CN108334872A - Based on the feature extracting method for improving HHT transformation - Google Patents
Based on the feature extracting method for improving HHT transformation Download PDFInfo
- Publication number
- CN108334872A CN108334872A CN201810266004.2A CN201810266004A CN108334872A CN 108334872 A CN108334872 A CN 108334872A CN 201810266004 A CN201810266004 A CN 201810266004A CN 108334872 A CN108334872 A CN 108334872A
- Authority
- CN
- China
- Prior art keywords
- imf components
- narrow band
- signal
- measured signal
- imf
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Signal Processing (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
A kind of feature extracting method based on improvement HHT transformation：Decomposition and reconstruction is carried out to measured signal using wavelet packet, obtains the narrow band signal of different frequency range；Empirical mode decomposition is done to obtained each narrow band signal, obtains the IMF components of all narrow band signals；The IMF components of all narrow band signals are screened, the final IMF components of entire measured signal are obtained；Hilbert transformation is carried out to final IMF components respectively, obtains the instantaneous attribute of measured signal；According to the instantaneous attribute of measured signal, extraction can reflect the time-frequency statistical nature of measured signal time-frequency characteristic, including：The mean value of the instantaneous amplitude of each IMF components, bandwidth, peak value and the variance of the marginal spectrum of all IMF components.The present invention is based on WAVELET PACKET DECOMPOSITIONs and the improvement HHT methods of rejecting falseness IMF components to improve the accuracy of the time frequency analysis of signal.The method of the present invention can be more efficient extraction reflection characteristics of signals feature, the final efficiency for improving data mining and pattern-recognition.
Description
Technical field
The present invention relates to a kind of feature extracting methods.More particularly to a kind of for non-linear, non-stationary time varying signal
The feature extracting method based on improvement HHT transformation of time frequency analysis.
Background technology
1, feature extraction
Feature extraction refers to by way of mathematic(al) manipulation or statistical analysis, and one or more is obtained from original signal
Parameter, these parameters can represent the characteristic of the signal in one aspect.These parameters are referred to as the feature of the signal, obtain this
The process of a little features is exactly feature extraction.Feature extraction is used for data mining and area of pattern recognition, and feature extraction is therein
One of key technology.By feature extraction, the parameter that can reflect data characteristic in signal can be obtained, the feature of extraction is good and bad
Determine the performance of data mining and pattern-recognition.
2、HHT
Signal is generally extracted using Time-Frequency Analysis Method due to its stronger time variation for non-linear, non-stationary signal
Feature.The Time-Frequency Analysis Method of signal can divide local component by signal on the basis of being unfolded in time frequency space
Analysis, this method can preferably reflect the time-frequency characteristic of audio signal.Common Time-Frequency Analysis Method has Fourier's change in short-term
(STFT), Wigner-Viller distributions (WVD), wavelet transformation (WT) are changed, although these methods can preferably analyze audio letter
Number time-frequency characteristic, but there is also certain problems, if STFT is there are Heisenberg uncertainty principle and single resolution problem,
There are serious cross-interference terms in WVD, wavelet transformation, which exists, excessively relies on the limitations such as wavelet basis selection and energy leakage.
Hilbert-Huang transform (Hilbert-Huang Transform, HHT), is doctor Huang E of U.S. NASA
A kind of Time-Frequency Analysis method that (Norden E.Huang) was proposed in 1998.Since it is based entirely on signal itself, have
Adaptivity, to the high treating effect of nonlinear and non local boundary value problem, especially suitable for non-stationary, Analysis of nonlinear signals, so
Be widely used in fields such as ocean signal, seismic signal analysis, biomedicine, health monitorings, and achieve it is good at
Fruit.HHT is largely divided into two parts, first to signal carry out empirical mode decomposition (Empirical Mode Decomposition,
EMD a series of intrinsic mode function (Intrinsic Mode Function, IMF) of frequencies from high to low) is obtained, it is then right
IMF components carry out Hilbert and convert to obtain significant instantaneous frequency (instantaneous attribute of signal).Wherein each natural mode of vibration letter
Number must satisfy following 2 conditions：(1) in entire data area, extreme point and the number of zero crossing must be equal or most
Poor one of multiphase；(2) at any point, the lower packet of coenvelope line and the formation of all minimum points that all maximum points are formed
The average value of winding thread is always zero.
