WO2018090467A1 - 基于模糊熵的含噪信号处理方法及迭代奇异谱软阈值去噪方法 - Google Patents
基于模糊熵的含噪信号处理方法及迭代奇异谱软阈值去噪方法 Download PDFInfo
- Publication number
- WO2018090467A1 WO2018090467A1 PCT/CN2016/113611 CN2016113611W WO2018090467A1 WO 2018090467 A1 WO2018090467 A1 WO 2018090467A1 CN 2016113611 W CN2016113611 W CN 2016113611W WO 2018090467 A1 WO2018090467 A1 WO 2018090467A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- signal
- noise
- fuzzy entropy
- denoising
- fuzzy
- Prior art date
Links
Images
Classifications
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03H—IMPEDANCE NETWORKS, e.g. RESONANT CIRCUITS; RESONATORS
- H03H17/00—Networks using digital techniques
- H03H17/02—Frequency selective networks
- H03H17/0219—Compensation of undesirable effects, e.g. quantisation noise, overflow
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R23/00—Arrangements for measuring frequencies; Arrangements for analysing frequency spectra
- G01R23/16—Spectrum analysis; Fourier analysis
- G01R23/165—Spectrum analysis; Fourier analysis using filters
-
- 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
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- 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
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03H—IMPEDANCE NETWORKS, e.g. RESONANT CIRCUITS; RESONATORS
- H03H17/00—Networks using digital techniques
- H03H17/02—Frequency selective networks
- H03H17/0248—Filters characterised by a particular frequency response or filtering method
- H03H17/0255—Filters based on statistics
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03H—IMPEDANCE NETWORKS, e.g. RESONANT CIRCUITS; RESONATORS
- H03H2222/00—Indexing scheme relating to digital filtering methods
- H03H2222/02—Indexing scheme relating to digital filtering methods using fuzzy logic
Definitions
- the invention claims Chinese Patent Application No. 201611010791.1, and the application date is November 17, 2016, and the priority is "the fuzzy entropy-based noise-containing signal processing method and the iterative singular spectrum soft threshold denoising method", and the application is introduced in its entirety. The way is combined here.
- the invention relates to signal denoising and filtering, in particular to a noisy signal processing method based on fuzzy entropy and an iterative singular spectral soft threshold denoising method.
- SSA Singular Spectrum Analysis
- the purpose of the SSA is to decompose the signal to be analyzed into the sum of multiple physically meaningful components, such as trends, oscillating components, and noise. Based on these components, the researchers proposed different denoising, change point detection, missing value interpolation, synchronization detection, feature extraction and prediction algorithms.
- the method In order to filter out noise, the method first calculates the eigenvalues of the delay covariance matrix and arranges them in descending order. These eigenvalues may form a relatively flat trailing plane from a certain order, the so-called "noise plane". .
- the component corresponding to the larger eigenvalue of the initial steep portion of the singular spectrum constitutes the base of the signal subspace, and the component corresponding to the noise plane portion is considered to be white noise and discarded.
- a high eigenvalue corresponds to the fundamental oscillating component of the signal, and the largest singular value is usually associated with a gradual trend.
- This method of truncating singular spectrum has been widely used for speech, ultrasound, Doppler radar signals, biomedical signals (EEG, ECG, EMG) and mechanical signals as well as denoising of hyperspectral images. In order to obtain a satisfactory denoising effect, previous research and invention focused on how to determine or find the optimal order of the noise plane.
- the initial starting point for the truncated SSA method is that any signal has an inherent "noise plane", but many signals with lower signal-to-noise ratio (SNR), especially those recorded on-site, often have a smooth exponential function shape. There is no clear noise plane.
- the singular spectrum algorithm itself only solves the problem of determining the best low-rank expression for a signal and noise mixing test matrix. An answer is given to how to obtain the best estimate of the low rank signal matrix. Therefore, one should not expect SSA denoising to get the best performance.
- singular spectrum uses a binary method that preserves part of the component and discards part of the component. Filtering out the high-frequency components of the signal does not significantly improve the signal-to-noise ratio.
- Fuzzy entropy is a chaotic invariant used to characterize the complexity of the system in chaos theory.
- fuzzy entropy spectrum we propose the concept of fuzzy entropy spectrum, and use this spectrum to characterize the signal noise plane, and invented a fuzzy entropy noise spectrum based on the noisy signal.
- An iterative SSA threshold denoising method is a chaotic invariant used to characterize the complexity of the system in chaos theory.
- the object of the present invention is to provide a noisy signal processing method spectrum and an iterative singular spectrum soft threshold denoising method based on the fuzzy entropy in view of the above-mentioned deficiencies and shortcomings in the prior art.
- N is white noise
- the same method constructs (d+1) dimensional vector And their corresponding similarities Fuzzy probability Fuzzy entropy is defined as:
- d and r are set to 2 and 0.2, respectively
- N is the length of the signal to be denoised
- the singular spectral distribution of all components is defined as the fuzzy entropy spectrum.
- the iterative singular spectrum soft threshold denoising method based on the above fuzzy entropy spectrum of the present invention comprises the following steps:
- the first component x 1 does not perform any threshold denoising, and sets a smaller threshold for the component x 2 to x k whose fuzzy entropy value is smaller than the fuzzy entropy value of the noise signal.
- ⁇ c is the component x c variance
- the present invention utilizes the fuzzy entropy of the quantization system complexity in chaos theory to characterize the noise plane, and the fuzzy entropy can quantitatively represent the noise level of each singular spectrum component relative to the white noise and the original noise signal, according to the fuzzy entropy spectrum and the SSA
- the filter characteristics through the simulation of four synthetic signals and two experimental signals, show that the denoising performance of the iterative singular spectrum soft threshold denoising method is significantly better than the traditional truncated singular spectrum method, slightly better than wavelet transform and empirical mode decomposition. Denoising method.
