CN113642245B - Construction method of ship radiation noise data set - Google Patents
Construction method of ship radiation noise data set Download PDFInfo
- Publication number
- CN113642245B CN113642245B CN202110975058.8A CN202110975058A CN113642245B CN 113642245 B CN113642245 B CN 113642245B CN 202110975058 A CN202110975058 A CN 202110975058A CN 113642245 B CN113642245 B CN 113642245B
- Authority
- CN
- China
- Prior art keywords
- noise
- data
- spectrum
- data set
- ship
- 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
- 230000005855 radiation Effects 0.000 title claims abstract description 29
- 238000010276 construction Methods 0.000 title abstract description 3
- 238000001228 spectrum Methods 0.000 claims abstract description 80
- 238000000034 method Methods 0.000 claims abstract description 38
- 238000005070 sampling Methods 0.000 claims abstract description 8
- 238000004088 simulation Methods 0.000 claims abstract description 8
- 230000003595 spectral effect Effects 0.000 claims description 19
- 230000008569 process Effects 0.000 claims description 12
- 238000009826 distribution Methods 0.000 claims description 11
- 238000013178 mathematical model Methods 0.000 claims description 10
- 238000010586 diagram Methods 0.000 claims description 4
- 230000002706 hydrostatic effect Effects 0.000 claims description 4
- 230000000007 visual effect Effects 0.000 claims description 4
- 238000005315 distribution function Methods 0.000 claims description 2
- 230000008485 antagonism Effects 0.000 claims 1
- 238000005516 engineering process Methods 0.000 abstract description 6
- 238000013135 deep learning Methods 0.000 abstract description 4
- 238000007781 pre-processing Methods 0.000 abstract description 2
- 238000005065 mining Methods 0.000 abstract 1
- 230000005236 sound signal Effects 0.000 abstract 1
- 238000012549 training Methods 0.000 description 21
- 238000004422 calculation algorithm Methods 0.000 description 8
- 238000004458 analytical method Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 229920006395 saturated elastomer Polymers 0.000 description 3
- 238000009827 uniform distribution Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000012217 deletion Methods 0.000 description 2
- 230000037430 deletion Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 240000007594 Oryza sativa Species 0.000 description 1
- 235000007164 Oryza sativa Nutrition 0.000 description 1
- 235000003181 Panax pseudoginseng Nutrition 0.000 description 1
- 244000131316 Panax pseudoginseng Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001010 compromised effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000000977 initiatory effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 235000009566 rice Nutrition 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/21—Design, administration or maintenance of databases
- G06F16/211—Schema design and management
- G06F16/212—Schema design and management with details for data modelling support
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/10—Noise analysis or noise optimisation
-
- 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)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Molecular Biology (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Algebra (AREA)
- Animal Behavior & Ethology (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
A method for constructing a ship radiation noise data set belongs to the field of underwater sound signal identification. The invention solves the problem that the radiation noise sample of the actual mining ship is rare. According to the invention, according to the existing ship radiation noise data actual sampling sample, the ship radiation noise meeting the actual requirement is generated by using a simulation technology, the simulated ship radiation noise and the actual marine noise are taken as an original noise data set, the LOFAR spectrum is used for preprocessing, key characteristics are reserved, and finally the sample expansion is realized by using GAN, so that more data sets are obtained to meet the requirement of deep learning on large data volume. The invention can be applied to the construction of the ship radiation noise data set.
Description
Technical Field
The invention belongs to the field of underwater acoustic signal identification, and particularly relates to a method for constructing a ship radiation noise data set.
Background
In recent years, deep learning related technology has made rapid application development in application fields such as image radar, sonar and the like, and has made a great contribution in many application fields. In some special fields, such as the field of underwater acoustic signal recognition, training data sets may be deficient due to safety concerns. The underwater target recognition field is difficult to collect and manufacture a data set due to confidentiality, safety and other reasons, and meanwhile, the industry currently lacks a unified and standard data set, so that the quantity of the actual collection ship radiation noise samples is scarce. Since deep learning techniques require a large number of data sets to ensure high accuracy, the lack of data sets can greatly degrade the performance of the algorithm.
