the intrinsic mode function), and combining to obtain a deviation profile, as shown in fig. 6.
And fourthly, taking the deviation contour map as the characteristic input of the classification model, and training the classification model. The classification model adopts a neural network model based on deep learning, specifically a Convolutional Neural Network (CNN) model. The convolutional neural network has 6 layers: three convolution layers, two full connection layers and one softmax classification layer. The first layer of convolutional layers has 32 3 x 3 convolutional kernels, followed by a 2 x 2 max pooling layer. The second convolutional layer has 8 3 × 3 convolutional kernels, followed by a 2 × 2 max pooling layer. The structure of the third layer of the convolution layer is completely the same as that of the second layer. The step size of all convolutional layers is 1, and each convolutional layer is activated by using the tanh activation function. And dividing the deviation profile map into a training set sample and a verification set sample, training the CNN model by using the training set, and adjusting parameters of the network model by using the verification set to obtain the trained CNN model.
And fifthly, carrying out classification and identification on the ship noise by using the trained classification model. And (4) processing the ship noise according to the steps from one step to three to obtain a deviation profile, inputting the deviation profile into the trained CNN model, and giving corresponding ship classification by the CNN model. The recognition accuracy obtained on the verification set is 99.6%, and the recognition accuracy of the comparative modulation Spectrum Method under the same conditions is 90.8% [ see the literature Liu, Zongwei, et al. "DEMON Spectrum Extraction Method Using Empirical Mode decomposition." OCEANS-MTS/IEEE Kobe Techno-Oceans (OTO). IEEE, 2018 ], which is a great improvement.

Claims (3)

1. A ship underwater noise deep learning identification method based on an intrinsic probability density function is characterized by comprising the following steps:
firstly, extracting a section of noise signal s with the total length T to obtain a normalized noise signal st(ii) a The step of normalizing is to unify the sampling rates of the different data samples to FSstdIn which FSstdGreater than 20 kHz; the dc component of the signal is then removed:
Figure 446774DEST_PATH_IMAGE001
where mean () represents the averaging operation; finally, the signal work is carried outRate normalization:
Figure 469307DEST_PATH_IMAGE002
wherein std () represents a take standard deviation operation;
secondly, calculating an intrinsic probability density function of the signal; first, for the signal stPerforming modal decomposition, and expressing as follows:
Figure 327672DEST_PATH_IMAGE003
,
Figure 988461DEST_PATH_IMAGE004
（1）
wherein, ci(t) isiAn intrinsic mode function component;
then, for each intrinsic mode function, the probability density function estimated by the kernel function is obtained
Figure 130729DEST_PATH_IMAGE005
；
Figure 468038DEST_PATH_IMAGE006
（2）
Wherein, K () is a kernel function, and h is a smoothing parameter;
the combination of the probability density functions of all the obtained intrinsic mode functions is the intrinsic probability density function of the signal;
thirdly, solving the deviation of the signal intrinsic probability density function and the Gaussian function, and combining to obtain a deviation profile;
fourthly, inputting the deviation contour map as the characteristic of the classification model, and training the classification model; the classification model adopts a neural network model based on deep learning, in particular a convolutional neural network model; dividing the amplitude modulation holographic spectrogram into a training set sample and a verification set sample, training the convolutional neural network model by using the training set, and performing parameter adjustment on the network model by using the verification set to obtain a trained convolutional neural network model;
fifthly, classifying and identifying the ship noise by using the trained classification model; and processing the ship noise according to the first step to the third step to obtain a deviation profile, inputting the deviation profile into the trained convolutional neural network model, and giving a corresponding ship classification result by the convolutional neural network model.
2. The method for deep learning and identifying the underwater noise of the ship based on the eigen probability density function as claimed in claim 1, wherein N in the formula (1) is greater than 8 for the ship noise.
3. The method for deep learning and identifying the underwater noise of the ship based on the eigen probability density function as claimed in claim 1, wherein the gaussian curve corresponding to the normal distribution is selected as the kernel function in the formula (2), and then
Figure 62968DEST_PATH_IMAGE007
（3）。
CN202210370774.8A 2022-04-11 2022-04-11 Ship underwater noise deep learning identification method based on intrinsic probability density function Active CN114462457B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210370774.8A CN114462457B (en) 2022-04-11 2022-04-11 Ship underwater noise deep learning identification method based on intrinsic probability density function

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210370774.8A CN114462457B (en) 2022-04-11 2022-04-11 Ship underwater noise deep learning identification method based on intrinsic probability density function

Publications (2)

Publication Number Publication Date
CN114462457A true CN114462457A (en) 2022-05-10
CN114462457B CN114462457B (en) 2022-07-08

Family

ID=81417401

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210370774.8A Active CN114462457B (en) 2022-04-11 2022-04-11 Ship underwater noise deep learning identification method based on intrinsic probability density function

Country Status (1)

Country Link
CN (1) CN114462457B (en)