- the fuzzy entropy spectrum proposed by the present invention can accurately provide the relative noise level of each component of the noisy signal, which denoises different signals or improves other signals.
- the denoising method provides an important foundation.
- the invention can be widely applied to denoising of mobile equipment, hearing aids, wearable equipment, medical instruments or biomedical signals, mechanical signals, radar signals.
- Figure 1 is the fuzzy entropy spectrum (solid line) and singular spectrum (dotted line) of the piecewise-regular signal with SNR of 0dB (left) and 15dB (right), respectively.
- the dashed and dotted lines are noiseless and noisy piecewise-
- the fuzzy entropy value of the regular signal, and the rectangle is the 95% confidence band of the Gaussian white noise fuzzy entropy;
- Figure 2 is the fuzzy entropy spectrum (solid line) and singular spectrum (dotted line) of the Riemann signal with SNR of 0dB (left) and 15dB (right), respectively.
- the dashed and dotted lines are the blur of noiseless and noisy Riemann signals, respectively.
- Entropy value, the rectangle is the 95% confidence band of Gaussian white noise fuzzy entropy;
- FIG. 3 is a schematic diagram of the present invention.
- Figure 4 (a) a noisy piecewise-regular signal sample
- 4(b) to 4(d) are the denoising signals of the SSA-IST algorithm iterating once (b), four times (c), and 14 times (d), respectively;
- Figure 4 (e) is an SNR improvement curve
- Figure 5 is SSA-IST (solid line), truncated SSA (dashed line), WT (dotted line) and EMD (dotted line) four algorithms filter piecewise-regular (a), Riemann (b), blocks (c), And the signal-to-noise ratio after the sineoneoverx(d) signal.
- Fuzzy entropy is a chaotic invariant used to characterize the complexity of a system in chaos theory.
- fuzzy entropy spectrum can reveal the relative noise level of each signal component and identify whether the component is a signal or a noise dominant.
- SSA-IST singular spectral iterative soft threshold
- N respectively represent equation (1) And the Hankel matrix constructed by n.
- trajectory matrix H ⁇ ⁇ m ⁇ d can be obtained by singular value decomposition (SVD):
- Diagonal elements ⁇ A singular value called H, whose set is singular. According to equation (3), the SVD of H can also be expressed as:
- Fuzzy entropy is a measure of the robust quantization signal complexity that is applicable to any nonlinear non-stationary signal.
- d and r are set to 2 and 0.2, respectively.
- the SSA is used to decompose the noise signal.
- the first component is the low frequency component, which represents the main trend of the original signal, while the remaining component has obvious oscillation and high frequency characteristics.
- Figure 1 shows the fuzzy entropy spectrum (solid line) and singular spectrum (dotted line) of the piecewise-regular signal at 0dB (left) and 15dB (right), no noise (dashed line) and noisy (dotted line). Fuzzy entropy of the signal and high The 95% confidence band (rectangle) of the white noise fuzzy entropy is also placed in the graph for comparison.
- Figure 2 provides the same information for the Riemann signal, and the fuzzy entropy alignment in the two graphs corresponds to the order of singular values from large to small.
- Figure 3 is a schematic diagram of the iterative SSA soft threshold denoising algorithm.
- the fuzzy entropy value is larger than the original noise signal, especially those that are close to or within the 95% confidence band of white noise. Large thresholds are used for filtering.
- components with fuzzy entropy values smaller than the noise signal have more signal components, and smaller thresholds are needed, especially the first component, which contains the main trend of the signal, and its fuzzy entropy value. It is very close to the clean signal, so the first component should remain completely in the signal recovery.
- the iterative SSA soft threshold denoising algorithm is summarized as follows:
- ⁇ c is the component x c variance and N is the signal length.
- the fuzzy entropy of the remaining components is not less than the noise signal, and a larger threshold is set:
- the first denoising signal is the sum of the first component and the remaining soft threshold denoising components:
- the four composite signal samples use four different sampling frequencies to generate samples of lengths of 1024, 2048, 4096, and 8192, respectively.
- the parameter is the mean of the signal-to-noise ratio after denoising 50 noise samples.
- Figure 4 illustrates the effect of the iterative mechanism in the SSA-IST algorithm by taking the piecewise-regular signal as an example.
- Figure 4(a) shows a piecewise-regular signal sample with a length of 8192 SNR
- Figure 4(b) shows SSA- After the IST denoises the signal waveform
- Figures 4(c) and (d) are the output of the SSA-IST algorithm for 4 and 14 iterations. From these time domain waveforms, we can see that the noise is significant as the number of iterations increases. Suppression, Figure 4(e) quantitatively shows the effect of SNR increasing with the number of iterations. With 14 iterations, the SNR of the noisy signal is increased from 15dB to about 25dB, which fully demonstrates the effectiveness of the SSA-IST algorithm for denoising.
- Figure 5 shows the SSA-IST method, and the denoising effect of the four denoising methods of truncating SSA, WT and EMD for the four composite signals.
- the curve on each subgraph corresponds to the SNR after denoising of different sample length signals. Before denoising, they are divided into 0dB on the left side and 15dB on the right side.
- the signal-to-noise ratio is higher than WT in other cases after SSA-IST denoising.
- EMD is a more competitive method. Compared with different input signal SNR, EMD improves the signal-to-noise ratio of speech signal denoising better than SSA-IST, but The EMG signal, SSA-IST is superior to EMD at different signal-to-noise ratios, and the variance of SSA-IST denoising SNRs is smaller than the EMD method under both signal conditions.
