CN114764579A - Denoising feature extraction method for ship radiation noise signal - Google Patents
Denoising feature extraction method for ship radiation noise signal Download PDFInfo
- Publication number
- CN114764579A CN114764579A CN202210520175.XA CN202210520175A CN114764579A CN 114764579 A CN114764579 A CN 114764579A CN 202210520175 A CN202210520175 A CN 202210520175A CN 114764579 A CN114764579 A CN 114764579A
- Authority
- CN
- China
- Prior art keywords
- denoising
- radiation noise
- noise signal
- signal
- mfcc
- 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
- 230000005855 radiation Effects 0.000 title claims abstract description 136
- 238000000605 extraction Methods 0.000 title claims abstract description 10
- 239000013598 vector Substances 0.000 claims abstract description 49
- 230000002457 bidirectional effect Effects 0.000 claims abstract description 26
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 13
- 238000000034 method Methods 0.000 claims abstract description 12
- 230000000694 effects Effects 0.000 claims abstract description 9
- 238000010220 Pearson correlation analysis Methods 0.000 claims abstract description 7
- 230000002776 aggregation Effects 0.000 claims abstract description 5
- 238000004220 aggregation Methods 0.000 claims abstract description 5
- 238000012549 training Methods 0.000 claims description 21
- 238000012360 testing method Methods 0.000 claims description 16
- 238000012706 support-vector machine Methods 0.000 claims description 5
- 238000012935 Averaging Methods 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 4
- 239000000284 extract Substances 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 42
- 230000009286 beneficial effect Effects 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000007429 general method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
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
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides a denoising feature extraction method of ship radiation noise signals, which is used for acquiring and recording the radiation noise signals of different classes of ships to obtain a noise signal sample set; performing modal decomposition by adopting a complete noise-assisted aggregation empirical mode decomposition algorithm to obtain a plurality of inherent modal function components, and performing Pearson correlation analysis to obtain a correlation coefficient; and obtaining a radiation noise signal sample set, extracting Mel frequency cepstrum coefficient characteristics to obtain a denoised signal MFCC characteristic vector set, constructing a bidirectional denoise self-encoder network, and obtaining identification accuracy so as to measure the effect of the extracted denoised characteristics. The method effectively extracts the denoising characteristic of the ship radiation noise signal, and improves the identification accuracy of the ship target. Compared with the traditional denoising self-encoder, the hidden layer denoising characteristic extracted by the bidirectional denoising self-encoder provided by the invention is more comprehensive and has better effect.
Description
Technical Field
The invention relates to the field of information signal processing, in particular to theories such as underwater sound signal processing, deep learning and the like, and particularly relates to a noise signal extraction method.
Background
Due to the influence of background noise of a complex marine environment, ship radiation noise signals received by the hydrophones often have low signal-to-noise ratio characteristics, and adverse influence is brought to the identification of ship targets. In view of this, at present, wavelet denoising, empirical mode decomposition, local projection, compressive sensing and other algorithms are mainly adopted to carry out research from the perspective of ship radiation noise signal denoising, and certain progress has been achieved. However, these methods also have respective problems. For example, wavelet de-noising requires efficient selection of wavelet bases and the number of decomposition layers; the empirical mode decomposition has the problem of mode aliasing; local projection denoising requires effective setting of parameters such as embedding dimension, delay time, neighborhood radius and the like to achieve the expected denoising effect; the compressed sensing denoising needs to strengthen the sparsity of signals, and the algorithm has high complexity and long operation time.
In recent years, deep learning methods typified by convolutional neural networks, cyclic neural networks, and autoencoders have been rapidly developed in the fields of image and speech recognition. The denoising autoencoder has a remarkable denoising effect in the fields of image denoising and voice denoising, but has no obvious breakthrough in the field of ship radiation noise signal denoising. Therefore, the invention expands the traditional denoising autoencoder to a bidirectional denoising autoencoder and provides a denoising feature extraction method of the ship radiation noise signal based on the bidirectional denoising autoencoder.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a method for extracting the denoising feature of a ship radiation noise signal.