Different from other Time-Frequency Analysis Methods based on Fourier transformation, there is no many of Fourier spectrum analysis to lack by HHT
Point, as HHT can handle any signal in the case of no priori, and not by Heisenberg uncertainty principles
Limitation.
3, falseness IMF components are screened
During HHT signal Analysis, due to during empirical mode decomposition there is data envelopment fitting, modal overlap,
The shortcomings of end effect and excessively decomposition, will produce some false IMF components in decomposition, be not enough to reflection signal itself
Feature.In practical applications, it needs to reject these false IMF components.The relevance of true IMF components and original signal is than pseudo-
The relevance of component and original signal is big, and the more false vacation IMF components of ratio that true IMF components account for are big.It rejects at present empty
The method of false IMF has correlation coefficient process, grey-relational degree, mutual information, the ratio of energy, K-S methods of inspection etc..
It is existing at present to extract the inaccurate problem of feature for non-linear, non-stationary signal.
Invention content
The technical problem to be solved by the invention is to provide it is a kind of can carry out efficient feature extraction based on improve HHT
The feature extracting method of transformation.
The technical solution adopted in the present invention is：A kind of feature extracting method based on improvement HHT transformation, including walk as follows
Suddenly：
1) it uses wavelet packet to carry out decomposition and reconstruction to measured signal x (t), obtains the narrow band signal of different frequency range；
2) empirical mode decomposition is done to each narrow band signal obtained in step 1), obtains IMF points of all narrow band signals
Amount；
3) the IMF components of all narrow band signals are screened, obtains the final IMF components of entire measured signal x (t)；
4) Hilbert transformation is carried out to final IMF components respectively, obtains the instantaneous attribute of measured signal x (t)；
5) according to the instantaneous attribute of measured signal x (t), extraction can reflect the time-frequency system of measured signal x (t) time-frequency characteristics
Feature is counted, including：The mean value of the instantaneous amplitude of each IMF components, bandwidth, peak value and the variance of the marginal spectrum of all IMF components.
Step 1) is to carry out N layers of WAVELET PACKET DECOMPOSITION and reconstruct to input measured signal x (t) using Daubechies small echos,
Obtain 2NThe narrow band signal of a different frequency range.
Step 2) includes：
(1) i-th of narrow band signal x is found outi(t) all local maximums and minimum；
(2) cubic spline interpolation is carried out respectively to all local maximums and minimum, obtained by all local maximums
The lower envelope line that the coenvelope line of composition and all local minimums are constituted, is denoted as u (t), v (t) respectively；
(3) mean value of upper and lower envelope is asked to be：
(4) h (t)=x is enabledi(t) whether-m (t), verification h (t) meet the conditions of IMF components, if satisfied, then h (t) is the
One IMF component；It is such as unsatisfactory for, narrow band signal x is replaced with h (t)i(t), (1) step is returned to, until obtaining first IMF points
Amount, and first IMF component is denoted as c1(t)；
(5) by r1(t)=xi(t)-c1(t) as new signal Analysis, (1) step is repeated to (4) step, obtains second
A IMF components c2(t), r is remembered at this time2(t)=r1(t)-c2(t)；
(6) (5) step is repeated, until obtaining remainder rm(t) it is a monotonic signal or remainder rm(t) value is less than pre-
First given remainder threshold value, decomposition terminate, and obtain a series of IMF components of i-th of narrow band signal；
(7) operation of (1) step to (6) step is carried out respectively to remaining narrow band signal, until all narrow band signals
Until all having carried out empirical mode decomposition, the IMF components of all narrow band signals are obtained.
Step 3) includes：The mutual information of each IMF components and measured signal x (t) and mutual information threshold value Mi are carried out
Comparison, filter out the true IMF components that can reflect measured signal x (t) features, by the true IMF components screened according to
The sequence sequence of frequency from high to low, obtains the final IMF components of entire measured signal x (t).
The instantaneous attribute of measured signal x (t) described in step 4) includes：The instantaneous frequency of each IMF components and instantaneous width
Degree, marginal spectrum and the Hilbert spectrum of all IMF components.
The feature extracting method based on improvement HHT transformation of the present invention based on WAVELET PACKET DECOMPOSITION and rejects falseness IMF components
Improvement HHT methods improve the accuracy of the time frequency analysis of signal.The present invention method can be more efficient extraction reflection letter
The feature of number characteristic, the final efficiency for improving data mining and pattern-recognition.