- EMG signals contain many spikes, and Riemann The signal has a distinct high frequency component and the waveform is very similar to 1/f noise.
- SNRs Signal-to-noise ratio
- signal-to-noise ratio variance of four methods for denoising speech and EMG signals
- the traditional truncated SSA filtering method uses a binary method of retaining part of the component, discarding other components, is equivalent to low-pass filtering in the frequency domain, and loses information of the high-frequency band.
- the method relies on subjective search in many cases and The noise plane does not exist.
- the present invention first proposes to replace the singular spectrum with a fuzzy entropy spectrum. Regardless of the signal property and the noise level, the fuzzy entropy spectrum can accurately provide the relative noise level of each component of the noisy signal, which is to invent different signals. Noise or other processing methods provide an important foundation. Based on the fuzzy entropy spectrum, we invented the iterative singular spectrum soft threshold denoising algorithm.
- Truncating the SSA is more effective in improving the signal-to-noise ratio of the noisy signal.
- This invention can be widely applied to the denoising of mobile equipment, hearing aids, wearable equipment, medical equipment or biomedical signals, mechanical signals, and radar signals.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Optimization (AREA)
- Theoretical Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Probability & Statistics with Applications (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Algebra (AREA)
- Computer Hardware Design (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Operations Research (AREA)
- Evolutionary Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
本发明公开了一种基于模糊熵的含噪信号处理方法及迭代奇异谱(SSA)软阈值去噪方法。该方法适用于含噪信号,假设长度N的含噪信号xin={x1,x2,…,xN},并假设其中的加性白噪声与信号不相关,利用原始信号xin构造d维矢量Xd
e={xd
e,xd
e+1,…,xd
e+d-1},并且定义相似度Sd
r(e)以及模糊概率Sd
r,同样方法构造(d+1)维矢量Xd+1
e及其相应的相似度Sd+1
r(e)和模糊概率Sd+1
r,模糊熵则定义为:FuzzyEn(d,r,N)=−ln(S
d+1
r
/S
d
r
),对利用已知信号分解方法获得的分量,其所有分量的奇异谱分布定义为模糊熵谱。本发明利用混沌理论中量化系统复杂度的模糊熵来表征噪声平面,为含噪信号的处理提供了更加有效的路径;其基于模糊熵谱的迭代奇异谱(SSA-IST)软阈值去噪方法,去噪性能优于传统的截断奇异谱方法,及小波变换和经验模态分解去噪方法。
Description
本发明要求中国专利申请号201611010791.1,申请日为2016年11月17日,名称为“基于模糊熵的含噪信号处理方法及迭代奇异谱软阈值去噪方法”的优先权,该申请通过全文引入的方式合并于此。
本发明涉及信号去噪和滤波,具体涉及一种基于模糊熵的含噪信号处理方法及迭代奇异谱软阈值去噪方法。
奇异谱分析(SSA)是一种结合了经典时间序列分析、线性代数、多变量统计学、动力系统的高级信号处理方法。SSA的目的在于将待分析信号分解为多个具有物理意义的分量之和,比如趋势、震荡分量和噪声。根据这些分量,研究人员提出了不同的去噪、变点检测、缺失值插补、同步检测、特征提取和预测算法。为了滤除噪声,该方法首先计算延时协方差矩阵的特征值并按降序排列,这些特征值可能会从某个阶数开始形成一个较为平坦的拖尾状平面,即所谓的“噪声平面”。对应于奇异谱上最初陡峭部分较大特征值的分量构成了信号子空间的基底,而对应于噪声平面部分的分量则认为是白噪声而被舍弃。高的特征值对应于信号中基本震荡成分,最大的一个奇异值通常与缓变趋势相关。这种截断奇异谱的方法已被广泛用于语音、超声、多普勒雷达信号,生物医学信号(脑电,心电,肌电)和机械信号以及高光谱图像的去噪。为了获得较为满意的去噪效果,以前的研究和发明侧重于如何确定或查找噪声平面的最佳阶数。
截断SSA方法的最初出发点是任何信号都有一个固有的“噪声平面”,但是许多信噪比(SNR)较低的信号,特别是工程现场记录的信号,其奇异谱往往是平滑的指数函数形状,并不存在一个清晰地噪声平面。另外,奇异谱算法本身仅解决了如果确定一个信号和噪声混合测试矩阵的最佳低秩表达问题,并没有
对如何获得低秩信号矩阵的最佳估计给出回答。因此,人们不应期望SSA去噪能获得最佳性能。相反,奇异谱采用保留部分分量,舍弃部分分量的二元法。形同滤除了信号的高频分量,并不能显著提高信噪比。