Aiming at the problems that the ship radiation noise signal is interfered by the background noise of the marine environment to present low signal-to-noise ratio characteristic and is not beneficial to ship target identification, the invention expands the traditional denoising autoencoder to a bidirectional denoising autoencoder and provides a denoising feature extraction method of the ship radiation noise signal based on the bidirectional denoising autoencoder, and the extracted denoising feature can effectively improve the identification accuracy of the ship target.
Aiming at the problems that the ship radiation noise signal is interfered by the background noise of the marine environment, so that the ship radiation noise signal has low signal-to-noise ratio characteristic and is not beneficial to ship target identification, the invention expands the traditional denoising autoencoder to a bidirectional denoising autoencoder and provides a denoising characteristic extraction method of the ship radiation noise signal based on the bidirectional denoising autoencoder.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
step 1, distributing hydrophones in the sea, and respectively collecting and recording radiation noise signals of different classes of ships to obtain a multi-class ship radiation noise signal sample set of original radiation noise;
Step 2, performing modal decomposition on each radiation noise signal sample in the multi-class ship radiation noise signal sample set by adopting a complete noise-assisted aggregation empirical mode decomposition (CEEMDAN) algorithm, and obtaining a plurality of inherent modal function components for each radiation noise signal;
step 3, performing Pearson correlation analysis on each radiation noise signal and a plurality of corresponding inherent modal function components respectively to obtain a correlation coefficient between each inherent modal function component and the original radiation noise signal;
step 4, obtaining a multi-class ship radiation noise signal sample set corresponding to the multi-class ship radiation noise signal sample set in the step 1 according to the final de-noising signal in the step 3;
step 5, extracting the Mel Frequency Cepstrum Coefficient (MFCC) characteristics of each radiation noise signal sample in the multi-class ship radiation noise signal sample set in the step 1 to obtain an MFCC characteristic vector corresponding to each radiation noise signal sample, thereby obtaining an MFCC characteristic vector set of the original radiation noise signal; extracting the Mel Frequency Cepstrum Coefficient (MFCC) characteristics of each denoised signal sample in the multi-class ship radiation noise denoised signal sample set in the step 4 to obtain an MFCC characteristic vector corresponding to each denoised signal sample, thereby obtaining a denoised signal MFCC characteristic vector set;
Step 6, respectively labeling the original radiation noise signal MFCC characteristic vector set and the de-noising signal MFCC characteristic vector set in the step 5 according to the target category; randomly taking one half of an original radiation noise signal MFCC feature vector set and one half of a corresponding denoising signal MFCC feature vector set as a training set, and taking the rest data in the original radiation noise signal MFCC feature vector set as a test set;
and 7, constructing a bidirectional denoising self-encoder network, which comprises the following steps:
and 8, taking the denoising features extracted by the bidirectional denoising autoencoder in the step 7 as the input of a support vector machine classifier to obtain the identification accuracy rate, thereby measuring the effect of the extracted denoising features.
The specific steps of the step 3 are as follows:
step 3, respectively carrying out Pearson correlation analysis on each radiation noise signal and a plurality of corresponding inherent modal function components to obtain a correlation coefficient between each inherent modal function component and the original radiation noise signal;
step 301, selecting a first threshold of a correlation coefficient, ignoring a first inherent mode function component, only reserving inherent mode function components of which the correlation coefficient is greater than or equal to the first threshold, and adding the reserved inherent mode function components to obtain a first denoising signal corresponding to an original radiation noise signal;
Step 302, selecting a second threshold of the correlation coefficient, ignoring the first inherent mode function component, only reserving the inherent mode function components of which the correlation coefficient is greater than or equal to the second threshold, and adding the reserved inherent mode function components to obtain a second denoising signal corresponding to the original radiation noise signal;
step 303, selecting a third threshold of the correlation coefficient, ignoring the first inherent mode function component, only reserving the inherent mode function components of which the correlation coefficient is greater than or equal to the third threshold, and adding the reserved inherent mode function components to obtain a third denoising signal corresponding to the original radiation noise signal;
step 304, averaging the first denoising signal, the second denoising signal and the third denoising signal to obtain an average denoising signal, and taking the average denoising signal as a final denoising signal corresponding to the original radiation noise signal.