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1609631A (en) * 2004-11-25 2005-04-27 上海交通大学 Method for testing and identifying underwater sound noise based on small wave area
CN104749568A (en) * 2013-12-26 2015-07-01 中国科学院声学研究所 Shallow sea target depth classification method based on hydrophone array
CN105092023A (en) * 2015-08-11 2015-11-25 西安科技大学 Bolt vibration signal correction method based on white noise statistical characteristics
CN107464226A (en) * 2017-07-31 2017-12-12 东南大学 A kind of image de-noising method based on improvement two-dimensional empirical mode decomposition algorithm
CN107609488A (en) * 2017-08-21 2018-01-19 哈尔滨工程大学 A kind of ship noise method for identifying and classifying based on depth convolutional network
CN109784410A (en) * 2019-01-18 2019-05-21 西安邮电大学 A kind of feature extraction and classification method of ships radiated noise signal
US20190164564A1 (en) * 2016-05-23 2019-05-30 The University Of New Hampshire Techniques for Empirical Mode Decomposition (EMD)-Based Signal De-Noising Using Statistical Properties of Intrinsic Mode Functions (IMFs)
CN109946389A (en) * 2019-01-31 2019-06-28 青岛理工大学 Damage Detection of Structures based on overall experience mode decomposition and convolutional neural networks
CN110186557A (en) * 2019-06-05 2019-08-30 国网江苏省电力有限公司检修分公司 A kind of Reactor Fault diagnostic method
CN112230205A (en) * 2020-10-16 2021-01-15 哈尔滨工程大学 Underwater target recognition system performance evaluation method using ship radiation noise simulation signal
CN113077813A (en) * 2021-03-22 2021-07-06 自然资源部第一海洋研究所 Ship noise identification method based on holographic spectrum and deep learning

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1609631A (en) * 2004-11-25 2005-04-27 上海交通大学 Method for testing and identifying underwater sound noise based on small wave area
CN104749568A (en) * 2013-12-26 2015-07-01 中国科学院声学研究所 Shallow sea target depth classification method based on hydrophone array
CN105092023A (en) * 2015-08-11 2015-11-25 西安科技大学 Bolt vibration signal correction method based on white noise statistical characteristics
US20190164564A1 (en) * 2016-05-23 2019-05-30 The University Of New Hampshire Techniques for Empirical Mode Decomposition (EMD)-Based Signal De-Noising Using Statistical Properties of Intrinsic Mode Functions (IMFs)
CN107464226A (en) * 2017-07-31 2017-12-12 东南大学 A kind of image de-noising method based on improvement two-dimensional empirical mode decomposition algorithm
CN107609488A (en) * 2017-08-21 2018-01-19 哈尔滨工程大学 A kind of ship noise method for identifying and classifying based on depth convolutional network
CN109784410A (en) * 2019-01-18 2019-05-21 西安邮电大学 A kind of feature extraction and classification method of ships radiated noise signal
CN109946389A (en) * 2019-01-31 2019-06-28 青岛理工大学 Damage Detection of Structures based on overall experience mode decomposition and convolutional neural networks
CN110186557A (en) * 2019-06-05 2019-08-30 国网江苏省电力有限公司检修分公司 A kind of Reactor Fault diagnostic method
CN112230205A (en) * 2020-10-16 2021-01-15 哈尔滨工程大学 Underwater target recognition system performance evaluation method using ship radiation noise simulation signal
CN113077813A (en) * 2021-03-22 2021-07-06 自然资源部第一海洋研究所 Ship noise identification method based on holographic spectrum and deep learning

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
ZONGWEI LIU等: "《DEMON Spectrum Extraction Method Using Empirical Mode Decomposition》", 《2018 OCEANS - MTS/IEEE KOBE TECHNO-OCEANS (OTO)》 *
张宁等: "《基于变分模态分解和概率密度估计的变压器绕组变形在线检测方法》", 《电网技术》 *

Also Published As

Publication number Publication date
CN114462457B (en) 2022-07-08

Similar Documents

Publication Publication Date Title
CN109800700B (en) Underwater acoustic signal target classification and identification method based on deep learning
CN100543842C (en) Realize the method that ground unrest suppresses based on multiple statistics model and least mean-square error
CN113077813B (en) Ship noise identification method based on holographic spectrum and deep learning
CN106772331B (en) Target identification method and Target Identification Unit
CN113707176B (en) Transformer fault detection method based on acoustic signal and deep learning technology
CN112731330B (en) Radar carrier frequency parameter change steady target identification method based on transfer learning
Luo et al. An underwater acoustic target recognition method based on combined feature with automatic coding and reconstruction
Wang et al. Deep learning assisted time-frequency processing for speech enhancement on drones
CN104732970A (en) Ship radiation noise recognition method based on comprehensive features
CN111010356A (en) Underwater acoustic communication signal modulation mode identification method based on support vector machine
CN111816200B (en) Multi-channel speech enhancement method based on time-frequency domain binary mask
Lu et al. Fundamental frequency detection of underwater acoustic target using DEMON spectrum and CNN network
CN111899750A (en) Speech enhancement algorithm combining cochlear speech features and hopping deep neural network
CN112666533B (en) Repetition frequency change steady target identification method based on spatial pyramid pooling network
Liu et al. Application of multi-algorithm mixed feature extraction model in underwater acoustic signal
CN114462457B (en) Ship underwater noise deep learning identification method based on intrinsic probability density function
Ashraf et al. Ambient-noise free generation of clean underwater ship engine audios from hydrophones using generative adversarial networks
Xie et al. Data augmentation and deep neural network classification based on ship radiated noise
CN114462458B (en) Ship underwater signal noise reduction and target enhancement method
Ashraf et al. Improved CycleGAN for underwater ship engine audio translation
Mohammed et al. A GTCC-based underwater HMM target classifier with fading channel compensation
CN108573698B (en) Voice noise reduction method based on gender fusion information
Song et al. Underwater acoustic signal noise reduction based on fully convolutional time domain separation network
Shi et al. Research on armored vehicle classification based on MFCC and SVM
Bansal et al. Performance of speaker recognition system using shifted mfcc, delta spectral cepstral coefficient (DSCC) and Fuzzy techniques

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