Description of the drawings
Fig. 1 is that the present invention is based on the flows for the feature extracting method for improving HHT transformation；
Fig. 2 is three layers of WAVELET PACKET DECOMPOSITION process schematic.
Specific implementation mode
With reference to embodiment and attached drawing being made specifically based on the feature extracting method for improving HHT transformation to the present invention
It is bright.
As shown in Figure 1, the feature extracting method based on improvement HHT transformation of the present invention, includes the following steps：
1) it uses wavelet packet to carry out decomposition and reconstruction to measured signal x (t), obtains the narrow band signal of different frequency range；Specifically
Be using in Daubechies small echos db4 or db6 or db7 or db8 N layers of WAVELET PACKET DECOMPOSITION are carried out to input measured signal x (t)
With reconstruct, 2 are obtainedNThe narrow band signal of a different frequency range, wherein three layers of WAVELET PACKET DECOMPOSITION schematic diagram are as shown in Fig. 2, in figure, S tables
Show measured signal x (t), A indicates that low frequency, D indicate that high frequency, the serial number at end indicate the number of plies of WAVELET PACKET DECOMPOSITION, and decomposing has
Relationship such as following formula：
S=AAA3+DAA3+ADA3+DDA3+AAD3+DAD3+ADD3+DDD3.
2) empirical mode decomposition is done to each narrow band signal obtained in step 1), obtains IMF points of all narrow band signals
Amount；Including：
(1) i-th of narrow band signal x is found outi(t) all local maximums and minimum；
(2) cubic spline interpolation is carried out respectively to all local maximums and minimum, obtained by all local maximums
The lower envelope line that the coenvelope line of composition and all local minimums are constituted, is denoted as u (t), v (t) respectively；
(3) mean value of upper and lower envelope is asked to be：
(4) h (t)=x is enabledi(t) whether-m (t), verification h (t) meet the conditions of IMF components, if satisfied, then h (t) is the
One IMF component；It is such as unsatisfactory for, narrow band signal x is replaced with h (t)i(t), (1) step is returned to, until obtaining first IMF points
Amount, and first IMF component is denoted as c1(t)；
The condition of the IMF components is：(1) in entire data area, extreme point must be equal with the number of zero crossing
Or at most differ one；(2) at any point, coenvelope line and all minimum points that all maximum points are formed are formed
The average value of lower envelope line be always zero.
(5) by r1(t)=xi(t)-c1(t) as new signal Analysis, (1) step is repeated to (4) step, obtains second
A IMF components c2(t), r is remembered at this time2(t)=r1(t)-c2(t)；
(6) (5) step is repeated, until obtaining remainder rm(t) it is a monotonic signal or remainder rm(t) value is less than pre-
First given remainder threshold value, decomposition terminate, and obtain a series of IMF components of i-th of narrow band signal；
(7) operation of (1) step to (6) step is carried out respectively to remaining narrow band signal, until all narrow band signals
Until all having carried out empirical mode decomposition, the IMF components of all narrow band signals are obtained.
3) the IMF components of all narrow band signals are screened, obtains the final IMF components of entire measured signal x (t)；
Including：The mutual information of each IMF components and measured signal x (t) is compared with mutual information threshold value Mi, mutual information threshold
Value Mi is generally taken as 0.1.The true IMF components of measured signal x (t) features can be reflected by filtering out, true by what is screened
IMF components sort according to the sequence of frequency from high to low, obtain the final IMF components of entire measured signal x (t).
4) Hilbert transformation is carried out to final IMF components respectively, obtains the instantaneous attribute of measured signal x (t)；It is described
The instantaneous attribute of measured signal x (t) include：The instantaneous frequency and instantaneous amplitude of each IMF components, the side of all IMF components
Border is composed and Hilbert spectrums.
5) according to the instantaneous attribute of measured signal x (t), extraction can reflect the time-frequency system of measured signal x (t) time-frequency characteristics
Feature is counted, including：The mean value of the instantaneous amplitude of each IMF components, bandwidth, peak value and the variance of the marginal spectrum of all IMF components.