模糊熵是混沌理论中用来表征系统复杂度的一个混沌不变量,在此,我们提出模糊熵谱的概念,并利用该谱来表征信号噪声平面,根据含噪信号模糊熵噪声谱发明了一种迭代SSA阈值去噪方法。
发明内容
本发明的目的在于:针对现有技术中存在的上述不足和欠缺,提供一种基于模糊熵的含噪信号处理方法谱及迭代奇异谱软阈值去噪方法。
本发明是通过以下技术方案实现的:
本发明的基于模糊熵的含噪信号处理方法是:对于含噪信号,假设长度N的含噪信号xin={x1,x2,…,xN},并假设其中的加性白噪声与信号不相关,即:这里代表无噪信号,n为白噪声,利用原始信号xin构造一系列d维矢量(1≤e≤N-d+1),并且定义相似度以及模糊概率同样方法构造(d+1)维矢量及其相应的相似度和模糊概率模糊熵则定义为:这里d和r分别设置为2和0.2,N为待去噪信号长度,对利用任何信号分解方法获得的分量,其所有分量的奇异谱分布定义为模糊熵谱。
本发明的基于上述模糊熵谱的迭代奇异谱软阈值去噪方法,包括如下步骤:
(2)、计算模糊熵谱:按上述模糊熵谱定义计算SSA分量的模糊熵谱和原始噪声信号的模糊熵值;
(4)、软阈值去噪:除第一分量的所有分量xc(c=2,…,d),利用相应的阈值进行去噪,即每一分量的某噪声信号数值绝对值如小于该分量阈值,输出为零,如信号数值大于阈值,则输出为信号数值减去阈值,剩余情况输出为信号数值加上去阈值,软阈值去噪后所有分量与x1之和为第一次估计信号估计噪声为
(6)、比较连续迭代获得的噪声方差,如噪声方差不再明显减小或达到指定迭代次数,迭代停止,否则重复(1)到(5);
综上所述,由于采用了上述技术方案,本发明的有益效果是:
1、本发明利用混沌理论中量化系统复杂度的模糊熵来表征噪声平面,模糊熵能够定量地表示每个奇异谱分量相对于白噪声和原始噪声信号的噪声水平,根据模糊熵谱和SSA的滤波器特性,通过四个合成信号和两个实验信号的仿真表明,迭代奇异谱软阈值去噪方法的去噪性能明显优于传统的截断奇异谱方法,稍优于小波变换和经验模态分解去噪方法。
2、无论信号性质和噪声水平,无论传统的“噪声平面”是否存在,本发明提出的模糊熵谱均能准确提供含噪信号各分量的相对噪声水平,这为不同信号去噪或改进其他信号去噪方法提供了重要基础。本发明可以广泛应用于移动装备、助听器、可穿戴装备、医疗器械或生物医学信号、机械信号、雷达信号的去噪。
本发明将通过例子并参照附图的方式说明,其中:
图1是SNR分别为0dB(左)和15dB(右)的piecewise-regular信号的模糊熵谱(实线)及奇异谱(点线),虚线和点划线分别为无噪及含噪piecewise-regular信号的模糊熵值,矩形为高斯白噪声模糊熵的95%置信带;
图2是SNR分别为0dB(左)和15dB(右)的Riemann信号的模糊熵谱(实线)及奇异谱(点线),虚线和点划线分别为无噪及含噪Riemann信号的模糊熵值,矩形为高斯白噪声模糊熵的95%置信带;
图3是本发明的原理图;
图4(a)一含噪piecewise-regular信号样本;
图4(b)至图4(d)分别是SSA-IST算法迭代一次(b),四次(c),和14次(d)的去噪信号;
图4(e)是SNR改进曲线;
图5是SSA-IST(实线),截断SSA(虚线),WT(点划线)和EMD(点线)四种算法滤波piecewise-regular(a),Riemann(b),blocks(c),和sineoneoverx(d)信号后的信噪比。
本说明书中公开的所有特征,或公开的所有方法或过程中的步骤,除了互
相排斥的特征和/或步骤以外,均可以以任何方式组合。
本说明书(包括任何附加权利要求、摘要和附图)中公开的任一特征,除非特别叙述,均可被其他等效或具有类似目的的替代特征加以替换。即,除非特别叙述,每个特征只是一系列等效或类似特征中的一个例子而已。
模糊熵是混沌理论中用来表征系统复杂度的一个混沌不变量,在此,我们首先提出模糊熵谱的概念并利用该谱来获得真实意义上的噪声平面。无论平坦的奇异谱是否存在,模糊熵谱都能够揭示每个信号分量的相对噪声水平并鉴别该分量是信号还是噪声占优。
当SSA分解一信号为其多组成分量时,基于SSA分量的模糊熵谱特性,我们发明了一种奇异谱迭代软阈值(SSA-IST)去噪方法:为了能够滤除每个分量中的噪声,根据这些SSA分量的模糊熵谱,我们将其划分为信号占优或噪声占优的两组,我们提出两个不同的阈值公式来分别过滤信号或噪声占优的分量。我们通过四个合成信号和两个实验信号(语音和肌电信号)的去噪实验来表明该方法的有效性及相对截断奇异谱方法的显著性能改善。
I.SSA和信号噪声模型
考虑长度N的含噪信号xin={x1,x2,…,xN},并假设其中的加性白噪声与信号不相关,即:
选择合适窗长度d,xin可通过SSA的第一步嵌入变换为一轨迹矩阵:
这里m=N-d+1是多维延迟矢量的数目,H为一Hankel矩阵,即其沿主对角i+j=const上的元素相等。H可表示为:
假设m≥d,轨迹矩阵H∈□m×d可通过奇异值分解(SVD)为:
H=UΣVT,U≡(u1,u2,…,um),V≡(v1,v2,…,vd) (4)
这里U∈□m×m和V∈□d×d分别由均正交的左右奇异矢量u和v构成,Σ=diag(λ1≥λ2≥…≥λd≥0).对角元素Σ称为H的奇异值,其集合即奇异谱。根据方程(3),H的SVD也可表达为:
从方程(6)到(8)我们可以发现截断SSA算法移除了噪声空间,但是保留了噪声在信号子空间的投影,这样的算法实际上包含了可能性最高的残存噪声。
II.模糊熵辅助的SSA阈值去噪方法
A.模糊熵
模糊熵是一个鲁棒的量化信号复杂度的测度,它适用于任何非线性非平稳信号。我们重新表达方程(2)中H的第e个矢量序列:
利用模糊数学中模糊概率的思想,矩阵H中所有矢量对相似的概率定义为:
这里d和r分别设置为2和0.2。
任何无噪或去噪后的带限信号具有较小的熵值,而含噪信号具有较大熵值,且随噪声水平升高,熵值变大,我们用5000个样本的蒙特卡洛模拟表明,高斯分布和均匀分布白噪声的模糊熵95%置信带的上下限分别为1.66和1.74。与其他多个量化信号复杂度的线性、非线性统计量相比,模糊熵具有更好的单调性、相对一致性,对任何一组通过SSA、WT、EMD或任何其他信号分解方法获得的分量,我们定义这些分量模糊熵的分布即为模糊熵谱模。
B.SSA模糊熵谱
利用SSA分解噪声信号,第一个分量为低频分量,代表了原始信号的主要趋势,而剩余分量则具有明显的震荡和高频特性。
图1为piecewise-regular信号在信噪比0dB(左)和15dB(右)下的模糊熵谱(实线)和奇异谱(点线),无噪(虚线)及含噪(点划线)信号的模糊熵以及高
斯白噪声模糊熵的95%置信带(矩形)也置于图形中供比较。图2提供了Riemann信号的同样信息,两图中的模糊熵排列对应于奇异值从大到小的顺序。
