CN111276154A - Wind noise suppression method and system and shot sound detection method and system - Google Patents
Wind noise suppression method and system and shot sound detection method and system Download PDFInfo
- Publication number
- CN111276154A CN111276154A CN202010119512.5A CN202010119512A CN111276154A CN 111276154 A CN111276154 A CN 111276154A CN 202010119512 A CN202010119512 A CN 202010119512A CN 111276154 A CN111276154 A CN 111276154A
- Authority
- CN
- China
- Prior art keywords
- signal
- wind noise
- decomposition
- noise
- mixed
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal 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
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal 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
- G10L21/0272—Voice signal separating
Abstract
The invention discloses a wind noise suppression method and system and a shot sound detection method and system, wherein the wind noise suppression method comprises the following steps: carrying out empirical mode decomposition on the collected wind noise signal, and then reconstructing the wind noise signal to obtain a reconstructed wind noise signal; performing convolution nonnegative matrix decomposition on the reconstructed wind noise signal and the mixed signal respectively; and constructing a signal-noise mixed separation model according to the decomposition result, and performing signal-noise separation on the mixed signals acquired in real time through the signal-noise mixed separation model to obtain pure target signals.
Description
Technical Field
The invention relates to a wind noise suppression method and system and a shot sound detection method and system, in particular to a wind noise suppression method and system and a shot sound detection method and system based on empirical mode decomposition and convolution nonnegative matrix joint characteristic decomposition.
Background
Artillery reconnaissance belongs to the subject of passive sound detection research, and a traditional artillery reconnaissance sound detection system positions an artillery position by using rifling waves generated when the artillery is launched. Gun projectiles generally generate rifling waves, ballistic waves and blast waves from firing to exploding; the time and space information of the rifling waves directly reflect the information of the gun position, so that the detection of the rifling waves is of great importance to the detection of the artillery. In the rifling wave detection of acoustic measurement, the convolution nonnegative matrix decomposition algorithm cannot inhibit wind noise, which is a difficult problem in the traditional technology. Under the conditions of single-channel measurement and software filtering, most of the existing wind noise suppression technologies are based on a Fourier transform framework, and a signal-noise separation method of convolution nonnegative matrix decomposition is adopted, so that the separation of signals and noise of ballistic waves, blast waves, voice and the like can be realized. However, in time-frequency domain analysis of the rifling wave signal and the wind noise under a Fourier transform frame, the frequency domains of the rifling wave signal and the wind noise are highly overlapped, and the rifling wave can not be effectively separated by the convolution nonnegative matrix decomposition method, so that signal-noise separation can not be realized by the convolution nonnegative matrix decomposition method when the rifling wave is detected. The detection direction of the rifling wave needs to use a convolution nonnegative matrix decomposition method to realize signal-noise separation so as to inhibit wind noise and improve the signal-noise ratio level.
Referring to fig. 1-3, fig. 1 is a flow chart of a conventional wind noise suppression method; FIG. 2 is a diagram of an original signal; fig. 3 is a schematic diagram of the filtered signal. As shown in fig. 1-3, a conventional convolution non-negative matrix factorization (CSNMF) method is specifically described, which mainly includes signal modeling, separation, and reconstruction. The CSNMF algorithm is operated in the STFT domain of data, so that the noisy target signal and the wind noise training data in FIG. 1 are reconstructed by the short-time amplitude spectrum of STFT. Firstly, carrying out signal acquisition on wind noise, and realizing characteristic learning of the wind noise to obtain a base matrix capable of representing wind noise characteristics; then according to a mixed model of the target signal and the wind noise, combining with the characteristic extraction of the noisy signal, carrying out decomposition operation on the noisy target signal to obtain a base matrix and a coding matrix, and realizing signal-noise separation; and finally, signal reconstruction is carried out to obtain a target signal after noise reduction.
Referring to fig. 2 and 3, the filter using the rifle wave of the CSNMF algorithm, as can be seen from comparing fig. 2 and 3, in the signal-noise separation process, the CSNMF algorithm removes the wind noise and simultaneously filters the rifle wave, so that the signal-noise separation task cannot be completed. The technical problem of the existing wind noise suppression technology is solved.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a wind noise suppression method, which includes:
carrying out empirical mode decomposition on the collected wind noise signal, and then reconstructing the wind noise signal to obtain a reconstructed wind noise signal;
performing convolution nonnegative matrix decomposition on the reconstructed wind noise signal and the mixed signal respectively;
and constructing a signal-noise mixed separation model according to the decomposition result, and performing signal-noise separation on the mixed signals acquired in real time through the signal-noise mixed separation model to obtain pure target signals.