The steps of constructing the bidirectional denoising self-encoder network are as follows:
step 701, constructing two denoising autoencoder networks, wherein the two networks have the same structure and comprise an input layer, a hidden layer and an output layer which are respectively marked as a denoising autoencoder I and a denoising autoencoder II;
step 702, respectively taking the original radiation noise signal MFCC eigenvector in the training set in the step 6 and the corresponding denoised signal MFCC eigenvector as the input and the label of the denoised self-encoder I; respectively taking the original radiation noise signal MFCC eigenvector in the training set in the step 6 and the corresponding denoised signal MFCC eigenvector as a label and an input of a denoised self-encoder II; simultaneously training a first denoising autoencoder and a second denoising autoencoder on the training set in the step 6;
And 703, simultaneously testing the two denoising autocoders trained in the step 702 on the original radiation noise signal MFCC characteristic vector test set in the step 6, namely simultaneously taking the original radiation noise signal MFCC characteristic vectors in the test set in the step 6 as the input of the denoising autocoder I and the denoising autocoder II, merging the hidden layers of the two denoising autocoders, and extracting the merged hidden layer representation, namely the denoising characteristic extracted by the bidirectional denoising autocoder.
The first threshold value is 0.1.
The second threshold is
maxcorr is the maximum value among the correlation coefficients of each radiated noise signal with all the corresponding natural mode function components.
The third threshold value is 0.3.
Compared with the traditional ship radiation noise signal denoising method, the ship radiation noise signal denoising feature extraction method based on the bidirectional denoising autoencoder provided by the invention can effectively extract the denoising feature of the ship radiation noise signal and improve the identification accuracy of the ship target. Compared with the traditional denoising self-encoder, the hidden layer denoising characteristic extracted by the bidirectional denoising self-encoder provided by the invention is more comprehensive and has better effect.
Drawings
FIG. 1 is a general method block diagram of the present invention.
FIG. 2 is a block diagram of a bi-directional de-noising self-encoder of the present invention.
Detailed Description
The invention is further illustrated with reference to the following figures and examples.
The invention expands a traditional denoising autoencoder to a bidirectional denoising autoencoder and provides a denoising feature extraction method of a ship radiation noise signal based on the bidirectional denoising autoencoder, aiming at the problems that the ship radiation noise signal is interfered by ocean environment background noise, has low signal-to-noise ratio and is not beneficial to ship target identification. The method mainly comprises the following steps:
step 1, distributing hydrophones in the sea, and respectively collecting and recording radiation noise signals of different classes of ships to obtain a multi-class ship radiation noise signal sample set;
step 2, performing modal decomposition on each radiation noise signal sample in the multi-class ship radiation noise signal sample set by adopting a complete noise-assisted aggregation empirical mode decomposition (CEEMDAN) algorithm, and obtaining a plurality of inherent modal function components for each radiation noise signal;
step 3, performing Pearson correlation analysis on each radiation noise signal and a plurality of corresponding inherent modal function components respectively to obtain a correlation coefficient between each inherent modal function component and the original radiation noise signal;
Step 301, selecting 0.1 as a first threshold of a correlation coefficient, ignoring a first inherent modal function component, only reserving inherent modal function components of which the correlation coefficient is greater than or equal to the first threshold, and adding the reserved inherent modal function components to obtain a first denoising signal corresponding to an original radiation noise signal;
step 302, select
As a second threshold value of the correlation coefficient, maxcorr is the maximum value of the correlation coefficients of each radiation noise signal and all the corresponding natural mode function components, ignoring the first natural mode function component, only preserving the natural mode function components with the correlation coefficients larger than or equal to the second threshold value, and preserving the natural mode function componentsAdding to obtain a second denoising signal corresponding to the original radiation noise signal;
step 303, selecting 0.3 as a third threshold of the correlation coefficient, ignoring the first inherent modal function component, only reserving the inherent modal function components of which the correlation coefficient is greater than or equal to the third threshold, and adding the reserved inherent modal function components to obtain a third denoising signal corresponding to the original radiation noise signal;
step 304, averaging the first denoising signal, the second denoising signal and the third denoising signal to obtain an average denoising signal, and taking the average denoising signal as a final denoising signal corresponding to the primary radiation noise signal;
Step 4, obtaining a multi-class ship radiation noise denoising signal sample set corresponding to the multi-class ship radiation noise signal sample set in the step 1 according to the step 3;
step 5, extracting the Mel Frequency Cepstrum Coefficient (MFCC) characteristics of each radiation noise signal sample in the multi-class ship radiation noise signal sample set in the step 1 to obtain an MFCC characteristic vector corresponding to each radiation noise signal sample, thereby obtaining an original radiation noise signal MFCC characteristic vector set; extracting the Mel Frequency Cepstrum Coefficient (MFCC) characteristics of each denoised signal sample in the multi-class ship radiation noise denoised signal sample set in the step 4 to obtain an MFCC characteristic vector corresponding to each denoised signal sample, thereby obtaining a denoised signal MFCC characteristic vector set;
step 6, respectively labeling the original radiation noise signal MFCC feature vector set and the de-noising signal MFCC feature vector set in the step 5 according to the target category; randomly taking one half of an original radiation noise signal MFCC feature vector set and one half of a corresponding de-noising signal MFCC feature vector set as a training set, and taking the rest data in the original radiation noise signal MFCC feature vector set as a test set;
and 7, constructing a bidirectional denoising self-encoder network, and comprising the following steps:
Step 701, constructing two denoising autoencoder networks, wherein the two networks have the same structure and comprise an input layer, a hidden layer and an output layer which are respectively marked as a denoising autoencoder I and a denoising autoencoder II;
step 702, respectively taking the original radiation noise signal MFCC eigenvector in the training set in the step 6 and the corresponding denoised signal MFCC eigenvector as the input and the label of a first denoised self-encoder; respectively taking the original radiation noise signal MFCC characteristic vector in the training set in the step 6 and the corresponding denoised signal MFCC characteristic vector as a label and an input of a denoising autoencoder II; simultaneously training a denoising autoencoder I and a denoising autoencoder II on the training set in the step 6;
step 703, testing the two denoising autocoders trained in step 702 on the original radiation noise signal MFCC feature vector test set in step 6 at the same time, that is, taking the original radiation noise signal MFCC feature vectors in the test set in step 6 as the input of the denoising autocoder one and the denoising autocoder two at the same time, merging the hidden layers of the two denoising autocoders, and extracting the merged hidden layer representation, that is, the denoising feature extracted by the bidirectional denoising autocoder;
And 8, taking the denoising feature extracted by the bidirectional denoising autoencoder in the step 703 as the input of a support vector machine classifier to obtain the identification accuracy, thereby measuring the effect of the extracted denoising feature.
The present invention will be further described with reference to the following drawings and specific examples, which include, but are not limited to, the following examples.
Examples
With reference to fig. 1 and fig. 2, the method for extracting the denoising feature of the ship radiation noise signal comprises the following steps:
step 1, distributing hydrophones in the ocean, and respectively acquiring and recording radiation noise signals of six types of ships to obtain six types of ship radiation noise signal sample sets;
step 2, performing modal decomposition on each radiation noise signal sample in the six classes of ship radiation noise signal sample set by adopting a complete noise-assisted aggregation empirical mode decomposition (CEEMDAN) algorithm, and obtaining a plurality of inherent modal function components for each radiation noise signal;
step 3, performing Pearson correlation analysis on each radiation noise signal and a plurality of corresponding inherent modal function components respectively to obtain a correlation coefficient between each inherent modal function component and the original radiation noise signal;
Step 301, selecting 0.1 as a first threshold of a correlation coefficient, ignoring a first inherent mode function component, only reserving inherent mode function components of which the correlation coefficient is greater than or equal to the first threshold, and adding the reserved inherent mode function components to obtain a first denoising signal corresponding to an original radiation noise signal;
step 302, select
As a second threshold of the correlation coefficient, maxcorr is the maximum value of the correlation coefficient between each radiation noise signal and all the corresponding inherent modal function components, neglecting the first inherent modal function component, only reserving the inherent modal function components with the correlation coefficient more than or equal to the second threshold, and adding the reserved inherent modal function components to obtain a second denoising signal corresponding to the original radiation noise signal;
step 303, selecting 0.3 as a third threshold of the correlation coefficient, ignoring the first inherent mode function component, only reserving the inherent mode function components of which the correlation coefficient is greater than or equal to the third threshold, and adding the reserved inherent mode function components to obtain a third denoising signal corresponding to the original radiation noise signal;
step 304, averaging the first denoising signal, the second denoising signal and the third denoising signal to obtain an average denoising signal, and taking the average denoising signal as a final denoising signal corresponding to the original radiation noise signal;
Step 4, obtaining a six-class ship radiation noise-removing signal sample set corresponding to the six-class ship radiation noise signal sample set in the step 1 according to the step 3;
step 5, extracting the Mel Frequency Cepstrum Coefficient (MFCC) characteristics of each radiation noise signal sample in the six types of ship radiation noise signal sample sets in the step 1 to obtain an MFCC characteristic vector corresponding to each radiation noise signal sample, and selecting 20 dimensions of the characteristic vectors to obtain an original radiation noise signal MFCC characteristic vector set; extracting the Mel Frequency Cepstrum Coefficient (MFCC) characteristics of each denoised signal sample in the six-class ship radiation noise denoised signal sample set in the step 4 to obtain an MFCC characteristic vector corresponding to each denoised signal sample, and selecting 20 dimensions of the characteristic vectors to obtain a denoised signal MFCC characteristic vector set;
step 6, respectively labeling the original radiation noise signal MFCC characteristic vector set and the de-noising signal MFCC characteristic vector set in the step 5 according to the target category; randomly taking one half of an original radiation noise signal MFCC feature vector set and one half of a corresponding denoising signal MFCC feature vector set as a training set, and taking the rest data in the original radiation noise signal MFCC feature vector set as a test set;
And 7, constructing a bidirectional denoising self-encoder network, and comprising the following steps:
step 701, constructing two denoising self-encoder networks, wherein the two networks have the same structure and comprise an input layer, a hidden layer and an output layer, the number of neurons of the input layer is 20, the number of neurons of the hidden layer is 40, the number of neurons of the output layer is 20, and the two networks are respectively marked as a denoising self-encoder I and a denoising self-encoder II;
step 702, respectively taking the original radiation noise signal MFCC eigenvector in the training set in the step 6 and the corresponding denoised signal MFCC eigenvector as the input and the label of the denoised self-encoder I; respectively taking the original radiation noise signal MFCC eigenvector in the training set in the step 6 and the corresponding denoised signal MFCC eigenvector as a label and an input of a denoised self-encoder II; simultaneously training a denoising autoencoder I and a denoising autoencoder II on the training set in the step 6, setting the learning rate to be 0.001 during training, selecting Smooth L1 loss as a loss function, and optimizing by adopting an Adam algorithm;
step 703, testing the two denoising autocoders trained in step 702 on the original radiation noise signal MFCC feature vector test set in step 6 at the same time, that is, taking the original radiation noise signal MFCC feature vectors in the test set in step 6 as the input of the denoising autocoder one and the denoising autocoder two at the same time, merging the hidden layers of the two denoising autocoders, and extracting the merged hidden layer representation, that is, the denoising feature extracted by the bidirectional denoising autocoder;
And 8, taking the denoising features extracted by the bidirectional denoising self-encoder in the step 703 as the input of a support vector machine classifier, wherein 80% of data is used for fitting a support vector machine classifier model, and 20% of data is used for prediction to obtain the prediction accuracy, so that the effect of the extracted denoising features is measured.
The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and all the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the appended claims.
Claims (6)
1. A noise-removing feature extraction method for ship radiation noise signals is characterized by comprising the following steps:
step 1, distributing hydrophones in the sea, and respectively collecting and recording radiation noise signals of different classes of ships to obtain a multi-class ship radiation noise signal sample set of original radiation noise;
step 2, performing modal decomposition on each radiation noise signal sample in the multi-class ship radiation noise signal sample set by adopting a complete noise-assisted aggregation empirical mode decomposition (CEEMDAN) algorithm, and obtaining a plurality of inherent modal function components for each radiation noise signal;
Step 3, respectively carrying out Pearson correlation analysis on each radiation noise signal and a plurality of corresponding inherent modal function components to obtain a correlation coefficient between each inherent modal function component and the original radiation noise signal;
step 4, obtaining a multi-class ship radiation noise signal sample set corresponding to the multi-class ship radiation noise signal sample set in the step 1 according to the final de-noising signal in the step 3;
step 5, extracting the Mel Frequency Cepstrum Coefficient (MFCC) characteristics of each radiation noise signal sample in the multi-class ship radiation noise signal sample set in the step 1 to obtain an MFCC characteristic vector corresponding to each radiation noise signal sample, thereby obtaining an MFCC characteristic vector set of the original radiation noise signal; extracting the Mel Frequency Cepstrum Coefficient (MFCC) characteristics of each denoised signal sample in the multi-class ship radiation noise denoised signal sample set in the step 4 to obtain an MFCC characteristic vector corresponding to each denoised signal sample, thereby obtaining a denoised signal MFCC characteristic vector set;
step 6, respectively labeling the original radiation noise signal MFCC feature vector set and the de-noising signal MFCC feature vector set in the step 5 according to the target category; randomly taking one half of an original radiation noise signal MFCC feature vector set and one half of a corresponding denoising signal MFCC feature vector set as a training set, and taking the rest data in the original radiation noise signal MFCC feature vector set as a test set;
And 7, constructing a bidirectional denoising self-encoder network, and comprising the following steps:
and 8, taking the denoising features extracted by the bidirectional denoising self-encoder in the step 7 as the input of a support vector machine classifier to obtain the identification accuracy rate, thereby measuring the effect of the extracted denoising features.
2. The method of extracting denoising features of a ship radiation noise signal according to claim 1, wherein:
the specific steps of the step 3 are as follows:
step 3, respectively carrying out Pearson correlation analysis on each radiation noise signal and a plurality of corresponding inherent modal function components to obtain a correlation coefficient between each inherent modal function component and the original radiation noise signal;
step 301, selecting a first threshold of a correlation coefficient, ignoring a first inherent modal function component, only reserving inherent modal function components of which the correlation coefficient is greater than or equal to the first threshold, and adding the reserved inherent modal function components to obtain a first denoising signal corresponding to a primary radiation noise signal;
step 302, selecting a second threshold of the correlation coefficient, ignoring the first inherent mode function component, only reserving the inherent mode function components of which the correlation coefficient is greater than or equal to the second threshold, and adding the reserved inherent mode function components to obtain a second denoising signal corresponding to the original radiation noise signal;
Step 303, selecting a third threshold of the correlation coefficient, ignoring the first inherent mode function component, only reserving the inherent mode function components of which the correlation coefficient is greater than or equal to the third threshold, and adding the reserved inherent mode function components to obtain a third denoising signal corresponding to the original radiation noise signal;
step 304, averaging the first denoising signal, the second denoising signal and the third denoising signal to obtain an average denoising signal, and taking the average denoising signal as a final denoising signal corresponding to the primary radiation noise signal.
3. The method of extracting denoising features of a ship radiation noise signal according to claim 1, wherein:
the steps of constructing the bidirectional denoising self-encoder network are as follows:
step 701, constructing two denoising autoencoder networks, wherein the two networks have the same structure and comprise an input layer, a hidden layer and an output layer which are respectively marked as a denoising autoencoder I and a denoising autoencoder II;
step 702, respectively taking the original radiation noise signal MFCC eigenvector in the training set in the step 6 and the corresponding denoised signal MFCC eigenvector as the input and the label of the denoised self-encoder I; respectively taking the original radiation noise signal MFCC eigenvector in the training set in the step 6 and the corresponding denoised signal MFCC eigenvector as a label and an input of a denoised self-encoder II; simultaneously training a first denoising autoencoder and a second denoising autoencoder on the training set in the step 6;
And 703, simultaneously testing the two denoising autocoders trained in the step 702 on the original radiation noise signal MFCC characteristic vector test set in the step 6, namely simultaneously taking the original radiation noise signal MFCC characteristic vectors in the test set in the step 6 as the input of the denoising autocoder I and the denoising autocoder II, merging the hidden layers of the two denoising autocoders, and extracting the merged hidden layer representation, namely the denoising characteristic extracted by the bidirectional denoising autocoder.
4. The method for extracting denoising features of the ship radiation noise signal according to claim 1, wherein:
the first threshold value is 0.1.
5. The method of extracting denoising features of a ship radiation noise signal according to claim 1, wherein:
the second threshold is
maxcorr is the maximum value among the correlation coefficients of each radiated noise signal with all the corresponding natural mode function components.
6. The method of extracting denoising features of a ship radiation noise signal according to claim 1, wherein:
the third threshold value is 0.3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210520175.XA CN114764579A (en) | 2022-05-12 | 2022-05-12 | Denoising feature extraction method for ship radiation noise signal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210520175.XA CN114764579A (en) | 2022-05-12 | 2022-05-12 | Denoising feature extraction method for ship radiation noise signal |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114764579A true CN114764579A (en) | 2022-07-19 |
Family
ID=82364602
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210520175.XA Pending CN114764579A (en) | 2022-05-12 | 2022-05-12 | Denoising feature extraction method for ship radiation noise signal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114764579A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| WO2021128510A1 (en) * | 2019-12-27 | 2021-07-01 | 江苏科技大学 | Bearing defect identification method based on sdae and improved gwo-svm |
| CN113205050A (en) * | 2021-05-09 | 2021-08-03 | 西北工业大学 | Ship radiation noise line spectrum extraction method based on GRU-AE network |
-
2022
- 2022-05-12 CN CN202210520175.XA patent/CN114764579A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| WO2021128510A1 (en) * | 2019-12-27 | 2021-07-01 | 江苏科技大学 | Bearing defect identification method based on sdae and improved gwo-svm |
| CN113205050A (en) * | 2021-05-09 | 2021-08-03 | 西北工业大学 | Ship radiation noise line spectrum extraction method based on GRU-AE network |
Non-Patent Citations (1)
| Title |
|---|
| 李余兴;李亚安;陈晓;: "基于EEMD的舰船辐射噪声特征提取方法研究", 振动与冲击, no. 05, 15 March 2017 (2017-03-15) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110600017B (en) | Training method of voice processing model, voice recognition method, system and device | |
| CN108962237B (en) | Hybrid speech recognition method, device and computer readable storage medium | |
| Ibrahim et al. | Automatic classification of grouper species by their sounds using deep neural networks | |
| KR102605736B1 (en) | Method and apparatus of sound event detecting robust for frequency change | |
| CN109890043B (en) | Wireless signal noise reduction method based on generative countermeasure network | |
| CN112349297A (en) | Depression detection method based on microphone array | |
| CN109977724B (en) | Underwater target classification method | |
| Yang et al. | Underwater acoustic signal denoising model based on secondary variational mode decomposition | |
| CN108627241B (en) | Dolphin widescriptae click signal detection method based on Gaussian mixture model | |
| CN110929842B (en) | Accurate intelligent detection method for non-cooperative radio signal burst time region | |
| CN114239657A (en) | Time sequence signal identification method based on complex value interference neural network | |
| Yin et al. | Underwater acoustic target classification based on LOFAR spectrum and convolutional neural network | |
| Noiboar et al. | Anomaly detection based on wavelet domain GARCH random field modeling | |
| Zhang et al. | Birdsoundsdenoising: Deep visual audio denoising for bird sounds | |
| CN110136746B (en) | Method for identifying mobile phone source in additive noise environment based on fusion features | |
| CN114863938A (en) | Bird language identification method and system based on attention residual error and feature fusion | |
| CN112052880A (en) | Underwater sound target identification method based on weight updating support vector machine | |
| CN117076858A (en) | Deep learning-based low-frequency geomagnetic strong interference suppression method and system | |
| JP6404780B2 (en) | Wiener filter design apparatus, sound enhancement apparatus, acoustic feature quantity selection apparatus, method and program thereof | |
| CN114764579A (en) | Denoising feature extraction method for ship radiation noise signal | |
| CN113901878B (en) | Three-dimensional ground penetrating radar image underground pipeline identification method based on CNN+RNN algorithm | |
| CN115223033A (en) | Synthetic aperture sonar image target classification method and system | |
| Kalkan et al. | Online feature selection and classification | |
| CN114818789A (en) | Ship radiation noise identification method based on data enhancement | |
| CN114626412A (en) | Multi-class target identification method and system for unattended sensor system |
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 |