Disclosure of Invention
The invention aims to solve the problem that the actual collection ship radiation noise sample is scarce, and provides a method for constructing a ship radiation noise data set.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a method for constructing a ship radiation noise data set specifically comprises the following steps:
firstly, establishing a mathematical model of actual marine noise data, simulating the simulated marine noise data by using the established mathematical model, and taking a set of the actual marine noise data and the simulated marine noise data as an original data set;
step two: the method comprises the steps of placing sound files in an original data set into simulation software to observe values, obtaining data of the sound files in the original data set, and converting the data of the sound files to obtain waveform patterns of the sound files;
step three: expanding the waveform spectrum obtained in the second step by using a generated countermeasure network to obtain an expanded waveform spectrum;
step four: and obtaining an expanded data set according to the expanded waveform map, and taking the expanded data set and the set of the original data set as a ship radiation noise data set.
The beneficial effects of the invention are as follows: the invention provides a method for constructing a ship radiation noise data set, which is characterized in that according to the existing ship radiation noise data actual sampling sample, the ship radiation noise meeting the actual requirement is generated by using a simulation technology, the simulated ship radiation noise and the actual marine noise are taken as an original noise data set, a LOFAR spectrum is used for preprocessing, key characteristics are reserved, and finally, the sample expansion is realized by using GAN, so that more data sets are obtained to meet the requirement of deep learning on a large data quantity, and the problem of scarcity of the actual ship radiation noise sample is solved.
Drawings
FIG. 1 is a simulated fitted noise waveform;
fig. 2 (a) is an Audio map;
FIG. 2 (b) is a DEMON map;
FIG. 2 (c) is a LOFAR profile;
FIG. 2 (d) is a Histogram map;
fig. 3 is a flow chart of the GAN algorithm.
Detailed Description
The first embodiment of the present invention provides a method for constructing a ship radiation noise data set, the method specifically including the following steps:
firstly, establishing a mathematical model of actual marine noise data, simulating the simulated marine noise data by using the established mathematical model, and taking a set of the actual marine noise data and the simulated marine noise data as an original data set;
the actual marine noise data and the simulated ship noise data are sound files in the original data set;
extracting characteristics of the sound file by utilizing LSTM and CNN networks, and checking the accuracy and the deletion rate of the obtained audio so as to detect the accuracy of the original data set; the specific process is as follows:
(1) Firstly, extracting audio characteristics, wherein the characteristic extraction mainly comprises four parts, namely the number of channels, the size of audio samples, the audio sampling rate and the total frame number; wherein the number of channels is fixed to be 1, the sampling size is 3, and the sampling rate is 52734; the quality of audio is code rate related, code rate = sample size × sample rate × number of channels;
(2) Then, feature recognition is performed, and from the feature extraction graph, we can see that the marine noise is the same in terms of sound except for the frame rate, but the marine noise cannot be said to have no difference in noise, and only the results obtained under the condition of the same sea area or the same marine environment can be said to be the same;
(3) After the characteristics of the audio are extracted and the data are read, the accuracy and the deletion rate of the sound audio can be simply detected, and the training frequency is set to be 100 times in order to observe the percentage of the accuracy more conveniently; from the feature extraction map, it can be seen that the data is reliable in most cases, and that the presence of a miss is acceptable in only individual cases.
Step two: the method comprises the steps of placing sound files in an original data set into simulation software to observe values, obtaining data of the sound files in the original data set, and converting the data of the sound files to obtain waveform patterns of the sound files;
step three: expanding the waveform spectrum obtained in the second step by using a generated countermeasure network (GAN) to obtain an expanded waveform spectrum;
the training times are controlled, the image quality and the definition of the extended data set under different training times are compared, the extended data set is ensured to be lifelike compared with the original data set, and the original characteristic frequency section and the characteristic value are reserved as much as possible;
step four: and obtaining an expanded data set according to the expanded waveform map, and taking the expanded data set and the set of the original data set as a ship radiation noise data set.
The second embodiment is different from the first embodiment in that: the mathematical model of the actual marine noise data is as follows:
S(t)＝1+[G(t)]×S x (t)+S l (t) (1)
wherein: s (t) is actual marine noise data, G (t) is modulation signal, S x (t) time domain waveform of continuous spectrum component, S l And (t) is a time domain waveform of the line spectrum component.