分析图1和图2,我们可以得出如下结论:第一,很多情况下,并无所谓的奇异谱噪声平面存在,比如我们考察的信号中,piecewise-regular信号在两个信噪比下,Riemann信号在15dB下均无平坦的噪声平面存在,但我们提出的模糊熵谱则清晰地勾画出了每个分量的相对噪声水平,提供了另一个“噪声平面”来有效地表征每个分量相对于白噪声、干净信号和含噪信号的噪声水平;第二,在每种情况下,第一个SSA分量的模糊熵均与干净信号的模糊熵非常接近;最后,也是最重要的发现是奇异谱对应的高频分量并不都在白噪声95%置信带内,说明这里高频分量中包含信号成分,也表明截断SSA剔除所有高频分量会导致信息损失,影响去噪性能。
C.迭代SSA软阈值去噪方法
图3为迭代SSA软阈值去噪算法原理图,根据前述对噪声信号模糊熵谱的考察,模糊熵值大于原始噪声信号特别是那些逼近或在白噪声95%置信带内的SSA分量,需要较大的阈值来滤波,另一方面,模糊熵值小于噪声信号的分量具有更多的信号成分,需要较小的阈值,特别是第一个分量,包含了信号的主要趋势,并且其模糊熵值与干净信号非常接近,因此第一分量应完全保留在信号恢复中,据此,迭代SSA软阈值去噪算法归纳如下:
1.SSA分解:噪声信号xin={x1,x2,…,xN}嵌入为m×d的Hankel矩阵,SVD分解Hankel矩阵为d个秩1矩阵之和,并重构为d个分量
2.计算模糊熵谱:按模糊熵谱公式(14)计算SSA分量的模糊熵谱和原始
噪声信号的模糊熵值;
3.设置阈值:第一分量x1不做任何阈值去噪,
对模糊熵值小于噪声信号模糊熵值的分量x2到xk,设置较小阈值:
这里σc为分量xc方差,N为信号长度,
剩余分量的模糊熵值均不小于噪声信号,设置较大阈值:
4.去噪:除第一分量的所有分量xc(c=2,…,d),利用相应的阈值公式(15)或(16)进行去噪:
5.信号恢复:第一次去噪信号为第一分量和其余软阈值去噪分量之和:
III.性能评估
除了piecewise-regular和Riemann信号,我们还利用另外两个代表性的合成信号,即blocks和sineoneoverx信号来表明我们去噪算法的性能,两个真实的实验信号,即语言和肌电信号也被用来评估该算法,并和截断SSA,WT,及EMD去噪算法进行比较,四个合成信号样本均采用了四个不同采样频率,生成长度分别为1024,2048,4096,和8192的样本,性能参数为50个噪声样本去噪后信噪比的均值。
A.迭代次数效果
图4以piecewise-regular信号为例,说明了SSA-IST算法中迭代机制的效果,图4(a)为一长度8192信噪比15的piecewise-regular信号样本,图4(b)为SSA-IST去噪一次后信号波形,图4(c)和(d)分别为SSA-IST算法迭代4和14次的输出,从这些时域波形上我们可以看出随迭代次数的增加,噪声被显著压制,图4(e)定量显示了SNR随迭代次数提高的效果,通过14次的迭代,含噪信号SNR从15dB提高到约25dB,这充分表明了SSA-IST算法去噪的有效性。
B.滤波合成信号的性能
图5显示了SSA-IST方法,以及截断SSA,WT和EMD四种去噪方法对四个合成信号的去噪效果,每个子图上的曲线对应于不同样本长度信号去噪后的SNR,按去噪前SNR它们被分左侧的0dB和右侧的15dB,可以看出在所有情况下SSA-IST(实线)去噪后信噪比都明显高于截断SSA(虚线)2到5个dB,另外,在大多数情况下,SSA-IST信噪比改善稍好于或非常接近于EMD(点线),唯一例外是在sineoneoverx信号长度8192情况下,EMD优于SSA-IST,最后,SSA-IST与WT(点
划线)相比,在低信噪比的情况下,SSA-IST去噪的优势更为明显,当SNR为15dB时,WT趋向于和SSA-IST性能相似或稍优,这种趋势在sineoneoverx信号中表现尤为明显。
C.滤波实验信号的性能
我们还评估了SSA-IST对两种真实实验信号,即语音和肌电信号的去噪性能,语音信号的采样频率为16kHz,肌电信号的采样频率为1kHz。表1为SSA-IST和其他三种方法对两种实验信号去噪后的信噪比总结,多个样本去噪后信噪比的方差是另一个,特别是针对真实实验信号有效的去噪性能评估方法,四种方法去噪后SNRs的方差也同时列于表1。与合成信号类似,在所有情况下SSA-IST去噪后的信噪比均高于截断SSA法,此外,除了语音信号15dB,SSA-IST去噪后信噪比在其他情况下比WT均高出1–2dB,与SSA-IST相比,EMD是较具竞争力的方法,在不同输入信号信噪比下,EMD对语音信号去噪的信噪比改善都优于SSA-IST,但是对肌电信号,SSA-IST在不同信噪比下都优于EMD,并且在两种信号情况下,SSA-IST去噪SNRs的方差均小于EMD法。
在所测试的六个信号中,几种方法对肌电信号去噪效果最差,其次是Riemann信号,这可能与这两种信号的复杂成分相关,EMG信号中包含了很多棘波,而Riemann信号具有明显的高频成分,波形与1/f噪声极为相似。
表1.四种方法对语音和肌电信号去噪后的信噪比(SNRs)以及信噪比的方差
IV.总结
传统的截断SSA滤波方法采用了保留部分分量,舍弃其他分量的二元方法,在频域等价于低通滤波,损失了高频带的信息,另外,该方法依赖于主观查找很多情况下并不存在的噪声平面,本发明首先提出了用模糊熵谱来替代奇异谱,无论信号性质和噪声水平,模糊熵谱均能准确提供含噪信号各分量的相对噪声水平,这为发明不同信号去噪或其他处理方法提供了重要基础。在模糊熵谱的基础上,我们发明了迭代奇异谱软阈值去噪算法,通过对四种仿真信号和两个实验信号的在不同信噪比下去噪的验证,我们发明的SSA-IST均比截断SSA更能有效地提高含噪信号信噪比,此发明可以广泛应用于移动装备,助听器,可穿戴装备,医疗器械或生物医学信号,机械信号,雷达信号的去噪。
以上所述的具体实施例,对本发明的目的、技术方案和有益效果进行了进一步详细说明,所应理解的是,以上所述仅为本发明的具体实施例而已,并不用于限制本发明。本发明扩展到任何在本说明书中披露的新特征或任何新的组合,以及披露的任一新的方法或过程的步骤或任何新的组合。
Claims (2)
- 一种基于模糊熵的迭代奇异谱软阈值去噪方法,其特征在于,包括如下步骤:(2)计算模糊熵谱:按权利要求1所述模糊熵谱定义计算SSA分量的模糊熵谱和原始噪声信号的模糊熵值;(4)软阈值去噪:除第一分量的所有分量xc(c=2,…,d),利用相应的阈值进行去噪,即每一分量的某噪声信号数值绝对值如小于该分量阈值,输出为零,如信号数值大于阈值,则输出为信号数值减去阈值,剩余情况输出为信号数值加上阈值,软阈值去噪后所有分量(c=2,…,d)与x1之和为第一次估计信号估计噪声为(6)比较连续迭代获得的噪声方差,如噪声方差不再明显减小或达到指定迭代次数,迭代停止,否则重复(1)到(5);
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611010791 | 2016-11-17 | ||
CN201611010791.1 | 2016-11-17 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2018090467A1 true WO2018090467A1 (zh) | 2018-05-24 |