In the wind noise suppression method, the step of obtaining the reconstructed wind noise signal includes:
acquiring the wind noise signal;
carrying out empirical mode decomposition on the wind noise signal, decomposing various intrinsic mode functions, and analyzing a plurality of component signals enveloped by the wind noise signal;
rejecting high frequency component signals in a plurality of the component signals;
reconstructing the remaining component signals into the reconstructed wind noise signal.
In the wind noise suppression method, the step of performing convolution nonnegative matrix factorization includes:
carrying out convolution nonnegative matrix decomposition on the reconstructed wind noise signal, and extracting a noise base matrix and a noise coding matrix of the reconstructed wind noise signal;
acquiring the mixed signal;
and carrying out convolution nonnegative matrix decomposition on the mixed signal, and extracting a signal base matrix and a signal coding matrix of the mixed signal.
In the wind noise suppressing method, the step of extracting the base matrix further includes ensuring an approximation effect in matrix decomposition by using an objective function.
In the wind noise suppression method, the step of obtaining a pure target signal includes:
constructing the signal-noise mixed separation model according to the noise basis matrix of the reconstructed wind noise signal and the signal basis matrix of the mixed signal;
and performing decomposition operation on the mixed signals acquired in real time through the signal-noise mixed separation model according to a CSNMF algorithm, and reconstructing the result of the decomposition operation to obtain pure target signals.
The present invention also provides a wind noise suppression system, comprising:
the wind noise signal reconstruction unit is used for carrying out empirical mode decomposition on the collected wind noise signals and then reconstructing the wind noise signals to obtain reconstructed wind noise signals;
the decomposition unit is used for respectively carrying out convolution nonnegative matrix decomposition on the reconstructed wind noise signal and the mixed signal;
and the signal-noise separation unit is used for constructing a signal-noise mixed separation model according to the decomposition result, and obtaining a pure target signal after performing signal-noise separation on the mixed signal acquired in real time through the signal-noise mixed separation model.
The above wind noise suppression system, wherein the wind noise signal reconstructing unit includes:
the first acquisition module acquires and acquires the wind noise signal;
the empirical mode decomposition module is used for performing empirical mode decomposition on the wind noise signal, decomposing various intrinsic mode functions and analyzing a plurality of component signals enveloped by the wind noise signal;
the rejecting module rejects high-frequency component signals in the component signals;
and the reconstruction module reconstructs the residual component signals into the reconstructed wind noise signals.
The above wind noise suppression system, wherein the decomposition unit includes:
the first extraction module is used for carrying out convolution nonnegative matrix decomposition on the reconstructed wind noise signal and extracting a noise basis matrix of the reconstructed wind noise signal;
the second acquisition module acquires and acquires the mixed signal;
and the second extraction module is used for carrying out convolution nonnegative matrix decomposition on the mixed signal and extracting a signal base matrix of the mixed signal.
In the wind noise suppression system, the first extraction module and the second extraction module further ensure an approximation effect in matrix decomposition through an objective function.
The wind noise suppression system described above, wherein the signal-to-noise separation unit includes:
the model building module is used for building the signal-noise mixed separation model according to the noise basis matrix of the reconstructed wind noise signal and the signal basis matrix of the mixed signal;
and the target signal obtaining module is used for carrying out decomposition operation on the mixed signals acquired in real time through the signal-noise mixed separation model according to a CSNMF algorithm and reconstructing the result of the decomposition operation so as to obtain the pure target signals.
The invention also provides a shot sound detection method, which comprises the wind noise suppression method for identifying and detecting the rifling wave signals.
The invention also provides a shot sound detection system, which comprises the wind noise suppression system for identifying and detecting the rifling wave signals.
Aiming at the prior art, the invention has the following effects: the invention combines an empirical mode decomposition method, utilizes an improved convolution nonnegative matrix combined characteristic decomposition technology to restrain wind noise and realizes the signal-noise separation of rifling waves. The invention combines an empirical mode decomposition method and a convolution non-negative matrix decomposition method, provides a wind noise suppression technology under combined characteristic decomposition, and realizes the application of the wind noise suppression technology in the detection of the shot sound; the detection and identification of the rifling wave signals are realized under the framework of a convolution nonnegative matrix decomposition method.
Drawings
FIG. 1 is a flow chart of a conventional wind noise suppression process;
FIG. 2 is a diagram of an original signal;
FIG. 3 is a schematic diagram of the filtered signal;
FIG. 4 is a flow chart of a wind noise suppression method according to the present invention;
FIG. 5 is a flowchart illustrating the substeps of step S1 in FIG. 4;
FIG. 6 is a flowchart illustrating the substeps of step S2 in FIG. 4;
FIG. 7 is a flowchart illustrating the substeps of step S3 in FIG. 4;
FIG. 8 is a schematic diagram of a filtered signal according to the present invention;
fig. 9 is a schematic structural diagram of a wind noise suppression system according to the present invention.
Wherein, the reference numbers:
reconstruction wind noise signal unit 11
Empirical mode decomposition module 112
Rejecting module 113
Signal-to-noise separating unit 13
Target signal acquisition module 132
Detailed Description
The detailed description and technical description of the present invention are further described in the context of a preferred embodiment, but should not be construed as limiting the practice of the present invention.
Aiming at the problem that the CSNMF algorithm cannot inhibit wind noise in rifling waves, the invention provides the EMD-CSNMF technology by combining the empirical mode decomposition technology and the convolution non-negative matrix decomposition technology, realizes the combined characteristic decomposition of wind noise signals and rifling signals, and realizes the signal-noise separation.
Firstly, collecting wind noise signals under the framework of a convolution nonnegative matrix decomposition technology, carrying out empirical mode decomposition on the wind noise signals, decomposing out various intrinsic mode functions, and analyzing component signals such as high-frequency components and low-frequency components enveloped by the wind noise signals; after the first high-frequency component is removed, wherein the first high-frequency component is a component signal with the highest frequency in the component signals, and the remaining component signals are reconstructed into wind noise signals; secondly, extracting a base matrix of the reconstructed wind noise signal from the reconstructed signal by applying a convolution nonnegative matrix decomposition technology to realize the feature extraction of the reconstructed wind noise; thirdly, performing convolution nonnegative matrix decomposition on the mixed signal with the shot noise by using a convolution nonnegative matrix decomposition technology, and extracting a base matrix of the mixed signal to realize feature extraction of the mixed signal; and finally, establishing a signal-noise mixed separation model, and realizing the reconstruction of the target signal according to the traditional convolution nonnegative matrix decomposition technology.
Referring to fig. 4-7, fig. 4 is a flow chart of a wind noise suppression method according to the present invention; FIG. 5 is a flowchart illustrating the substeps of step S1 in FIG. 4; FIG. 6 is a flowchart illustrating the substeps of step S2 in FIG. 4; fig. 7 is a flowchart illustrating a substep of step S3 in fig. 4. As shown in fig. 4 to 7, the wind noise suppression method of the present invention includes:
step S1: and carrying out empirical mode decomposition on the collected wind noise signals and then reconstructing to obtain reconstructed wind noise signals.
In step S1, the method includes:
step S11: and acquiring the wind noise signal.
Step S12: and carrying out empirical mode decomposition on the wind noise signal, decomposing various intrinsic mode functions, and analyzing a plurality of component signals enveloped by the wind noise signal.
Specifically, the empirical mode decomposition method is proposed by ne.huang and the like in the study of the nonlinear problem and the hilbert transform, and can enable signal decomposition to have uniqueness and simultaneously have good localization property in a time domain and a frequency domain. Once the decomposition of the signal is finished, the reconstruction can be flexibly realized according to the requirements of engineering problems.
Assuming that any signal is composed of different Intrinsic Mode Functions (IMFs), each of which may be linear or non-linear, the IMF components must satisfy the following two conditions: the number of extreme points and the number of zero-crossing points are the same or have one difference at most, and the upper envelope and the lower envelope are locally symmetrical about a time axis. So that any one signal can be decomposed into a sum of a finite number of IMFs.
The decomposition process is based on the following assumptions: (a) the signal has at least one maximum and one minimum; (b) the time domain characteristic is determined by the extremum interval; (c) if the data sequence completely lacks extreme values but only contains inflection points, it can also reveal extreme points by differentiating one or more times, and the final result can be obtained by integrating these components. The specific method is completed by a screening process, and the decomposition process is as follows:
(1) firstly, finding all maximum value points in the wind noise signal w (t) and fitting the maximum value points into an upper envelope line of the original data sequence by using a cubic spline function, and fitting all minimum value points into a lower envelope line of the original data sequence by using the cubic spline function.
(2) Calculate the mean value of the upper and lower envelope, and is recorded as m1(t); subtracting the mean value from the original data sequence w (t):
w(t)-m1(t)＝h1(t) (1)
(3) because of h1(t) is still generally not an IMF component sequence for which the above process is repeated. Repeating the above process k times until h1(t) meets the definition requirement of IMF, and the obtained mean value approaches zero, so that the 1 st IMF component c is obtained1(t), which represents the highest frequency component of the signal x (t):
h1(k-1)(t)-m1(k-1)(t)＝h1k(t) (2)
c1(t)＝h1k(t) (3)
(4) c is to1(t) extracting from w (t) to obtain a difference signal r with high frequency components removed1(t), namely:
r1(t)＝w(t)-c1(t) (4)
(5) will r is1(t) repeating the steps (1), (2) and (3) as the original data to obtain a second IMF component c2And (t), repeating the steps for n times to obtain n IMF components. Namely:
when c is going ton(t) or rn(t) satisfies a given termination condition (r)n(t) monotone), the cycle ends. The following equations (1) to (5) can be obtained:
wherein r isn(t) is a residual function; c. CjAnd (t) is an IMF component, which comprises components with different time characteristics and sizes, and the sizes are sequentially from small to large. At this point, the empirical mode decomposition method completes the decomposition of each characteristic component.
Step S13: rejecting high frequency component signals in the plurality of component signals;
this step rejects the first high frequency feature component c1(t) of (d). The first high frequency component in the wind noise represents the overlapping portion of the wind noise and the rifled hilbert (Hilber) spectrum after transformation. By means of a measure of eliminating the first high-frequency characteristic component, a Hilber spectrum overlapping area between the rifling wave signal and the processed wind noise is reduced, correlation between the rifling wave signal and the processed wind noise is reduced, and the effect of signal-noise separation is improved.
The hilbert (Hilber) spectrum refers to a spectrum analysis method corresponding to Empirical Mode Decomposition (EMD), which is different from a spectrum analysis method under fourier transform and has better time-frequency aggregation. The spectrum analysis method under the Fourier change corresponds to a Fourier (Fourier) spectrum, the frequency of a time domain local signal of the Fourier (Fourier) spectrum needs a complete oscillation wave period to realize statistics, but a short-time domain pulse signal without the complete oscillation wave cannot be effectively counted under the Fourier (Fourier) spectrum. The spectrogram of a time-domain pulse signal under a Fourier (Fourier) spectrum shows the characteristics of wide frequency spectrum band, large dispersion and lack of aggregation, which is also the root cause of the CSNMF algorithm being incapable of processing the pulse signal under a low signal-to-noise ratio.
Unlike the Fourier spectrum, the temporal frequency defined in the hilbert (Hilber) spectrum can study the local frequency characteristics of the signal by defining the temporal frequency without requiring a complete cycle of the oscillation wave, so that the hilbert (Hilber) spectrum can better reflect the local characteristics of the signal, which is superior to the Fourier spectrum. The Hilbert (Hilber) spectrum can better analyze non-stationary signals, and even if the frequency of the signals changes along with time, the vibration characteristics of the signals can be reflected.
First high-frequency characteristic component c1(t) represents the high frequency signal of the outermost envelope of the wind noise signal in the time domain, which is highly coincident with the outermost envelope of the rifling waves. Eliminating the first high-frequency characteristic component c1After (t), when the CSNMF algorithm is used again, the time-frequency overlapping area between the rifling wave signals and the wind noise is reduced, and therefore signal-noise separation is achieved.
Step S14: reconstructing the remaining component signals to obtain the reconstructed wind noise signal.
Step S2: and carrying out convolution nonnegative matrix decomposition on the acquired mixed signal and the reconstructed wind noise signal respectively.
In step S2, the method includes:
step S21: carrying out convolution nonnegative matrix decomposition on the reconstructed wind noise signal, and extracting a noise base matrix and a noise coding matrix of the reconstructed wind noise signal;
step S22: acquiring the mixed signal;
step S23: and carrying out convolution nonnegative matrix decomposition on the mixed signal, and extracting a signal base matrix and a signal coding matrix of the mixed signal. And extracting the noise base matrix and the noise coding matrix of the reconstructed wind noise signal and extracting the signal base matrix and the signal coding matrix of the mixed signal by the same method.
Here, the wind noise is to eliminate the first high-frequency component c1Residual noise after (t); the noisy signal is a mixed signal of noise and shot sound acquired in real time.
In particular, given a non-negative matrix X (usually a time-frequency representation of a noisy signal), a decomposition method is sought that is equal to the product of two non-negative matrices D and C, using a convolution model, i.e. a convolution model
Wherein X, Λ ∈ R≥0,M×N,D∈R≥0,M×RAnd C ∈ R≥0,R×N. The matrix D is a base matrix (Dictionarymatrix), the column vectors contained in the matrix D are base vectors, and the column vectors in the matrix X are formed by linear combination of the base vectors in the matrix D; the matrix C is called a code matrix (code matrix) and includes a combination of basis vectors to construct the matrix X. Typically, R is less than M and N, and a large number of data vectors are characterized by a small number of basis vectors.
In matrix decomposition, a completely precise decomposition process is difficult to realize, and therefore, an approximation effect in matrix decomposition is generally ensured by defining an objective function. The objective function can be generally constructed by the difference between two matrices, and the commonly used construction method has a least squares criterion (LS) or a divergence criterion:
its iteration is disclosed as:
the judgment of the convergence of the objective function can be generally completed by calculating the relative transformation rate of the objective function, and if the relative transformation rate of the objective function is smaller than a threshold epsilon, the objective function can be considered to be converged.
Step S3: and constructing a signal-noise mixed separation model according to the decomposition result, and performing signal-noise separation on the mixed signals acquired in real time through the signal-noise mixed separation model to obtain pure target signals.
In step S3, the method includes:
step S31: and constructing the signal-noise mixed separation model according to the noise basis matrix of the reconstructed wind noise signal and the signal basis matrix of the mixed signal.
Specifically, a signal-noise mixed separation model is established, and a signal consists of a shot signal and a noise signal:
x(t)＝s(t)+w(t) (11)
carrying out short-time Fourier transform (STFT) on the signal with noise to obtain a signal-noise mixture separation model, and separating the signal-noise mixture separation model into a pure target signal and wind noise:
step S32: and based on the signal-noise mixed separation model, performing decomposition operation on the mixed signals acquired in real time by adopting a CSNMF algorithm, and reconstructing the result of the decomposition operation to obtain pure target signals.
After wind noise is subjected to signal decomposition, a base matrix D capable of representing wind noise characteristics is obtainedwτ. Then, according to the mixed model of the target signal and the wind noise in the formula (12), the CSNMF algorithm is used for carrying out decomposition operation on the noisy target signal to obtain a base matrix DsτAnd a coding matrix
And finally, signal reconstruction is carried out according to the formula (13) to obtain a target signal S after noise reduction, namely a pure target signal.
Fig. 8 is a schematic diagram of the filtered signal according to the present invention, and fig. 2, 3 and 8 are combined to describe the filtering effect of the present invention.
From fig. 8 and fig. 2 of the EMD-CSNMF algorithm, it can be seen that fig. 8 successfully isolated the rifling signal around the 28 second time. Therefore, in the signal-noise separation process of the EMD-CSNMF algorithm, the rifling wave signals are successfully reserved while wind noise is removed, the signal-noise separation task under the CSNMF algorithm framework is realized, wind noise is inhibited, and the signal-to-noise ratio level of the rifling wave is improved.
The invention also provides a shot sound detection method, wherein the wind noise suppression method is used for identifying and detecting rifling wave signals, in particular to the problem that the CSNMF algorithm cannot suppress wind noise in the rifling waves.
Referring to fig. 9, fig. 9 is a schematic structural diagram of a wind noise suppression system according to the present invention. As shown in fig. 9, the wind noise suppression system of the present invention includes:
the reconstructed wind noise signal unit 11 is used for performing empirical mode decomposition on the collected wind noise signal and reconstructing the wind noise signal to obtain a reconstructed wind noise signal;
a decomposition unit 12, which performs convolution nonnegative matrix decomposition on the reconstructed wind noise signal and the mixed signal respectively;
and the signal-noise separation unit 13 is used for constructing a signal-noise mixed separation model according to the decomposition result, and obtaining a pure target signal after performing signal-noise separation on the mixed signal acquired in real time through the signal-noise mixed separation model.
Wherein the reconstructed wind noise signal unit 11 comprises:
the first acquisition module 111 acquires and acquires the wind noise signal;
an empirical mode decomposition module 112, which performs empirical mode decomposition on the wind noise signal to decompose various intrinsic mode functions and analyze a plurality of component signals enveloped by the wind noise signal;
a rejecting module 113 that rejects a high-frequency component signal of the plurality of component signals;
a reconstruction module 114 for reconstructing the remaining component signals into the reconstructed wind noise signal.
Wherein the decomposition unit 12 comprises:
a first extraction module 121, configured to perform convolution nonnegative matrix decomposition on the reconstructed wind noise signal, and extract a noise basis matrix of the reconstructed wind noise signal;
a second collecting module 122, which collects and obtains the mixed signal;
and a second extraction module 123, configured to perform convolution nonnegative matrix decomposition on the mixed signal, and extract a signal basis matrix of the mixed signal.
The first extraction module and the second extraction module also ensure the approximation effect in matrix decomposition through an objective function.
Wherein the signal-to-noise separating unit 13 includes:
the model building module 131 builds the signal-noise mixed separation model according to the noise basis matrix of the reconstructed wind noise signal and the signal basis matrix of the mixed signal;
the target signal obtaining module 132 performs decomposition operation on the mixed signal acquired in real time according to the CSNMF algorithm through the signal-noise mixed separation model, and reconstructs a result of the decomposition operation to obtain a pure target signal.
The invention also provides a shot sound detection method, which comprises the wind noise suppression method for identifying and detecting the rifling wave signals.
The invention also provides a shot sound detection system, which comprises the wind noise suppression system and is used for identifying and detecting the rifling wave signals.
In conclusion, the wind noise suppression technology of the combined characteristic decomposition realizes the rifling wave detection and the wind noise suppression under the convolution nonnegative matrix decomposition framework.
While the invention has been described with reference to specific embodiments, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (12)
1. A wind noise suppression method, comprising:
carrying out empirical mode decomposition on the collected wind noise signal and then reconstructing to obtain a reconstructed wind noise signal;
carrying out convolution nonnegative matrix decomposition on the collected mixed signal and the reconstructed wind noise signal respectively;
and constructing a signal-noise mixed separation model according to the decomposition result, and performing signal-noise separation on the mixed signals acquired in real time through the signal-noise mixed separation model to obtain pure target signals.
2. The wind noise suppression method of claim 1, wherein the step of obtaining a reconstructed wind noise signal comprises:
acquiring the wind noise signal;
carrying out empirical mode decomposition on the wind noise signal, decomposing various intrinsic mode functions, and analyzing a plurality of component signals enveloped by the wind noise signal;
rejecting high frequency component signals in the plurality of component signals;
reconstructing the remaining component signals to obtain the reconstructed wind noise signal.
3. The wind noise suppression method of claim 1, wherein said step of performing a convolutional nonnegative matrix factorization comprises:
carrying out convolution nonnegative matrix decomposition on the reconstructed wind noise signal, and extracting a noise base matrix and a noise coding matrix of the reconstructed wind noise signal;
acquiring the mixed signal;
and carrying out convolution nonnegative matrix decomposition on the mixed signal, and extracting a signal base matrix and a signal coding matrix of the mixed signal.
4. The wind noise suppression method of claim 3, wherein the step of extracting the basis matrix further comprises ensuring an approximation effect in matrix decomposition by an objective function.
5. The wind noise suppression method of claim 3, wherein the obtaining a clean target signal step comprises:
constructing the signal-noise mixed separation model according to the noise basis matrix of the reconstructed wind noise signal and the signal basis matrix of the mixed signal;
and based on the signal-noise mixed separation model, performing decomposition operation on the mixed signals acquired in real time by adopting a CSNMF algorithm, and reconstructing the result of the decomposition operation to obtain pure target signals.
6. A wind noise suppression system, comprising:
carrying out empirical mode decomposition on the collected wind noise signal and then reconstructing to obtain a reconstructed wind noise signal;
the decomposition unit is used for respectively carrying out convolution nonnegative matrix decomposition on the acquired mixed signal and the reconstructed wind noise signal;
and the signal-noise separation unit is used for constructing a signal-noise mixed separation model according to the decomposition result, and performing signal-noise separation on the mixed signals acquired in real time through the signal-noise mixed separation model so as to obtain pure target signals.
7. The wind noise suppression system of claim 6, wherein the reconstructing the wind noise signal unit comprises:
the first acquisition module acquires and acquires the wind noise signal;
the empirical mode decomposition module is used for performing empirical mode decomposition on the wind noise signal, decomposing various intrinsic mode functions and analyzing a plurality of component signals enveloped by the wind noise signal;
the rejecting module rejects high-frequency component signals in the plurality of component signals;
and the reconstruction module is used for reconstructing the residual component signals to obtain the reconstructed wind noise signals.
8. The wind noise suppression system of claim 6, wherein the decomposition unit comprises:
the first extraction module is used for carrying out convolution nonnegative matrix decomposition on the reconstructed wind noise signal and extracting a noise basis matrix of the reconstructed wind noise signal;
the second acquisition module acquires and acquires the mixed signal;
and the second extraction module is used for carrying out convolution nonnegative matrix decomposition on the mixed signal and extracting a signal base matrix of the mixed signal.
9. The wind noise suppression system of claim 8, wherein the first extraction module and the second extraction module further guarantee an approximation effect in matrix decomposition by an objective function.
10. The wind noise suppression system of claim 8, wherein the signal-to-noise separation unit comprises:
the model building module is used for building the signal-noise mixed separation model according to the noise basis matrix of the reconstructed wind noise signal and the signal basis matrix of the mixed signal;
and the target signal obtaining module is used for carrying out decomposition operation on the mixed signals acquired in real time by adopting a CSNMF algorithm based on the signal-noise mixed separation model and reconstructing the result of the decomposition operation so as to obtain the pure target signals.
11. A shot detection method, characterized in that rifling signals are identified and detected using the wind noise suppression method according to any one of claims 1 to 5.
12. A shot detection system comprising a wind noise suppression system as claimed in any one of claims 6 to 10 for identifying and detecting rifling signals.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010119512.5A CN111276154B (en) | 2020-02-26 | 2020-02-26 | Wind noise suppression method and system and shot sound detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010119512.5A CN111276154B (en) | 2020-02-26 | 2020-02-26 | Wind noise suppression method and system and shot sound detection method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111276154A true CN111276154A (en) | 2020-06-12 |
| CN111276154B CN111276154B (en) | 2022-12-09 |
Family
ID=71003914
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010119512.5A Active CN111276154B (en) | 2020-02-26 | 2020-02-26 | Wind noise suppression method and system and shot sound detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111276154B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112148059A (en) * | 2020-10-12 | 2020-12-29 | 四川科陆新能电气有限公司 | MPPT maximum power tracking method for photovoltaic power station |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030033094A1 (en) * | 2001-02-14 | 2003-02-13 | Huang Norden E. | Empirical mode decomposition for analyzing acoustical signals |
| JP2011227417A (en) * | 2010-04-23 | 2011-11-10 | Yamaha Corp | Sound processing apparatus |
| CN103926599A (en) * | 2014-04-17 | 2014-07-16 | 东南大学 | GNSS multipath effect suppression method based on EMD iteration threshold value smoothing |
| CN104685562A (en) * | 2012-11-21 | 2015-06-03 | 华为技术有限公司 | Method and device for reconstructing a target signal from a noisy input signal |
| CN105957537A (en) * | 2016-06-20 | 2016-09-21 | 安徽大学 | Voice denoising method and system based on L1/2 sparse constraint convolution non-negative matrix decomposition |
| CN106601265A (en) * | 2016-12-15 | 2017-04-26 | 中国人民解放军第四军医大学 | Method for eliminating noise in millimeter wave biological radar voice |
| WO2017144007A1 (en) * | 2016-02-25 | 2017-08-31 | 深圳创维数字技术有限公司 | Method and system for audio recognition based on empirical mode decomposition |
| CN107516065A (en) * | 2017-07-13 | 2017-12-26 | 天津大学 | The sophisticated signal denoising method of empirical mode decomposition combination dictionary learning |
| US20180138896A1 (en) * | 2016-11-17 | 2018-05-17 | Hwashen Electronics Co., Ltd. | Fuzzy entropy based noisy signal processing method and iterative singular spectrum analysis soft threshold de-noising method |
| CN110688964A (en) * | 2019-09-30 | 2020-01-14 | 哈尔滨工程大学 | Wavelet threshold and EMD combined denoising method based on sparse decomposition |
-
2020
- 2020-02-26 CN CN202010119512.5A patent/CN111276154B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030033094A1 (en) * | 2001-02-14 | 2003-02-13 | Huang Norden E. | Empirical mode decomposition for analyzing acoustical signals |
| JP2011227417A (en) * | 2010-04-23 | 2011-11-10 | Yamaha Corp | Sound processing apparatus |
| CN104685562A (en) * | 2012-11-21 | 2015-06-03 | 华为技术有限公司 | Method and device for reconstructing a target signal from a noisy input signal |
| CN103926599A (en) * | 2014-04-17 | 2014-07-16 | 东南大学 | GNSS multipath effect suppression method based on EMD iteration threshold value smoothing |
| WO2017144007A1 (en) * | 2016-02-25 | 2017-08-31 | 深圳创维数字技术有限公司 | Method and system for audio recognition based on empirical mode decomposition |
| CN105957537A (en) * | 2016-06-20 | 2016-09-21 | 安徽大学 | Voice denoising method and system based on L1/2 sparse constraint convolution non-negative matrix decomposition |
| US20180138896A1 (en) * | 2016-11-17 | 2018-05-17 | Hwashen Electronics Co., Ltd. | Fuzzy entropy based noisy signal processing method and iterative singular spectrum analysis soft threshold de-noising method |
| CN106601265A (en) * | 2016-12-15 | 2017-04-26 | 中国人民解放军第四军医大学 | Method for eliminating noise in millimeter wave biological radar voice |
| CN107516065A (en) * | 2017-07-13 | 2017-12-26 | 天津大学 | The sophisticated signal denoising method of empirical mode decomposition combination dictionary learning |
| CN110688964A (en) * | 2019-09-30 | 2020-01-14 | 哈尔滨工程大学 | Wavelet threshold and EMD combined denoising method based on sparse decomposition |
Non-Patent Citations (1)
| Title |
|---|
| 陈志全等: "基于EEMD的异常声音特征提取", 《计算机与数字工程》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112148059A (en) * | 2020-10-12 | 2020-12-29 | 四川科陆新能电气有限公司 | MPPT maximum power tracking method for photovoltaic power station |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111276154B (en) | 2022-12-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107272066B (en) | A kind of noisy seismic signal first-arrival traveltime pick-up method and device | |
| Chacon-Rodriguez et al. | Evaluation of gunshot detection algorithms | |
| CN110598166B (en) | Wavelet denoising method for adaptively determining wavelet layering progression | |
| CN110806602B (en) | Intelligent seismic data random noise suppression method based on deep learning | |
| EP1103951A2 (en) | Adaptive wavelet extraction for speech recognition | |
| US7492814B1 (en) | Method of removing noise and interference from signal using peak picking | |
| US7676046B1 (en) | Method of removing noise and interference from signal | |
| WO2001033547B1 (en) | Methods and apparatuses for signal analysis | |
| CN106772331A (en) | Target identification method and Target Identification Unit | |
| CN105242309B (en) | The drawing method and device of rule interference in a kind of geological data | |
| CN105607125A (en) | Seismic data noise suppression method based on block matching algorithm and singular value decompression | |
| CN108670291A (en) | The heart sound kind identification method of improved MFCC is combined based on EMD | |
| CN109932624B (en) | Cable partial discharge period narrow-band interference denoising method based on Gaussian scale space | |
| CN111276154B (en) | Wind noise suppression method and system and shot sound detection method and system | |
| CN109557587B (en) | VSP seismic data wellbore wave frequency domain filtering method and device | |
| CN108198558B (en) | Voice recognition method based on CSI data | |
| CN113887398A (en) | GPR signal denoising method based on variational modal decomposition and singular spectrum analysis | |
| CN108682433A (en) | The heart sound kind identification method of first-order difference coefficient based on MFCC | |
| CN108903914A (en) | A kind of heart sound kind identification method of the MFCC decomposed based on EMD | |
| CN110503967B (en) | Voice enhancement method, device, medium and equipment | |
| CN109102818B (en) | Denoising audio sampling algorithm based on signal frequency probability density function distribution | |
| Borsos et al. | Micaugment: One-shot microphone style transfer | |
| CN112213785B (en) | Seismic data desert noise suppression method based on feature-enhanced denoising network | |
| CN112764108B (en) | Novel seismic data noise suppression algorithm based on improved empirical wavelet transformation | |
| CN108037533A (en) | It is a kind of that the seismic prospecting noise drawing method for it is expected log-likelihood is recombinated based on block |
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 |