Other steps and parameters are the same as in the first embodiment.
And respectively carrying out modeling analysis on the continuous spectrum characteristics and the line spectrum statistical rules of the ship noise.
The third embodiment is different from one or two embodiments in that: the method utilizes the established mathematical model to simulate the noise data of the simulated ship, and comprises the following specific processes:
the method comprises the following steps: spectral peak frequency f 0 Is influenced by factors such as the target speed, the depth, the displacement, the size of a propeller, the water pressure and the like of the ship; wherein Ross proposes the spectral peak frequency f 0 The empirical formula of the relation with the target working condition is:
wherein B represents the number of propeller blades, sigma represents the number of cavitation bubbles, and P 0 Represents hydrostatic pressure, ρ represents density, and "-" represents parameters related thereto;
step two: assuming hydrostatic pressure P 0 Invariable, the density ρ is:
wherein v is the target speed, H is the depth, J p For the running speed coefficient;
step one, three: substituting formula (3) into formula (2) to obtain formula (4):
wherein: d is the diameter of the propeller;
step four: cavitation bubblesVelocity of onset of motion V si And cavitation saturation navigational speed V sy The method comprises the following steps of:
wherein: sigma (sigma) i Constant, ω is the wake fraction;
as the continuous spectrum is mainly propeller noise, the spectrum peak and spectrum level are affected by cavitation degree, when the target navigational speed is smaller than cavitation primary speed V si When the propeller does not cavitation, the spectrum peak frequency and spectrum level change are not great; when the target navigational speed is greater than the cavitation saturated navigational speed V sy When the propeller is saturated and cavitated, the spectrum peak frequency and spectrum level are not greatly changed; based on this, the first and second light sources,
combined cavitation bubble generation velocity V si And cavitation saturation navigational speed V sy Determining min (max (V, V) si ),V sy ) Then according to min (max (V, V) si ),V sy ) Determining spectral peak frequency f 0 ；
Step five: extracting continuous spectrum attenuation law by actual marine noise data, wherein the spectrum peak spectrum level is related to the diameter of a propeller, the linear speed of propeller movement, the static pressure and the spectrum peak frequency, then obtaining the spectrum level at 5kHz according to an empirical formula, and obtaining the spectrum peak frequency f by utilizing the spectrum level at 5kHz and the extracted continuous spectrum attenuation law 0 Spectral peak spectral level SL at f0 ；
By means of spectral peak frequency f 0 Spectral peak spectral level SL f0 Continuous spectrum attenuation law C (t) simulates continuous spectrum characteristic S conti (t)；
The continuous spectrum simulation of the ship noise can be directly realized by establishing a filter with the same continuous spectrum frequency response as the ship noise and passing Gaussian white noise through the filter;
the noise line spectrum of the ship mainly comprises three types of mechanical noise line spectrum, propeller blade high-efficiency line spectrum and modulation noise line spectrum, so that the following assumption can be made on a noise line spectrum basic model of the ship:
step one, six: suppose that emissions are those using high frequency mechanical microwave noiseThe number of signal source phases, the fundamental frequency phase frequency, the frequency multiplication phase number, the frequency multiplication frequency distribution and the frequency amplitude corresponding to the phases which can be generated in the line spectrum obey a certain statistical rule, and the mechanical noise line spectrum S mac (t) is represented by formula (6):
wherein j is an imaginary unit, n is the number of mechanical noise sources, and the mechanical noise sources are assumed to be subjected to uniform distribution; m is the harmonic frequency number of the corresponding noise source, and is assumed to obey the rice distribution; f (m, n) represents line spectrum frequency constrained by fundamental frequency and harmonic frequency, and the fundamental frequency and frequency multiplication distribution can be assumed to be respectively subject to continuous uniform distribution and chi square distribution; a (m, n) 2 For the frequency amplitude of the corresponding noise source, since the frequency amplitude is reduced along with the frequency multiplication number under the general condition, the chip square distribution can be assumed to be obeyed; t is the moment;
seventhly,: because the propeller blade speed line spectrum has close relation with the propeller rotating speed and the number of blades, the line spectrum is relatively stable, and is an important basis for target detection and identification; therefore, corresponding statistical rule assumptions can be made for the frequency multiplication number, frequency multiplication distribution and corresponding amplitude values, and the propeller blade velocity line spectrum S prop (t) is represented by formula (7):
wherein f (m) represents a line spectrum frequency constrained by the harmonic number m;
step one, eight: the propeller modulation line spectrum modulates the broadband continuous spectrum with the axial frequency of the blade and the resonance frequency thereof, shows obvious characteristics and rules, modulates the noise line spectrum S modu (t) is represented by formula (8):
step one, nine: because of the stability of the modulation spectrum, the harmonic frequency m and the frequency multiplication distribution can be assumed to obey uniform distribution, and the corresponding amplitude is larger because of the integral multiple of the blade number amplitude, and the integral multiple of the frequency amplitude can be assumed to be increased by a multiple on the basis of chi square distribution;
joint simulation continuum S conti (t) obtaining a formula for simulating ship noise data, as shown in formula (9):
S(t)＝[1+S modu (t)]×S conti (t)+S mac (t)+S prop (t) (9)
step one, ten: and obtaining a waveform diagram through the formula simulation of the simulated ship noise data, and generating a ship noise data wav file through the waveform diagram.