Family
ID=61154966
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2016/113611 WO2018090467A1 (zh) | 2016-11-17 | 2016-12-30 | 基于模糊熵的含噪信号处理方法及迭代奇异谱软阈值去噪方法 |
Country Status (3)
Country | Link |
---|---|
US (1) | US10361680B2 (zh) |
CN (1) | CN107688120A (zh) |
WO (1) | WO2018090467A1 (zh) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110108689A (zh) * | 2019-05-17 | 2019-08-09 | 广西科技大学 | 快速消除3d荧光本底噪声的方法 |
CN110349106A (zh) * | 2019-07-09 | 2019-10-18 | 北京理工大学 | 一种基于Renyi熵的小波软阈值图像去噪方法 |
CN111581596A (zh) * | 2020-05-29 | 2020-08-25 | 广东电网有限责任公司 | 一种变压器油中溶解气体浓度预测方法 |
CN111650654A (zh) * | 2020-05-13 | 2020-09-11 | 吉林大学 | 联合emd与wt算法的地面磁共振信号尖峰噪声剔除方法 |
CN112084845A (zh) * | 2020-07-30 | 2020-12-15 | 中国人民解放军63892部队 | 基于多尺度小波系数自相关的低频1/f噪声消除方法 |
CN114900246A (zh) * | 2022-05-25 | 2022-08-12 | 中国电子科技集团公司第十研究所 | 噪声基底估计方法、装置、设备及存储介质 |
CN116804971A (zh) * | 2023-08-22 | 2023-09-26 | 上海安般信息科技有限公司 | 一种基于信息熵的模糊测试方法 |
Families Citing this family (34)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10775418B2 (en) * | 2017-01-03 | 2020-09-15 | University Of Electronic Science And Technology Of China | Method and system for oscilloscope triggering |
CN108551412B (zh) * | 2018-05-03 | 2021-05-25 | 网宿科技股份有限公司 | 监控数据降噪处理方法和装置 |
CN108801630B (zh) * | 2018-06-22 | 2020-04-14 | 石家庄铁道大学 | 单通道盲源分离的齿轮故障诊断方法 |
CN109635699B (zh) * | 2018-12-05 | 2023-08-29 | 陕西科技大学 | 一种基于k-svd与omp算法结合的心电信号去噪方法 |
CN109688078B (zh) * | 2018-12-24 | 2021-06-08 | 长安大学 | 一种基于混沌同步的弱正弦信号提取方法 |
CN109696598B (zh) * | 2018-12-26 | 2021-02-26 | 四川大学 | 一种考虑继发性故障的配电网故障选线方法 |
CN109815849A (zh) * | 2019-01-02 | 2019-05-28 | 东莞理工学院 | 基于奇异值分解的混沌信号去噪算法 |
CN109657646B (zh) * | 2019-01-07 | 2023-04-07 | 哈尔滨工业大学(深圳) | 生理时间序列的特征表示与提取方法、装置及存储介质 |
CN109858109A (zh) * | 2019-01-14 | 2019-06-07 | 北京工业大学 | 一种基于相关性的emd与形态奇异值分解相结合的齿轮信号降噪方法 |
CN110069868B (zh) * | 2019-04-26 | 2024-04-02 | 中国人民解放军战略支援部队信息工程大学 | Gnss测站非线性运动建模方法与装置 |
US11709245B2 (en) * | 2019-04-29 | 2023-07-25 | Adnoviv Inc. | System and methods for radar-based detection of people in a room |
CN110346143B (zh) * | 2019-08-06 | 2021-02-09 | 江苏今创车辆有限公司 | 超高频噪声辅助经验模态分解方法 |
CN110490257A (zh) * | 2019-08-21 | 2019-11-22 | 四川长虹电器股份有限公司 | 一种基于去趋势项的生理电信号熵分析方法 |
CN110503060B (zh) * | 2019-08-28 | 2022-05-10 | 中南大学 | 一种光谱信号去噪方法及其系统 |
CN110659621A (zh) * | 2019-09-27 | 2020-01-07 | 山东科技大学 | 一种基于变分模态分解和排列熵的联合降噪方法 |
CN110909480B (zh) * | 2019-11-29 | 2023-03-14 | 华北水利水电大学 | 一种水轮机振动信号的去噪方法与装置 |
CN111276154B (zh) * | 2020-02-26 | 2022-12-09 | 中国电子科技集团公司第三研究所 | 风噪声抑制方法与系统以及炮声检测方法与系统 |
CN111487063B (zh) * | 2020-03-20 | 2021-05-14 | 天津大学 | 一种参数自适应的内燃机噪声源分离方法 |
CN112350787B (zh) * | 2020-10-13 | 2021-09-07 | 电子科技大学 | 一种无线电信号异常检测方法 |
CN112284520B (zh) * | 2020-10-25 | 2022-06-24 | 广东石油化工学院 | 一种利用最佳秩逼近的振声检测信号重构方法和系统 |
CN113160069B (zh) * | 2021-02-26 | 2024-02-02 | 桂林电子科技大学 | 一种基于图信号的高光谱图像去噪方法 |
CN113229829B (zh) * | 2021-04-15 | 2023-09-01 | 广东工业大学 | 一种四元数脑电信号提取方法及系统 |
CN113408336B (zh) * | 2021-05-08 | 2022-06-14 | 南昌大学 | 一种基于鲸鱼优化算法的小波去噪最优阈值整定方法 |