Claims (5)
1. a kind of based on the feature extracting method for improving HHT transformation, which is characterized in that include the following steps：
1) it uses wavelet packet to carry out decomposition and reconstruction to measured signal x (t), obtains the narrow band signal of different frequency range；
2) empirical mode decomposition is done to each narrow band signal obtained in step 1), obtains the IMF components of all narrow band signals；
3) the IMF components of all narrow band signals are screened, obtains the final IMF components of entire measured signal x (t)；
4) Hilbert transformation is carried out to final IMF components respectively, obtains the instantaneous attribute of measured signal x (t)；
5) according to the instantaneous attribute of measured signal x (t), extraction can reflect that the time-frequency statistics of measured signal x (t) time-frequency characteristics is special
Sign, including：The mean value of the instantaneous amplitude of each IMF components, bandwidth, peak value and the variance of the marginal spectrum of all IMF components.
2. according to claim 1 based on the feature extracting method for improving HHT transformation, which is characterized in that step 1) is to adopt
N layers of WAVELET PACKET DECOMPOSITION and reconstruct are carried out to input measured signal x (t) with Daubechies small echos, obtain 2NA different frequency range
Narrow band signal.
3. according to claim 1 based on the feature extracting method for improving HHT transformation, which is characterized in that step 2) includes：
(1) i-th of narrow band signal x is found outi(t) all local maximums and minimum；
(2) cubic spline interpolation is carried out respectively to all local maximums and minimum, obtains being made of all local maximums
Coenvelope line and all local minimums constitute lower envelope line, be denoted as u (t), v (t) respectively；
(3) mean value of upper and lower envelope is asked to be：
(4) h (t)=x is enabledi(t) whether-m (t), verification h (t) meet the condition of IMF components, if satisfied, then h (t) is first
IMF components；It is such as unsatisfactory for, narrow band signal x is replaced with h (t)i(t), (1) step is returned, until first IMF component is obtained, and
First IMF component is denoted as c1(t)；
(5) by r1(t)=xi(t)-c1(t) as new signal Analysis, (1) step is repeated to (4) step, obtains second IMF
Component c2(t), r is remembered at this time2(t)=r1(t)-c2(t)；
(6) (5) step is repeated, until obtaining remainder rm(t) it is a monotonic signal or remainder rm(t) value is less than gives in advance
Fixed remainder threshold value, decomposition terminate, and obtain a series of IMF components of i-th of narrow band signal；
(7) carry out operation of (1) step to (6) step respectively to remaining narrow band signal, until all narrow band signals all into
It has gone until empirical mode decomposition, has obtained the IMF components of all narrow band signals.
4. according to claim 1 based on the feature extracting method for improving HHT transformation, which is characterized in that step 3) includes：
The mutual information of each IMF components and measured signal x (t) is compared with mutual information threshold value Mi, filtering out can reflect
The true IMF components of measured signal x (t) features arrange the true IMF components screened according to the sequence of frequency from high to low
Sequence obtains the final IMF components of entire measured signal x (t).
5. according to claim 1 based on the feature extracting method for improving HHT transformation, which is characterized in that step 4) is described
The instantaneous attribute of measured signal x (t) include：The instantaneous frequency and instantaneous amplitude of each IMF components, the side of all IMF components
Border is composed and Hilbert spectrums.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810266004.2A CN108334872A (en) | 2018-03-28 | 2018-03-28 | Based on the feature extracting method for improving HHT transformation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810266004.2A CN108334872A (en) | 2018-03-28 | 2018-03-28 | Based on the feature extracting method for improving HHT transformation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108334872A true CN108334872A (en) | 2018-07-27 |
Family
ID=62931515
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810266004.2A Pending CN108334872A (en) | 2018-03-28 | 2018-03-28 | Based on the feature extracting method for improving HHT transformation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108334872A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109460614A (en) * | 2018-11-12 | 2019-03-12 | 广西交通科学研究院有限公司 | Signal time based on instant bandwidth-frequency decomposition method |