Other steps and parameters are the same as in the first or second embodiment.
The fourth embodiment is different from one of the first to third embodiments in that: in the fifth step, continuous spectrum attenuation law is extracted by actual marine noise data, and the specific process is as follows:
and extracting continuous spectrum attenuation curves from the actual marine noise data through a linear phase FIR low-pass filter to obtain continuous spectrum attenuation rules.
A large number of data observations indicate that the continuous spectrum peak frequencies are typically between 100 and 1000 Hz. Assuming that the target is before cavitation bubble initiation (i.e., V.ltoreq.V si ) Decays with a law of-10 dB/oct, and when cavitation is saturated (i.e. V is more than or equal to V sy ) Attenuation is carried out according to a law of-6 dB/oct; evolution region (i.e. V si ≤V≤V sy ) At f>At 1kHz, attenuation is regular at-2.1 dB/oct (i.e., q=0.7), f 0 <f<Approximately q=1.2N/N at 1000Hz si -0.4, wherein q is the decay index, N/N si To evaluate the magnitude of the degree of cavitation of the propeller. The rising edges then increase with 3, 3q, 6dB/oct, respectively. From this, a continuous spectrum decay law curve C (t) is derived.
Other steps and parameters are the same as in one to three embodiments.
The fifth embodiment is different from one to four embodiments in that: the specific process of the second step is as follows:
step two,: after the sound files in the original data set are put into simulation software, visual data corresponding to the sound files are obtained; sampling visual data corresponding to the sound file to obtain sampled data corresponding to the sound file;
observing the corresponding characteristics of the sound file by using the visualized data, wherein the sampling rate is fs;
step two: converting sampled data corresponding to the sound file into a waveform chart;
step two, three: and (3) performing short-time Fourier transform on the waveform map obtained in the second step to obtain the waveform map of the sound file.
In processing signals, fourier transforms can reveal a dominant characteristic of the signal (i.e., their frequencies and components); for a vector x containing n uniformly sampled points, its fourier transform formula is defined as equation (10):
wherein: ω=e -2πi/n Is one of n complex unit roots, i is an imaginary unit; for x and y, the indices j and k range from 0 to n-1;
other steps and parameters are the same as in one to four embodiments.
The sixth embodiment is different from one of the first to fifth embodiments in that: the waveform spectrum of the sound file is a LOFAR spectrum.
Other steps and parameters are the same as in one of the first to fifth embodiments.
The seventh embodiment is different from one of the first to sixth embodiments in that: the loss function of the generated countermeasure network is as follows:
wherein E (x) represents the expected value of the distribution function, x is the real sample, p data (x) Representing true senseDistribution of real samples, z is noise, p noise (z) is a noise distribution defined in a low dimension, D (x) represents a discriminator, D (G (z)) represents a generator, log D (x) is a probability that the discriminator decides that real data is real data, log (1-D (G (z))) is a probability that the discriminator still decides that false data generated by the generator is false data, and V (D, G) represents an optimized objective function of GAN.
