CN107369444A - A kind of underwater manoeuvre Small object recognition methods based on MFCC and artificial neural network - Google Patents
A kind of underwater manoeuvre Small object recognition methods based on MFCC and artificial neural network Download PDFInfo
- Publication number
- CN107369444A CN107369444A CN201610308112.2A CN201610308112A CN107369444A CN 107369444 A CN107369444 A CN 107369444A CN 201610308112 A CN201610308112 A CN 201610308112A CN 107369444 A CN107369444 A CN 107369444A
- Authority
- CN
- China
- Prior art keywords
- mrow
- mfcc
- characteristic quantities
- neural network
- small object
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 27
- 239000002131 composite material Substances 0.000 claims abstract description 27
- 239000013598 vector Substances 0.000 claims abstract description 21
- 210000005036 nerve Anatomy 0.000 claims abstract description 13
- 230000005236 sound signal Effects 0.000 claims abstract description 13
- 238000000605 extraction Methods 0.000 claims abstract description 12
- 239000000284 extract Substances 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 17
- 238000001228 spectrum Methods 0.000 claims description 10
- 230000003595 spectral effect Effects 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 5
- 238000001914 filtration Methods 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 238000000205 computational method Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 abstract description 6
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 6
- 241000238631 Hexapoda Species 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 230000007935 neutral effect Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 244000287680 Garcinia dulcis Species 0.000 description 2
- 241000124008 Mammalia Species 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000010365 information processing Effects 0.000 description 2
- 230000000873 masking effect Effects 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 241000545442 Radix Species 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000004090 dissolution Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000036039 immunity Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 239000004575 stone Substances 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/16—Speech classification or search using artificial neural networks
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
- G10L15/063—Training
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
- G10L21/0224—Processing in the time domain
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Quality & Reliability (AREA)
- Signal Processing (AREA)
- Evolutionary Computation (AREA)
- Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)
Abstract
A kind of underwater manoeuvre Small object recognition methods based on MFCC and artificial neural network, methods described include:The original sound signal s (n) of target to be identified is pre-processed, obtains the time-domain signal x (n) of each speech frame;Extract time-domain signal x (n) MFCC characteristic quantities, Further Feature Extraction, which is carried out, using MFCC characteristic quantities obtains difference MFCC characteristic quantities, the crest frequency of MFCC characteristic quantities, difference MFCC characteristic quantities and x (n) is combined, form MFCC composite character amounts, the MFCC composite characters vector is input to the artificial nerve network classifier trained to be identified, exports the type of identification.The apish auditory properties of MFCC composite characters of the method extraction of the present invention, the outstanding sound signal processing ability of human ear is effectively applied to the classification of underwater manoeuvre Small object;The interframe feature of signal is make use of simultaneously, is reduced error caused by ambient noise, is improved the discrimination of underwater manoeuvre Small object.
Description
Technical field
The present invention relates to the identification field of underwater Small object, and in particular to a kind of water based on MFCC and artificial neural network
Lower small maneuvering target recognition methods.
Background technology
Cold War period, various countries are in the war state of alert, and sea detection and the object defendd are mainly the warship of hostile nations
The large-scale target such as ship and submarine.After the end of cold war, particularly the Soviet Union's dissolution, Underwater Battery equipment minimizes rapid
Development, the technical equipment such as frogman, underwater research vehicle and underwater robot it is increasingly mature, this kind of target has good concealment, breaks
Bad power waits by force the attack pattern of obvious " asymmetric " advantage, it has also become terrorist carries out the important way of terrorist activity.
In recent years, wait the detection to the underwater Small object such as frogman and identification to carry out correlative study both at home and abroad, but be directed to water
The Classification and Identification of lower small maneuvering target is less.In Active Acoustic field of signal identification, the SQS-26 sonars for equipping USN have
Active target classification feature.In addition, the torpedo homing system of many foreign navies, has also had been provided with identification ship key now
The ability at position.Document [1] (Wu state, Li Jing, Li Xunhao, Chen Yaoming, Yuan Yi ship noises identification (III)-dual spectrum peace
The feature extraction of equal power spectrum and Prototype drawing acoustic journals, 1999,24 (2):191-196 pages;Wu Guoqing, Li Jing, Li Xunfa,
Chen Yaoming, Yuan Yi ship noises identify (IV)-fuzzy neural network acoustic journals, 1999,24 (3):275-280 pages) et al.
Substantial amounts of research has been carried out to the radiated noise on naval vessel, has been extracted numerous features of ship-radiated noise, and utilize statistical model
Ship noise is identified the mode that identification and fuzzy neural network are combined.Document [2] (LI Qihu, WANG Jinlin,
WEI Wei.An application of expert system in recognition of radiated noise of
under-water target.Beijing:Institute of Acoustics,Chinese Academy of
Sciences, 1989.404-408) Acoustic Object expert's identifying system is have developed, in most cases, when signal to noise ratio is not less than
During 3dB, the discrimination of signal has exceeded 75%.
In field of underwater acoustic signal processing, currently used method is that time-domain signal is transformed into time-frequency combination domain, when passing through
The non-stationary and frequency that frequency Joint Distribution discloses signal changes with time feature.(it is small that stone supermale is based on lifting to document [3]
Application [J] the acoustic techniques of the MFCC of wave conversion in target identification, 2014,33 (4):372-375.) propose using lifting
The MFCC of wavelet transformation method, to underwater passive sonar target classification and identification.Its emulation experiment shows, lifting wavelet transform
Method extraction MFCC has the advantages of discrimination is higher, preferable to noise robustness.(Zhu Leqing, a true are based on document [4]
MFCC and GMM insect sound automatic identification [J] insect journals, 2012,55 (4):466-471.) by MFCC and Gaussian Mixture
Model (GMM) is applied to the identification of insect in forest, is assessed, achieved in the Sample Storehouse comprising 58 kinds of insect sound
Higher correct recognition rata and comparatively ideal time performance.These researchs show that the method based on MFCC can be used for complex situations
Under voice signal property extraction identification.It is low to be currently based on the universal noise immunity of recognition methods of the MFCC to submarine target, more applications
In the higher situation of signal to noise ratio, therefore for the Small object of current signal to noise ratio is relatively low, plurality of target coexists complicated underwater environment
Discrimination is relatively low.
Because artificial neural network has very strong learning ability, adaptive ability, robustness and fault-tolerant ability, to information
Processing procedure closer to human brain thinking activities, the traditional algorithm of complicated and time consumption can be replaced.Utilize neutral net simultaneously
Highly-parallel arithmetic ability, it is difficult to other mathematical computations radixes realize Optimal Signals Processing Algorithm can be in real time inhale
Receive.Due to the These characteristics of artificial neural network, marine acoustics has gradually been obtained in Acoustic Object identification field artificial neural network
The concern and application of person.It is particularly suitable for as the relevant information processing of this kind of perception with people of underwater manoeuvre Small object.It can be with
By from sample learning, realizing self and the adjustment of network.After the feature for extracting signal, signal will be deployed to know
Do not work.Because underwater sound signal has very strong uneven stability, plus the complexity of underwater environment;The priori of other underwater sound signal
Mathematical knowledge is difficult to grasp, in field of underwater acoustic signal processing, it is difficult to carry out accurate description to it.So the classification of underwater sound signal is asked
Topic is a nonlinear problem, extremely complex.
The content of the invention
It is underwater it is an object of the invention to overcome current MFCC methods to be used for the complexity that signal to noise ratio is relatively low, plurality of target coexists
The problem of discrimination existing for the Small object identification of environment is relatively low, it is proposed that a kind of underwater based on MFCC and artificial neural network
Small maneuvering target recognition methods, this method extract the mixing MFCC characteristic quantities of target audio signal, including:Difference MFCC features and
MFCC features, then mixing MFCC characteristic quantities are identified using the BP artificial neural networks trained, so as to realize to mesh
Target is classified, and experiment shows that the side of the present invention has higher object recognition rate.