CN113541729B (zh) * | 2021-07-12 | 2022-06-07 | 电子科技大学 | 一种基于时频矩阵的时频图去噪方法 |
CN113568058B (zh) * | 2021-07-20 | 2022-06-03 | 湖南师范大学 | 一种基于多分辨率奇异值分解的大地电磁信噪分离方法及系统 |
CN113640891B (zh) * | 2021-08-11 | 2022-11-08 | 吉林大学 | 一种基于奇异谱分析的瞬变电磁探测数据噪声滤除方法 |
CN113820564B (zh) * | 2021-09-24 | 2023-03-10 | 国家电网有限公司 | 一种适应于源网荷储复杂电网的故障检测方法 |
CN113974658B (zh) * | 2021-10-28 | 2024-01-26 | 天津大学 | 基于eeg分时频谱黎曼的语义视觉图像分类方法及装置 |
CN114288500B (zh) * | 2021-12-02 | 2024-03-19 | 中国科学院深圳先进技术研究院 | 一种基于模糊熵特征提取的人机通气异步检测模型及装置 |
CN114098763B (zh) * | 2021-12-13 | 2023-05-19 | 清华大学深圳国际研究生院 | 一种脑电去噪方法 |
CN115438301B (zh) * | 2022-10-26 | 2023-06-20 | 北京建筑大学 | 一种基于ICEEMDAN二次分解耦合informer模型的设备运行趋势预测方法 |
CN116953784A (zh) * | 2023-07-17 | 2023-10-27 | 哈尔滨工程大学 | 一种冰层振源信号到时拾取方法和装置 |
CN116825121B (zh) * | 2023-08-30 | 2023-11-10 | 北京神码数字技术有限公司 | 基于人工智能的智能语音识别与机器翻译系统 |
CN117743835B (zh) * | 2024-02-20 | 2024-04-30 | 齐鲁工业大学(山东省科学院) | 基于改进蜣螂算法优化变分模态分解的水声信号降噪方法 |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102903083A (zh) * | 2012-09-25 | 2013-01-30 | 哈尔滨工程大学 | 基于Context模型和双树复小波变换的水下声纳图像的去噪方法 |
CN104615877A (zh) * | 2015-01-28 | 2015-05-13 | 辽宁工程技术大学 | 一种基于小波包进行信号去噪的方法 |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TW200734888A (en) * | 2006-03-01 | 2007-09-16 | Univ Nat Chiao Tung | Visualization system of acoustic source energy distribution and the method thereof |
US9532748B2 (en) * | 2013-04-22 | 2017-01-03 | Personal Neuro Devices Inc. | Methods and devices for brain activity monitoring supporting mental state development and training |
-
2016
- 2016-12-30 WO PCT/CN2016/113611 patent/WO2018090467A1/zh active Application Filing
-
2017
- 2017-06-02 US US15/612,567 patent/US10361680B2/en not_active Expired - Fee Related
- 2017-08-25 CN CN201710742788.7A patent/CN107688120A/zh active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102903083A (zh) * | 2012-09-25 | 2013-01-30 | 哈尔滨工程大学 | 基于Context模型和双树复小波变换的水下声纳图像的去噪方法 |
CN104615877A (zh) * | 2015-01-28 | 2015-05-13 | 辽宁工程技术大学 | 一种基于小波包进行信号去噪的方法 |
Non-Patent Citations (1)
Title |
---|
"Denoising Nonlinear Time Series Using Singular Spectrum Analysis and Fuzzy Entropy", CHINESE PHYSICS LETTERS, 31 October 2016 (2016-10-31) * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110108689A (zh) * | 2019-05-17 | 2019-08-09 | 广西科技大学 | 快速消除3d荧光本底噪声的方法 |
CN110108689B (zh) * | 2019-05-17 | 2021-07-30 | 广西科技大学 | 快速消除3d荧光本底噪声的方法 |
CN110349106A (zh) * | 2019-07-09 | 2019-10-18 | 北京理工大学 | 一种基于Renyi熵的小波软阈值图像去噪方法 |
CN111650654A (zh) * | 2020-05-13 | 2020-09-11 | 吉林大学 | 联合emd与wt算法的地面磁共振信号尖峰噪声剔除方法 |
CN111650654B (zh) * | 2020-05-13 | 2022-03-22 | 吉林大学 | 联合emd与wt算法的地面磁共振信号尖峰噪声剔除方法 |
CN111581596A (zh) * | 2020-05-29 | 2020-08-25 | 广东电网有限责任公司 | 一种变压器油中溶解气体浓度预测方法 |
CN112084845A (zh) * | 2020-07-30 | 2020-12-15 | 中国人民解放军63892部队 | 基于多尺度小波系数自相关的低频1/f噪声消除方法 |
CN112084845B (zh) * | 2020-07-30 | 2023-05-02 | 中国人民解放军63892部队 | 基于多尺度小波系数自相关的低频1/f噪声消除方法 |
CN114900246A (zh) * | 2022-05-25 | 2022-08-12 | 中国电子科技集团公司第十研究所 | 噪声基底估计方法、装置、设备及存储介质 |