Other steps and parameters are the same as in one of the first to sixth embodiments.
The working process of the generated countermeasure network is as follows:
step three: the GAN algorithm mainly comprises 4 parts, namely a part (data) for reading pictures and processing data, a part (net) for constructing a generator and a discriminator, a part (train) for training a data set and a data expansion part (do_expansion) after training is finished finally;
step three, two: reading the picture by using an OpenCV related algorithm;
and step three: after the data is read, converting the data into an array form so as to facilitate the subsequent operation, and recovering the read data into original pixels again so as to facilitate the subsequent reading of the generated picture by a subsequent generator and a discriminator;
and step three, four: reading the pictures into a generator and a discriminator, training the pictures, and defining the pixel types of the images to be 128 x 3 because a LOFAR map is selected, wherein 128 x 128 is the pixel size of the pictures, and 3 is that channels represent color pictures;
step three, five: connecting the generator and the discriminator together to form a code of a complete GAN network structure part;
step III: training the obtained data, wherein the grouping batches are required to form groups by multiples of 2, and the training data are divided into 8 groups; the training times are selected, and the training times can be used for performing the comparison of the training after the training, and the quality of the finally trained image can be influenced, so that the difference of the finally trained image can be observed in the aspects of data analysis and comparison analysis;
step three, pseudo-ginseng: in the aspect of data expansion, the variable requirements of the latent space are consistent with those of the part in training, 100 is adopted as the number of the finally required pictures in the experiment, the quantity of samples and the basic requirements of contrast data are ensured, and the required pictures can be generated according to the actual requirements in the actual process;
step three, eight: finally, the training quality of the training data set is compared.
The eighth embodiment is different from one of the first to seventh embodiments in that: the spectrum level at 5kHz is obtained according to an empirical formula; the specific process is as follows:
when no cavitation exists:
SL 5kHz ＝SL 0 +40log(U/10) (12)
when cavitation exists:
SL 5kHz ＝SL 0 +40log(U/10)+25+δSL (13)
δSL＝13.75N/N si -16.5 (14)
wherein SL is provided with 5kH At spectrum level of 5kHz, SL 0 Represents the spectral level of the radiated noise, delta SL is an intermediate variable, N/N si To evaluate the magnitude of the degree of cavitation of the propeller, U represents the speed of the blade tip of the propeller, u=pi nD; in the present invention, log defaults to 10.
Other steps and parameters are the same as those of one of the first to seventh embodiments.
The ninth embodiment is different from one of the first to eighth embodiments in that: the radiated noise spectrum level SL 0 The method comprises the following steps:
SL 0 ＝72.7+1.55D-0.1D 2 (15)。
other steps and parameters are the same as in one to eight of the embodiments.
Referring to fig. 1, a waveform of the fitting noise of the ocean and the ship obtained by the formula simulation is shown.
The original data set is obtained by some sound files, so that the sound files are needed to be put into simulation software to observe values, then the data of the sound files are obtained, and the data are converted into waveform patterns to observe corresponding data.
Referring to fig. 2 (a) to 2 (d) for classification of four waveform patterns, 4 patterns are better in terms of processing underwater acoustic signals, namely an Audio pattern, a deman pattern, a lover pattern and a Histogram pattern, all of which can embody some characteristics and frequency bands of ship noise, and the 4 patterns are subjected to availability analysis:
each kind of atlas has its own corresponding characteristic part, but not every kind of atlas is suitable for being used as the atlas of final GAN technology expansion, because the complexity of the picture itself is different, the part available for observation is also different, like the Audio oscillogram and the Histogram oscillogram have characteristics in observation, the characteristics can be clearly observed, but the two oscillograms have more data and too dense waveforms because of the picture itself, the picture structure is complex, and the quality and effect of the effect after expanding the data may be greatly compromised. Although the DEMON spectrum can also read the corresponding characteristic frequency band in the data band of the spectrum itself, because the data size of the DEMON spectrum is scarce, a clear spectrum and frequency band can not be generated in the GAN technology through an countermeasure network. The LOFAR spectrum is huge in quantity and clear in characteristics, and only a line with a characteristic frequency segment needs to be found to know whether the line is the characteristic frequency segment with sound or not, so that the LOFAR spectrum is more suitable to be used as a data set which is expanded finally. In order to better compare the quality, definition and characteristics of the corresponding pictures, other maps can also be expanded with corresponding data sets.