To achieve these goals, the present invention proposes the underwater Small object recognition methods based on MFCC and neutral net,
Composite character vector is built based on MFCC and Fourier transformation, then target is known using neural network classifier
Not;Methods described includes:The original sound signal s (n) of target to be identified is pre-processed, obtains the time domain of each speech frame
Signal x (n);Time-domain signal x (n) 2K MFCC characteristic quantity is extracted, 2K difference MFCC is obtained using 2K MFCC characteristic quantity
Characteristic quantity, the MFCC characteristic quantities of even number position, the difference MFCC characteristic quantities of even number position and crest frequency f are combined, shape
Into MFCC composite character amounts, the MFCC composite characters vector is input to the artificial nerve network classifier trained and known
Not, the type of identification is exported.
In above-mentioned technical proposal, methods described specifically includes:
Step 1) pre-processes to the original sound signal s (n) of target to be identified, obtains the time domain letter of each speech frame
Number x (n);
Step 2) calculates time-domain signal x (n) 2K MFCC coefficient, and the MFCC coefficients of even number position are formed into MFCC spies
Sign amount;
Step 3) carries out Further Feature Extraction to 2K MFCC coefficient, calculates 2K difference MFCC coefficient, and by even bit
The MFCC coefficients put form difference MFCC characteristic quantities;
Step 4) calculates time-domain signal x (n) crest frequency f;
MFCC characteristic quantities, difference MFCC characteristic quantities and crest frequency f are combined by step 5), form MFCC composite characters
Amount;
The MFCC composite characters vector is input to the artificial nerve network classifier trained and is identified by step 6),
Until identifying the type for successfully, exporting identification.
In above-mentioned technical proposal, the step 2) specifically includes:
Time-domain signal x (n) is obtained linear spectral X (k) by step 201) after discrete Fourier transform:
Wherein, N is the points for representing Fourier transformation;
Step 202) calculates linear spectral X (k) energy spectrum, and band logical filter is carried out to energy spectrum by 2K bandpass filter
Ripple;
Step 203) calculates the logarithmic energy that each bandpass filter group exports:
Step 204) carries out discrete cosine transform to the logarithmic energy S (m) of output and obtains MFCC coefficients:
Step 205) time-domain signal x (n) MFCC characteristic quantities are:[C(2),C(4),...C(2K)].
In above-mentioned technical proposal, the specific implementation process of the step 3) is:
The computational methods of difference MFCC coefficients are:
Wherein, k is constant, k=2;
Difference MFCC characteristic quantities are:[d(2),d(4),...d(2K)].
In above-mentioned technical proposal, the MFCC composite character amounts in the step 5) are:
[C(2),C(4),...,C(2K),d(2),d(4),...,d(2K),f]
Its dimension is 2K+1.
In above-mentioned technical proposal, also include before the step 6):Train artificial nerve network classifier;Specific bag
Include:
Step S1) training sample is classified;
Step S2) pretreatment is carried out to training sample form time-domain signal;
Step S3) using each training sample time-domain signal as x (t), according to step 1), step 2), step 3), step 4)
The MFCC composite characters vector of the time-domain signal of each training sample is built with step 5);
Step S4) using the MFCC composite characters vector of all training samples as the input of artificial neural network, train
Artificial nerve network classifier.
In above-mentioned technical proposal, the nodes of the input layer of the artificial nerve network classifier are 2K+1;Hidden layer section
Count as 2 (2K+1)+1;The node of output layer is the number of types of identification.
The advantage of the invention is that:
1st, the apish auditory properties of MFCC composite characters of method of the invention extraction, by the outstanding voice signal of human ear
Disposal ability is effectively applied to the classification of underwater manoeuvre Small object;The interframe feature of signal is make use of simultaneously, is reduced environment and is made an uproar
Error caused by sound;When carrying out target classification to MFCC composite characters using artificial neural network, ANN is made full use of
The learning ability and fault-tolerant ability of network;
2nd, recognition methods proposed by the present invention designs specially for the underwater environment that a variety of small maneuvering targets coexist, due to underwater ring
The complexity in border, but every kind of target has its distinctive frequency distribution, therefore frequency domain character is also a kind of important spy of target
Sign, frequency domain character addition MFCC characteristic vectors can be taken full advantage of into signal frequency domain feature.
Brief description of the drawings
Fig. 1 is the FB(flow block) of the method for the present invention;
Fig. 2 is the flow chart of extraction MFCC characteristic vectors.
Embodiment
The present invention is further detailed with specific embodiment below in conjunction with the accompanying drawings.
In order to describe acoustically to differentiate the impression of volume up-down so as to introduce the concept of tone in phonetics.It is low to frequency
Sound, people, which sound, feels that its tone is low, the sound high to frequency, and people, which sound, feels that its tone is high.But
The frequency of tone and tone color is not directly proportional relation.That is, frequency progress of the ear of people to sound is non-linear
Processing, i.e., have different perceptions to the voice of different frequency.In order to preferably describe tone, people make use of Mel frequencies
Rate scale is, it is specified that the unit of tone is Mel.Critical bandwidth is the important evidence of division Mel frequency scales.And critical broadband is drawn
Enter is to describe masking effect of the narrow-band noise to pure tone.When increasing the bandwidth of noise, this narrow-band noise is to pure tone
Masking amount is initially increase, but would not be increased when after a certain bandwidth, and this bandwidth is thus referred to as critical bandwidth.
Experiment shows:When centre frequency is in below 1kHz, the substantially linear distribution of critical bandwidth, about 100Hz;When centre frequency surpasses
When crossing 1kHz, with the increase of centre frequency, critical bandwidth is in logarithm increase.Therefore the frequency scale of the auditory system of people is met
Division should have higher frequency resolution in low frequency part, and have relatively low frequency resolution in HFS.Therefore
The present invention proposes a kind of composite character recognition methods based on MFCC.
As shown in figure 1, a kind of underwater manoeuvre Small object recognition methods based on MFCC composite characters and artificial neural network,
Methods described includes:
Step 1) gathers the original sound signal s (n) of target to be identified;Original sound signal s (n) passes through preemphasis, divided
The processing such as frame, adding window, obtains the time-domain signal x (n) of each speech frame;
Step 2) calculates time-domain signal x (n) 2K MFCC coefficient, and the MFCC coefficients of even number position are formed into MFCC spies
Sign amount;
As described in Figure 2, the step 2) specifically includes:
Time-domain signal x (n) is obtained linear spectral X (k) by step 201) after discrete Fourier transform (DFT):
Wherein, N is the points for representing Fourier transformation.
Step 202) calculates linear spectral X (k) energy spectrum, and band logical is carried out to energy spectrum by several bandpass filters
Filtering;
Square of linear spectral X (k) amplitude of calculating, i.e. energy spectrum;By the triangular filter group of one group of Mel yardstick,
Frequency domain carries out bandpass filtering to energy spectrum;Wherein Mel frequency filters group be set in the spectral range of voice several
Bandpass filter Hm(k), its centre frequency is that f (m), m=1,2 ... M, M are the number of bandpass filter, and M is even number;At this
In embodiment, M=24;Each bandpass filter has triangle filtering characteristic, and its transmission function is:
Wherein
Step 203) calculates the logarithmic energy that each bandpass filter group exports:
Step 204) carries out discrete cosine transform (DCT) to the logarithmic energy S (m) of output and obtains MFCC coefficients:
Step 205) time-domain signal x (n) MFCC characteristic quantities are:[C(2),C(4),...,C(M)];
In step 202), the number of bandpass filter is M, can obtain M MFCC coefficient, wherein selection even number number
MFCC coefficients form MFCC characteristic quantities;The MFCC coefficients of odd number number are used to construct follow-up MFCC characteristic quantities;
Step 3) carries out Further Feature Extraction to 2K MFCC coefficient, calculates 2K difference MFCC coefficient, and by even bit
The MFCC coefficients put form difference MFCC characteristic quantities;
Further Feature Extraction is that original sequence of feature vectors is analyzed again, by characteristic vector with weighting, it is poor
The methods of dividing, screening, further separate the target signature for being hidden in voice signal behind.Can to the first-order difference of characteristic vector
To obtain the pace of change of characteristic vector, the change of characteristic vector embodies the change of target sound signal.
The computational methods of difference MFCC coefficients are:
Wherein, k is constant, it is preferred that k=2;Obtained difference Mel cepstrum coefficient d (n) are front cross frame and rear two frame
Linear combination.
Difference MFCC characteristic quantities are:[d(2),d(4),...,d(M)].
Step 4) calculates crest frequency f;
MFCC characteristic quantities, MFCC Differential Characteristics amount and crest frequency are combined by step 5), form MFCC composite characters
Amount;
The radiated noise signals of underwater manoeuvre Small object have the characteristics of similar to voice signal, therefore use MFCC and (poor
The method for dividing MFCC to be combined, the behavioral characteristics (transition feature) of radiated noise are also taken in, and such echo signal is moved
State feature and static nature information form complementation, can effectively improve the performance of system.
Because underwater Small object radiated noise signals have the characteristics of different from pure voice signal, every kind of target radiated noise again
Signal has each independent frequency distribution again, therefore adds characteristic vector using the crest frequency of echo signal as one of feature.
Composite character amount is:[C (2), C (4) ..., C (M), d (2), d (4) ..., d (M), f], dimension 25, wherein envelope 12 are tieed up
MFCC characteristic quantities, 12 dimension difference MFCC amounts and crest frequency.
MFCC composite character vectors are input to several artificial nerve network classifiers trained and known by step 6)
Not, the type of identification is exported;
Also include before the step 6):Train artificial nerve network classifier;Specifically include:
Step S1) training sample is classified;
Step S2) pretreatment is carried out to training sample form time-domain signal;
Step S3) using each training sample time-domain signal as x (t), according to step 1), step 2), step 3), step 4)
The MFCC composite characters vector of the time-domain signal of each training sample is built with step 5);
Step S4) using the MFCC composite characters vector of all training samples as the input of artificial neural network, training of human
Artificial neural networks grader.
Due to being extracted the identification feature amount of one 25 dimension, the artificial nerve network classifier input layer contains 25 sections
Point.Hidden layer extracts feature from input layer, is mapped with neural fusion, its number of nodes drastically influence network performance.
Kolmogorov theorems are pointed out:For the artificial neural network of single hidden layer, if input layer number is N (input layers here
Nodes are that 25), then node in hidden layer can be taken as 2N+1, therefore the node in hidden layer of the present invention is 51.Output layer contains
Four nodes, for showing recognition result, to represent that underwater frogman, sounding mammal, underwater robot and the water surface are fast respectively
Ship.
Specifically training process is:According to the square error between the hope output of sample and reality output, gradient is utilized
Descent method, since output layer, layer-by-layer correction weight coefficient.
When the sample of a unknown classification is applied to input, the output of output node is investigated, and by this sample
Kind judging be classification corresponding to that maximum node of output valve.
Example:The underwater manoeuvre Small object radiated noise data of outfield experiments are extracted, first, use underwater frogman, sounding
Mammal, underwater robot, the data BP ANN identification model of water surface speedboat, then using Model Identification
Underwater manoeuvre Small object.It shown below is traditional MFCC characteristic recognition methods and mixing MFCC feature recognitions proposed by the present invention
The contrast of method, table 1 are the correct recognition rata contrasts of tertiary target:
Table 1
Underwater frogman | Underwater robot | Water surface speedboat | |
Traditional MFCC features | 85.8% | 79.0% | 84.5% |
Mix MFCC features | 91.2% | 92.4% | 90.4% |
The composite character recognition methods based on MFCC is effective it can be seen from the data processed result of table 1, system
Discrimination is significantly improved, and has reached more than 90% for the discrimination of experimental data;Grader based on neutral net
Classification to these targets is effective.Identifying system can be used for the Classification and Identification of underwater manoeuvre Small object.The present invention's
Threat of the method to dealing with underwater manoeuvre Small object is significant.
Claims (7)
1. a kind of underwater manoeuvre Small object recognition methods based on MFCC and artificial neural network, methods described include:Treat knowledge
The original sound signal s (n) of other target is pre-processed, and obtains the time-domain signal x (n) of each speech frame;Extract time-domain signal x
(n) MFCC characteristic quantities, carry out Further Feature Extraction using MFCC characteristic quantities and obtain difference MFCC characteristic quantities, by MFCC features
The crest frequency of amount, difference MFCC characteristic quantities and x (n) is combined, and forms MFCC composite character amounts, the MFCC is mixed special
Sign vector is input to the artificial nerve network classifier trained and is identified, and exports the type of identification.
2. the underwater manoeuvre Small object recognition methods according to claim 1 based on MFCC and artificial neural network, it is special
Sign is that methods described specifically includes:
Step 1) pre-processes to the original sound signal s (n) of target to be identified, obtains the time-domain signal x of each speech frame
(n);
Step 2) calculates time-domain signal x (n) 2K MFCC coefficient, and the MFCC coefficients of even number position are formed into MFCC features
Amount;
Step 3) carries out Further Feature Extraction to 2K MFCC coefficient, calculates 2K difference MFCC coefficient, and by even number position
MFCC coefficients form difference MFCC characteristic quantities;
Step 4) calculates time-domain signal x (n) crest frequency f;
MFCC characteristic quantities, difference MFCC characteristic quantities and crest frequency f are combined by step 5), form MFCC composite character amounts;
The MFCC composite characters vector is input to the artificial nerve network classifier trained and is identified by step 6), until
Identify the type for successfully, exporting identification.
3. the underwater manoeuvre Small object recognition methods according to claim 2 based on MFCC and artificial neural network, it is special
Sign is that the step 2) specifically includes:
Time-domain signal x (n) is obtained linear spectral X (k) by step 201) after discrete Fourier transform:
<mrow>
<mi>X</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>n</mi>
<mo>=</mo>
<mn>0</mn>
</mrow>
<mrow>
<mi>N</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</munderover>
<mi>x</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<mi>j</mi>
<mn>2</mn>
<mi>&pi;</mi>
<mi>n</mi>
<mi>k</mi>
<mo>/</mo>
<mi>N</mi>
</mrow>
</msup>
<mo>,</mo>
<mn>0</mn>
<mo>&le;</mo>
<mi>k</mi>
<mo>&le;</mo>
<mi>N</mi>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, N is the points for representing Fourier transformation;
Step 202) calculates linear spectral X (k) energy spectrum, and bandpass filtering is carried out to energy spectrum by 2K bandpass filter;
Step 203) calculates the logarithmic energy that each bandpass filter group exports:
<mrow>
<mi>S</mi>
<mrow>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>l</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>0</mn>
</mrow>
<mrow>
<mi>N</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</munderover>
<mo>|</mo>
<msub>
<mi>X</mi>
<mi>a</mi>
</msub>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
<msup>
<mo>|</mo>
<mn>2</mn>
</msup>
<msub>
<mi>H</mi>
<mi>m</mi>
</msub>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
<mo>)</mo>
<mo>,</mo>
<mn>0</mn>
<mo>&le;</mo>
<mi>m</mi>
<mo>&le;</mo>
<mn>2</mn>
<mi>K</mi>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
</mrow>
Step 204) carries out discrete cosine transform to the logarithmic energy S (m) of output and obtains MFCC coefficients:
<mrow>
<mi>C</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>m</mi>
<mo>=</mo>
<mn>0</mn>
</mrow>
<mrow>
<mi>N</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</munderover>
<mi>S</mi>
<mrow>
<mo>(</mo>
<mi>m</mi>
<mo>)</mo>
</mrow>
<mi>c</mi>
<mi>o</mi>
<mi>s</mi>
<mrow>
<mo>(</mo>
<mi>&pi;</mi>
<mi>n</mi>
<mo>(</mo>
<mrow>
<mi>m</mi>
<mo>-</mo>
<mn>0.5</mn>
</mrow>
<mo>)</mo>
<mo>/</mo>
<mi>M</mi>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mn>0</mn>
<mo>&le;</mo>
<mi>n</mi>
<mo><</mo>
<mn>2</mn>
<mi>K</mi>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
</mrow>
Step 205) time-domain signal x (n) MFCC characteristic quantities are:[C(2),C(4),...C(2K)].
4. the underwater manoeuvre Small object recognition methods according to claim 3 based on MFCC and artificial neural network, it is special
Sign is that the specific implementation process of the step 3) is:
The computational methods of difference MFCC coefficients are:
<mrow>
<mi>d</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<msqrt>
<mrow>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mo>-</mo>
<mi>k</mi>
</mrow>
<mi>k</mi>
</munderover>
<msup>
<mi>i</mi>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
</mfrac>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mo>-</mo>
<mi>k</mi>
</mrow>
<mi>k</mi>
</munderover>
<mi>C</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>+</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>4</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein, k is constant, k=2;
Difference MFCC characteristic quantities are:[d(2),d(4),...d(2K)].
5. the underwater manoeuvre Small object recognition methods according to claim 4 based on MFCC and artificial neural network, it is special
Sign is that the MFCC composite character amounts in the step 5) are:
[C(2),C(4),...,C(2K),d(2),d(4),...,d(2K),f]
Its dimension is 2K+1.
6. the underwater manoeuvre Small object recognition methods according to claim 2 based on MFCC and artificial neural network, it is special
Sign is, also includes before the step 6):Train artificial nerve network classifier;Specifically include:
Step S1) training sample is classified;
Step S2) pretreatment is carried out to training sample form time-domain signal;
Step S3) using each training sample time-domain signal as x (t), according to step 1), step 2), step 3), step 4) and step
The MFCC composite characters vector of the rapid time-domain signal for 5) building each training sample;
Step S4) using the MFCC composite characters vector of all training samples as the input of artificial neural network, train manually
Neural network classifier.
7. the underwater manoeuvre Small object recognition methods according to claim 6 based on MFCC and artificial neural network, it is special
Sign is that the nodes of the input layer of the artificial nerve network classifier are 2K+1;Node in hidden layer is 2 (2K+1)+1;
The node of output layer is the number of types of identification.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610308112.2A CN107369444A (en) | 2016-05-11 | 2016-05-11 | A kind of underwater manoeuvre Small object recognition methods based on MFCC and artificial neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610308112.2A CN107369444A (en) | 2016-05-11 | 2016-05-11 | A kind of underwater manoeuvre Small object recognition methods based on MFCC and artificial neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107369444A true CN107369444A (en) | 2017-11-21 |
Family
ID=60303500
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610308112.2A Pending CN107369444A (en) | 2016-05-11 | 2016-05-11 | A kind of underwater manoeuvre Small object recognition methods based on MFCC and artificial neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107369444A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109800700A (en) * | 2019-01-15 | 2019-05-24 | 哈尔滨工程大学 | A kind of underwater sound signal target classification identification method based on deep learning |
CN115910077A (en) * | 2022-11-15 | 2023-04-04 | 生态环境部南京环境科学研究所 | Insect identification method based on deep learning of sound |
CN117198331A (en) * | 2023-11-08 | 2023-12-08 | 东南大学 | Underwater target intelligent identification method and system based on logarithmic ratio adjustment |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2002091358A1 (en) * | 2001-05-08 | 2002-11-14 | Intel Corporation | Method and apparatus for rejection of speech recognition results in accordance with confidence level |
CN101976564A (en) * | 2010-10-15 | 2011-02-16 | 中国林业科学研究院森林生态环境与保护研究所 | Method for identifying insect voice |
CN102799892A (en) * | 2012-06-13 | 2012-11-28 | 东南大学 | Mel frequency cepstrum coefficient (MFCC) underwater target feature extraction and recognition method |
CN103325382A (en) * | 2013-06-07 | 2013-09-25 | 大连民族学院 | Method for automatically identifying Chinese national minority traditional instrument audio data |
-
2016
- 2016-05-11 CN CN201610308112.2A patent/CN107369444A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2002091358A1 (en) * | 2001-05-08 | 2002-11-14 | Intel Corporation | Method and apparatus for rejection of speech recognition results in accordance with confidence level |
CN101976564A (en) * | 2010-10-15 | 2011-02-16 | 中国林业科学研究院森林生态环境与保护研究所 | Method for identifying insect voice |
CN102799892A (en) * | 2012-06-13 | 2012-11-28 | 东南大学 | Mel frequency cepstrum coefficient (MFCC) underwater target feature extraction and recognition method |
CN103325382A (en) * | 2013-06-07 | 2013-09-25 | 大连民族学院 | Method for automatically identifying Chinese national minority traditional instrument audio data |
Non-Patent Citations (2)
Title |
---|
HONGJIAN SONG ET AL.: "《An Automatic Identification Algorithm of Yangtze Finless Porpoise》", 《ICSPCC2015》 * |
王巍巍: "《鱼类声信号的分析及特征提取研究》", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109800700A (en) * | 2019-01-15 | 2019-05-24 | 哈尔滨工程大学 | A kind of underwater sound signal target classification identification method based on deep learning |
CN109800700B (en) * | 2019-01-15 | 2022-08-02 | 哈尔滨工程大学 | Underwater acoustic signal target classification and identification method based on deep learning |
CN115910077A (en) * | 2022-11-15 | 2023-04-04 | 生态环境部南京环境科学研究所 | Insect identification method based on deep learning of sound |
CN117198331A (en) * | 2023-11-08 | 2023-12-08 | 东南大学 | Underwater target intelligent identification method and system based on logarithmic ratio adjustment |
CN117198331B (en) * | 2023-11-08 | 2024-03-15 | 东南大学 | Underwater target intelligent identification method and system based on logarithmic ratio adjustment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105488466B (en) | A kind of deep-neural-network and Acoustic Object vocal print feature extracting method | |
CN109800700A (en) | A kind of underwater sound signal target classification identification method based on deep learning | |
CN107609488A (en) | A kind of ship noise method for identifying and classifying based on depth convolutional network | |
CN106529428A (en) | Underwater target recognition method based on deep learning | |
CN111723701B (en) | Underwater target identification method | |
CN104657948B (en) | A kind of denoising of Laser Underwater image and Enhancement Method for marine exploration | |
CN110456332A (en) | A kind of underwater sound signal Enhancement Method based on autocoder | |
Wang et al. | ia-PNCC: Noise Processing Method for Underwater Target Recognition Convolutional Neural Network. | |
CN105957537A (en) | Voice denoising method and system based on L1/2 sparse constraint convolution non-negative matrix decomposition | |
CN108630209A (en) | A kind of marine organisms recognition methods of feature based fusion and depth confidence network | |
Wang et al. | Feature extraction of underwater target in auditory sensation area based on MFCC | |
CN112017682B (en) | Single-channel voice simultaneous noise reduction and reverberation removal system | |
CN107368840A (en) | A kind of underwater manoeuvre Small object recognition methods based on MFCC and SVMs | |
CN107369444A (en) | A kind of underwater manoeuvre Small object recognition methods based on MFCC and artificial neural network | |
CN104732970A (en) | Ship radiation noise recognition method based on comprehensive features | |
CN113640768B (en) | Low-resolution radar target identification method based on wavelet transformation | |
CN108091345A (en) | A kind of ears speech separating method based on support vector machines | |
Shen et al. | Improved auditory inspired convolutional neural networks for ship type classification | |
CN112183582A (en) | Multi-feature fusion underwater target identification method | |
CN112652326B (en) | Ambient sound identification method based on pseudo-color time-frequency image and convolution network | |
CN107368841A (en) | A kind of underwater manoeuvre Small object recognition methods based on HHT and artificial neural network | |
CN112307926A (en) | Acoustic passive ship target classification method based on generation countermeasure network | |
Vahidpour et al. | An automated approach to passive sonar classification using binary image features | |
CN116992263A (en) | Method for extracting, identifying and separating self-noise characteristics of underwater vehicle based on multitask learning | |
Song et al. | Method of underwater acoustic signal denoising based on dual-path transformer 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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171121 |