| CN110175508A (en) * | 2019-04-09 | 2019-08-27 | 杭州电子科技大学 | A kind of Eigenvalue Extraction Method applied to ultrasonic partial discharge detection |
| CN110426005A (en) * | 2019-07-01 | 2019-11-08 | 中国铁道科学研究院集团有限公司节能环保劳卫研究所 | Rail in high speed railway wave based on IMF energy ratio grinds acoustics diagnostic method |
| CN113792628A (en) * | 2021-08-30 | 2021-12-14 | 荆州市明德科技有限公司 | HHT-based intelligent automatic waveform analysis method |
| CN115597901A (en) * | 2022-12-13 | 2023-01-13 | 江苏中云筑智慧运维研究院有限公司（Cn） | Method for monitoring damage of bridge expansion joint |
-
2018
- 2018-03-28 CN CN201810266004.2A patent/CN108334872A/en active Pending
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109460614A (en) * | 2018-11-12 | 2019-03-12 | 广西交通科学研究院有限公司 | Signal time based on instant bandwidth-frequency decomposition method |
| CN109460614B (en) * | 2018-11-12 | 2023-06-30 | 广西交通科学研究院有限公司 | Signal time-frequency decomposition method based on instantaneous bandwidth |
| CN110175508A (en) * | 2019-04-09 | 2019-08-27 | 杭州电子科技大学 | A kind of Eigenvalue Extraction Method applied to ultrasonic partial discharge detection |
| CN110426005A (en) * | 2019-07-01 | 2019-11-08 | 中国铁道科学研究院集团有限公司节能环保劳卫研究所 | Rail in high speed railway wave based on IMF energy ratio grinds acoustics diagnostic method |
| CN110426005B (en) * | 2019-07-01 | 2020-11-20 | 中国铁道科学研究院集团有限公司节能环保劳卫研究所 | High-speed railway rail corrugation acoustic diagnosis method based on IMF energy ratio |
| CN113792628A (en) * | 2021-08-30 | 2021-12-14 | 荆州市明德科技有限公司 | HHT-based intelligent automatic waveform analysis method |
| CN113792628B (en) * | 2021-08-30 | 2024-04-12 | 荆州市明德科技有限公司 | HHT-based intelligent automatic waveform analysis method |
| CN115597901A (en) * | 2022-12-13 | 2023-01-13 | 江苏中云筑智慧运维研究院有限公司（Cn） | Method for monitoring damage of bridge expansion joint |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108334872A (en) | Based on the feature extracting method for improving HHT transformation | |
| Gilles | Empirical wavelet transform | |
| US7401008B2 (en) | Method, computer program, and system for intrinsic timescale decomposition, filtering, and automated analysis of signals of arbitrary origin or timescale | |
| CN107316653B (en) | Improved empirical wavelet transform-based fundamental frequency detection method | |
| CN107103908B (en) | Polyphonic music polyphonic pitch height estimation method and application of pseudo bispectrum in polyphonic pitch estimation | |
| CN109101910B (en) | A kind of Magnetotelluric signal denoising method screened based on noise | |
| CN105760347B (en) | It is a kind of based on data/extreme value joint symmetric continuation HHT end effect suppressing method | |
| CN108844617B (en) | The adaptive sparse tree construction noise-reduction method of main reducing gear very noisy vibration signal | |
| CN101017201A (en) | Signal processing method of laser radar based on empirical mode decomposition | |
| CN109359633B (en) | Signal joint classification method based on Hilbert-Huang transform and wavelet ridge line | |
| Yang et al. | Ship recognition via its radiated sound: The fractal based approaches | |
| CN103209036A (en) | Transient signal detection method based on Hilbert-Huang transform (HHT) double noise reduction | |
| Le et al. | Weak LFM signal dectection based on wavelet transform modulus maxima denoising and other techniques | |
| CN113670186A (en) | GNSS deformation monitoring denoising method combining variational modal decomposition and permutation entropy | |
| CN107729289A (en) | A kind of adaptive time-frequency conversion method of Polynomial Phase Signals based on genetic optimization | |
| Yang et al. | Radar emitter signal recognition based on time-frequency analysis | |
| CN108761202B (en) | Harmonic detection method combining pole symmetric modal decomposition and Hilbert transform | |
| Limin et al. | Low probability of intercept radar signal recognition based on the improved AlexNet model | |
| CN109460614B (en) | Signal time-frequency decomposition method based on instantaneous bandwidth | |
| Ren et al. | Automatic modulation recognition for overlapping radar signals based on multi-domain se-resnext | |
| Solomon et al. | Signal denoising using Ramanujan periodic transform | |
| Shreyamsha Kumar et al. | Target identification using harmonic wavelet based ISAR imaging | |
| Auger et al. | Quadratic Time‐Frequency Analysis I: Cohen's Class | |
| Biswal et al. | Time frequency analysis and FPGA implementation of modified S-transform for de-noising | |
| He et al. | Sparsity-assisted signal representation for rotating machinery fault diagnosis using the tunable Q-factor wavelet transform with overlapping group shrinkage |
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: 20180727 |