Referring to fig. 3, a flowchart of the GAN algorithm is shown. As shown in the figure, after a real sample is placed in the discrimination network D, whether the picture is correct or not is judged first, then the picture is trained to generate a fake picture, the fake picture enters the generation network G to be judged, if the picture is judged to be the fake picture by the discrimination network, the fake picture is cut off, if the picture is judged to be the real picture, the judgment is continued in the discrimination network, and the process is circulated until the picture which is similar to or even the same as the real picture is obtained.
The GAN algorithm is characterized by the following two main points:
(1) Compared with the characteristics of some traditional networks, the GAN algorithm has two different networks, namely a generation network G and a discrimination network D, and the two networks can not perform single training any more, and the countermeasure mode is a countermeasure generation training mode.
(2) Most of the information of the G generation network in GAN to generate the picture comes from the discrimination network D, not from the original dataset, which is also a feature of GAN technology.
The GAN algorithm has the following advantages:
(1) GAN is a model of the generation class, using only some back-propagation knowledge, and no markov chain, compared to the original ones.
(2) GAN can generate more realistic and clear samples, ensuring the sharpness of the desired data.
(3) GAN employs an unsupervised learning training approach, which can be more widely applied to unsupervised learning and semi-supervised learning training approaches.
(4) The GAN can save many troubles such as super resolution, denoising and the like which cannot be finished at ordinary times only by a discriminator.
(5) The GAN can generate perfect and clear pictures or audio with high quality on the premise of good training of a trainer.
The above examples of the present invention are only for describing the calculation model and calculation flow of the present invention in detail, and are not limiting of the embodiments of the present invention. Other variations and modifications of the above description will be apparent to those of ordinary skill in the art, and it is not intended to be exhaustive of all embodiments, all of which are within the scope of the invention.
Claims (7)
1. The method for constructing the ship radiation noise data set is characterized by comprising the following steps of:
firstly, establishing a mathematical model of actual marine noise data, simulating the simulated marine noise data by using the established mathematical model, and taking a set of the actual marine noise data and the simulated marine noise data as an original data set;
the mathematical model of the actual marine noise data is as follows:
S(t)＝1+[(t)]×S x (t)+S l (t) (1)
wherein: s (t) is actual marine noise data, G (t) is modulation signal, S x (t) time domain waveform of continuous spectrum component, S l (t) is a time domain waveform of the line spectral component;
the method utilizes the established mathematical model to simulate the noise data of the simulated ship, and comprises the following specific processes:
the method comprises the following steps: spectral peak frequency f 0 The empirical formula of the relation with the target working condition is:
wherein B represents the number of propeller blades, sigma represents the number of cavitation bubbles, and P 0 Represents hydrostatic pressure, v represents density;
step two: hydrostatic pressure P 0 Invariable, the density ρ is:
wherein v is the target speed, H is the depth, J p For the running speed coefficient;
step one, three: substituting formula (3) into formula (2) to obtain formula (4):
wherein: d is the diameter of the propeller;
step four: cavitation bubble primary velocity V si And cavitation saturation navigational speed V sy The method comprises the following steps of:
wherein: sigma (sigma) i Constant, ω is the wake fraction;
combined cavitation bubble generation velocity V si And cavitation saturation navigational speed V sy Determining min (max (V, V) si ),V sy ) Then according to min (max (V, V) si ),V sy ) Determining spectral peak frequency f 0 ；
Step five: extracting continuous spectrum attenuation law by actual marine noise data, then solving a spectrum level at 5kHz according to an empirical formula, and solving a spectrum peak frequency f by utilizing the spectrum level at 5kHz and the extracted continuous spectrum attenuation law 0 Spectral peak spectral level at