CN116804971A (zh) * | 2023-08-22 | 2023-09-26 | 上海安般信息科技有限公司 | 一种基于信息熵的模糊测试方法 |
CN116804971B (zh) * | 2023-08-22 | 2023-11-07 | 上海安般信息科技有限公司 | 一种基于信息熵的模糊测试方法 |
Also Published As
Publication number | Publication date |
---|---|
CN107688120A (zh) | 2018-02-13 |
US20180138896A1 (en) | 2018-05-17 |
US10361680B2 (en) | 2019-07-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2018090467A1 (zh) | 基于模糊熵的含噪信号处理方法及迭代奇异谱软阈值去噪方法 | |
Smital et al. | Adaptive wavelet wiener filtering of ECG signals | |
Tsolis et al. | Signal denoising using empirical mode decomposition and higher order statistics | |
Damasevicius et al. | BoostEMD: an extension of EMD method and its application for denoising of EMG signals | |
Phinyomark et al. | EMG denoising estimation based on adaptive wavelet thresholding for multifunction myoelectric control | |
El Bouny et al. | ECG signal denoising based on ensemble emd thresholding and higher order statistics | |
Georgieva-Tsaneva | Wavelet based interval varying algorithm for optimal non-stationary signal denoising | |
Patil et al. | New approach of threshold estimation for denoising ECG signal using wavelet transform | |
Isa et al. | Optimal selection of wavelet thresholding algorithm for ECG signal denoising | |
de Sá et al. | A principal component-based algorithm for denoising in single channel data (PCA for denoising in single channel data) | |
CN117158999A (zh) | 一种基于ppmcc和自适应vmd的脑电信号去噪方法及系统 | |
ÖZAYDIN et al. | Speech enhancement using maximal overlap discrete wavelet transform | |
Sasirekha et al. | A novel wavelet based thresholding for denoising fingerprint image | |
Rajankar et al. | An optimum ECG denoising with wavelet neural network | |
Hao et al. | An improved multivariate wavelet denoising method using subspace projection | |
Khaldi et al. | Voiced speech enhancement based on adaptive filtering of selected intrinsic mode functions | |
Xie et al. | Fuzzy entropy spectrum analysis for biomedical signals de-noising | |
Oktar et al. | Speech denoising using discrete wavelet packet decomposition technique | |
Gandham et al. | An iterative method of Ensemble Empirical Mode Decomposition for enhanced ECG signal denoising | |
Adamou-Mitiche et al. | Three levels discrete wavelet transform elliptic estimation for ECG denoising | |
Hadhami et al. | Speech denoising based on empirical mode decomposition and improved thresholding | |
Gon et al. | Removal of Noises from an ECG Signal Using an Adaptive S-Median Thresholding Technique | |
Bansod et al. | A new approach for removal of baseline wande in ECG signal using Empirical Mode Decomposition & Hurst Exponent | |
NSVN et al. | Optimal threshold estimation using cultural algorithm for EMD-DWT based ECG denoising | |
Mitra et al. | Comparative study on thresholding techniques of discrete wavelet transform (DWT) to de-noise corrupted ECG signals |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 16921804 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 16921804 Country of ref document: EP Kind code of ref document: A1 |