By means of spectral peak frequency f 0 Spectral peak spectral levelContinuous spectrum attenuation law C (t) simulates continuous spectrum characteristic S conti (t)；
Step one, six: mechanical noise line spectrum S mac (t) is represented by formula (6):
wherein j is an imaginary unit, and n is the number of mechanical noise sources; m is the harmonic frequency number of the corresponding noise source; f (m, n) represents line spectral frequencies constrained by fundamental and harmonic frequencies; a (m, n) 2 Frequency amplitude of the corresponding noise source; t is the moment;
seventhly,: propeller blade velocity profile S prop (t) is represented by formula (7):
wherein f (m) represents a line spectrum frequency constrained by the harmonic number m;
step one, eight: modulation noise line spectrum S modu (t) is represented by formula (8):
step one, nine: joint simulation continuum S conti (t) obtaining a formula for simulating ship noise data, as shown in formula (9):
S(t)＝[1+S modu (t)]×S conti (t)+S mac (t)+S prop (t) (9)
step one, ten: obtaining a waveform diagram through formula simulation of the simulated ship noise data, and generating a ship noise data wav file through the waveform diagram;
step two: the method comprises the steps of placing sound files in an original data set into simulation software to observe values, obtaining data of the sound files in the original data set, and converting the data of the sound files to obtain waveform patterns of the sound files;
step three: expanding the waveform spectrum obtained in the second step by using a generated countermeasure network to obtain an expanded waveform spectrum;
step four: and obtaining an expanded data set according to the expanded waveform map, and taking the expanded data set and the set of the original data set as a ship radiation noise data set.
2. The method for constructing a ship radiation noise data set according to claim 1, wherein in the fifth step, a continuous spectrum attenuation rule is extracted by using ocean noise data, and the specific process is as follows:
and extracting continuous spectrum attenuation curves from the actual marine noise data through a linear phase FIR low-pass filter to obtain continuous spectrum attenuation rules.
3. The method for constructing a ship radiation noise data set according to claim 2, wherein the specific process of the second step is as follows:
step two,: after the sound files in the original data set are put into simulation software, visual data corresponding to the sound files are obtained; sampling visual data corresponding to the sound file to obtain sampled data corresponding to the sound file;
step two: converting sampled data corresponding to the sound file into a waveform chart;
step two, three: and (3) performing short-time Fourier transform on the waveform map obtained in the second step to obtain the waveform map of the sound file.
4. A method of constructing a ship radiation noise data set according to claim 3, wherein the waveform profile of the sound file is a LOFAR profile.
5. The method of constructing a ship radiated noise dataset of claim 4, wherein the generated antagonism network's loss function is:
wherein E (x) represents the expected value of the distribution function, x is the real sample, p data (x) Representing the distribution of real samples, z being noise, p noise (z) is a noise distribution, D (x) represents a discriminator, D (G (z)) represents a generator, log D (x) is a probability that the discriminator decides that real data is real data, log (1-D (G (z))) is a probability that the discriminator decides that false data generated by the generator is still false data, and V (D, G) represents an optimized objective function of GAN.
6. The method for constructing a ship radiation noise data set according to claim 5, wherein the spectrum level at 5kHz is obtained according to an empirical formula; the specific process is as follows:
when no cavitation exists:
SL 5kHz ＝SL o +40log(U/10) (12)
when cavitation exists:
SL 5kHz ＝SL 0 +40log(U/10)+25+δSL (13)
δSL＝13.75N/N si -16.5 (14)
wherein SL is provided with 5kHz At spectrum level of 5kHz, SL 0 Represents the spectral level of the radiated noise, delta SL is an intermediate variable, N/N si To evaluate the magnitude of the degree of cavitation of the propeller, U represents the propeller blade tip speed.
7. The method for constructing a ship radiation noise data set according to claim 6, wherein the radiation noise spectrum level SL 0 The method comprises the following steps:
SL 0 ＝72.7+1.55D-0.1D 2 (15)。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110975058.8A CN113642245B (en) | 2021-08-24 | 2021-08-24 | Construction method of ship radiation noise data set |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110975058.8A CN113642245B (en) | 2021-08-24 | 2021-08-24 | Construction method of ship radiation noise data set |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113642245A CN113642245A (en) | 2021-11-12 |
| CN113642245B true CN113642245B (en) | 2024-03-19 |
Family
ID=78423585
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110975058.8A Active CN113642245B (en) | 2021-08-24 | 2021-08-24 | Construction method of ship radiation noise data set |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113642245B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115081510B (en) * | 2022-05-10 | 2024-05-10 | 哈尔滨工程大学 | Cross-modal high-concealment countermeasure sample generation method and system for underwater sound intelligent camouflage |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8111581B1 (en) * | 1985-08-09 | 2012-02-07 | Qinetiq Limited | Monitoring system for a ship'S radiated noise |
| CN102645265A (en) * | 2012-03-21 | 2012-08-22 | 西北工业大学 | Ship radiated noise level measuring method based on virtual time reversal mirror |
| CN110081964A (en) * | 2019-04-25 | 2019-08-02 | 西北工业大学 | Underwater sound source position and power spectrum combined estimation method based on sparse spectrum fitting |
| 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 |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200335086A1 (en) * | 2019-04-19 | 2020-10-22 | Behavioral Signal Technologies, Inc. | Speech data augmentation |
-
2021
- 2021-08-24 CN CN202110975058.8A patent/CN113642245B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8111581B1 (en) * | 1985-08-09 | 2012-02-07 | Qinetiq Limited | Monitoring system for a ship'S radiated noise |
| CN102645265A (en) * | 2012-03-21 | 2012-08-22 | 西北工业大学 | Ship radiated noise level measuring method based on virtual time reversal mirror |
| CN110081964A (en) * | 2019-04-25 | 2019-08-02 | 西北工业大学 | Underwater sound source position and power spectrum combined estimation method based on sparse spectrum fitting |
| 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 |
Non-Patent Citations (1)
| Title |
|---|
| 不确定航迹自适应预测模型;崔亚奇;熊伟;何友;;航空学报(第05期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113642245A (en) | 2021-11-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107609488B (en) | Ship noise identification and classification method based on deep convolutional network | |
| Liu et al. | Underwater target recognition using convolutional recurrent neural networks with 3-D Mel-spectrogram and data augmentation | |
| CN111178260B (en) | Modulation signal time-frequency diagram classification system based on generation countermeasure network and operation method thereof | |
| CN112230205B (en) | Performance evaluation method of underwater target recognition system by utilizing ship radiation noise simulation signals | |
| CN106529428A (en) | Underwater target recognition method based on deep learning | |
| Ren et al. | UALF: A learnable front-end for intelligent underwater acoustic classification system | |
| CN111724770B (en) | Audio keyword identification method for generating confrontation network based on deep convolution | |
| CN103323821B (en) | Water surface target radiation noise simulation method | |
| CN111931820B (en) | Method for extracting line spectrum of target radiation noise LOFAR spectrogram in water based on convolution residual error network | |
| CN110501683B (en) | Sea-land clutter classification method based on four-dimensional data characteristics | |
| AU2021101336A4 (en) | A Classification System Of Modulation Signal Time-Frequency Image Based On Generative Adversarial Network And Its Operation Method | |
| CN112163450A (en) | Based on S3High-frequency ground wave radar ship target detection method based on D learning algorithm | |
| CN113642245B (en) | Construction method of ship radiation noise data set | |
| CN111695461A (en) | Radar clutter intelligent classification method based on image frequency characteristics | |
| CN112766381B (en) | Attribute-guided SAR image generation method under limited sample | |
| Rosier et al. | Predicting ocean-induced ice-shelf melt rates using deep learning | |
| Qi et al. | Underwater acoustic target recognition with fusion feature | |
| CN109632071A (en) | A kind of method and device generating Underwater Acoustic Environment noise data based on time-frequency characteristics | |
| CN116430317A (en) | Radiation source modulation pattern and individual identification method and system | |
| CN115909040A (en) | Underwater sound target identification method based on self-adaptive multi-feature fusion model | |
| CN112185396B (en) | Offshore wind farm biological monitoring method and system based on passive acoustics | |
| CN112034512A (en) | Seismic data discontinuity detection method and system based on deep learning model | |
| Yu et al. | Simulated Sonar Image Generation Method Based on Improved CycleGAN | |
| CN117951504B (en) | Method and system for extracting spectral lines in time-frequency spectrogram by utilizing hough transformation | |
| CN117648851A (en) | Sonar data simulation generation method based on generation countermeasure 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |