A wavelet denoising method based on improved threshold function
By designing the adjustable threshold function in the wavelet denoising method, the constant deviation and discontinuity of the threshold function in the prior art are solved, and better denoising effect and signal feature retention are achieved, which is suitable for signal processing of different noise intensity. .
Patent Information
- Application Number
- CN202210275028.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-21
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2042-03-21
AI Technical Summary
In the existing wavelet denoising method, the threshold function has constant deviation, discontinuity and structural fixedness, resulting in poor denoising effect.
A wavelet denoising method based on an improved threshold function is adopted to calculate the critical threshold by estimating the noise variance, and an adjustable threshold function is designed, combining wavelet transform and inverse transform to realize signal denoising processing. The adjustable threshold function is continuous and highly derivable within the domain, and can be converted into soft or hard threshold function, suitable for different data systems.
It effectively overcomes the shortcomings of soft and hard threshold functions, reduces signal distortion and low-frequency information loss, significantly improves the noise denoising effect, retains the local characteristics of the signal and reduces the noise impact.
Smart Images

Figure CN114581674B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to signal processing, in particular to a wavelet denoising method based on an improved threshold function. Background Art
[0002] In real life, noise is everywhere. FIR, IIR and moving average methods lead to weak signal correlation after denoising. Wavelet denoising improves the shortcomings of general denoising methods that globally filter signals in the frequency domain, and has the same characteristics in the time domain. It also has advantages such as low entropy, and has obvious advantages in denoising signals. Therefore, wavelet denoising has been widely used in many fields such as astronomy, medical imaging and computer vision.
[0003] Noise is filtered out in different stages of wavelet transform: wavelet transform threshold denoising method, wavelet coefficient spatial correlation denoising method and wavelet coefficient modulus maximum denoising method (Sheng Jing, Zeng Yuping, Liu Guoman, et al. Determination of Weak Knock Characteristics for Two-Stroke Spark Ignition UAV Engines Based on Mallat Decomposition Algorithm [J]. Mathematical Problems in Engineering, 2021, 2021:). Since the wavelet threshold denoising algorithm is simple and easy to understand, has better noise filtering ability, and has a small running time and workload, it is also widely used in denoising processing.
[0004] Generally, wavelet threshold denoising algorithms use soft and hard threshold functions to filter out noise components in the signal, but the soft threshold function has a constant deviation problem, which leads to a decrease in the accuracy of the denoised signal. The discontinuity of the hard threshold function itself will cause local abnormal spikes in the denoised signal (Guo Hua, Yue Leihui, Song Peng, et al. Denoising of an ultraviolet light received signal based on improved wavelet transform threshold and threshold function. [J]. Applied optics, 2021, 60(28):). Prior art Wu Anquan, Shen Changsheng, Xiao Jinbiao, et al. Wavelet denoising based on a progressive semi-soft threshold function [J]. Electronic Devices, 2017, 40(02): 396-399. The semi-soft threshold function is proposed to improve the traditional threshold function, but it does not overcome the shortcomings of the soft and hard thresholds, so there are still problems of deviation and discontinuity. Another example is the prior art Hong YeGao. Wavelet Shrinkage Denoising Using the Non-Negative Garrote[J]. Journal of Computational and Graphical Statistics, 2012, 7(4): The Garrote threshold function is proposed. Although it improves the shortcomings of the soft and hard threshold functions, as the number of wavelet transform layers increases, the useful signal will also be filtered out, and the signal will be distorted. Huo Yuanlian, Zhang Jian, Lian Peijun, et al. Research on lightning electric field signal denoising based on improved wavelet threshold function[J]. Journal of Sensor Technology, 2021, 34(02): 218-222. The advantages of the above threshold functions are summarized, and the problems of soft and hard threshold functions are solved by changing the construction form of the threshold function. Although these improved threshold functions solve the shortcomings of soft and hard threshold functions, the function form is generally fixed.
[0005] In summary, the existing wavelet denoising still has the problem that the threshold function has constant deviation, discontinuity, fixed structure and lacks flexibility and inclusiveness, resulting in poor denoising effect. Summary of the invention
[0006] Purpose of the invention: In view of the above shortcomings, the present invention provides a wavelet denoising method based on an improved threshold function with continuity, flexibility and small constant deviation.
[0007] Technical solution: To solve the above problems, the present invention adopts a wavelet denoising method based on an improved threshold function, comprising the following steps:
[0008] (1) Input a noisy signal;
[0009] (2) Select the wavelet basis function and decompose the noisy signal through the wavelet basis to obtain the wavelet decomposition coefficient w j,k ;
[0010] (3) Estimate the noise variance σ in the noisy signal and find the value of the critical threshold λ;
[0011] (4) Improving the threshold function to obtain an adjustable threshold function;
[0012]
[0013] Among them, a and b are adjustment parameters of the adjustable threshold function, and λ is the critical threshold;
[0014] According to the improved adjustable threshold function, the wavelet coefficient w j,k Perform filtering to obtain the estimated wavelet coefficients
[0015] (5) Use inverse wavelet transform to reconstruct the signal and obtain the denoised signal.
[0016] Furthermore, the calculation formula of the critical threshold value λ in step (3) is:
[0017]
[0018] Wherein, N is the length of the noisy signal. The noise variance σ is estimated by the formula:
[0019]
[0020] Among them, w 1,k is the high frequency coefficient after the first wavelet transform, 0.6745 is the adjustment coefficient of the noise standard deviation, and median represents the median of the sequence arranged in order of size.
[0021] Furthermore, in step (4), a∈[0,1], b>0; the values of a and b are set to apply different denoising situations. When a=0 or b=1, the adjustable threshold function is converted into a soft threshold function; when b→0, the adjustable threshold function is approximately converted into a hard threshold function.
[0022] The hard threshold function formula is:
[0023]
[0024] The soft threshold function formula is:
[0025]
[0026] Furthermore, the noisy signal is a signal containing Gaussian white noise, and after the signal is denoised, the denoising ability is detected using the signal-to-noise ratio (SNR) and the Pearson coefficient.
[0027] The signal-to-noise ratio SNR coefficient is:
[0028]
[0029] The Pearson coefficient is:
[0030]
[0031] Among them, y(n) is the signal sequence reconstructed by the threshold function, x(n) is the original signal sequence, and N is the data length.
[0032] Beneficial effects: Compared with the prior art, the significant advantage of the present invention is that the adjustable threshold function can be converted into a hard threshold function or a soft threshold function to process the signal when taking a special value. In the application, the values of a and b can be set to denoise different data systems. The adjustable threshold function is continuous in the definition domain and is high-order differentiable, which makes up for the deficiency of the hard threshold function that produces discontinuity at ±λ, and no additional oscillation points will appear, which can better retain the energy of the decomposed wavelet coefficients. The asymptote of the adjustable threshold function is It can well preserve the local characteristics of the signal and reduce the influence of uncontrollable noise. j,k As | increases, the signal gradually increases and the noise gradually decreases. With w j,k The deviation between them gradually decreases, overcoming the problems existing in the soft threshold function and effectively reducing the loss of low-frequency signal information. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 A waveform comparison diagram of the adjustable threshold function of the present invention and the traditional threshold function;
[0034] Figure 2 It is a flow chart of the wavelet denoising method based on the improved threshold function of the present invention;
[0035] Figure 3 (a) is the noise signal, Figure 3 (b) is the denoising effect of the hard threshold function. Figure 3 (c) is the denoising effect of the soft threshold function. Figure 3 (d) is the denoising effect diagram of the adjustable threshold function;
[0036] Figure 4 (a) is a comparison chart of SNR correlation coefficient evaluation noise performance under different noise intensities; Figure 4(b) is a comparison chart of noise performance evaluation using the Pearson correlation coefficient under different noise intensities. DETAILED DESCRIPTION
[0037] like Figure 2 As shown, in this embodiment, a wavelet denoising method based on an improved threshold function includes the following steps:
[0038] (1) Input a noisy signal;
[0039] (2) Select the wavelet basis function and decompose the noisy signal through the wavelet basis to obtain the wavelet decomposition coefficient w j,k ;
[0040] (3) Estimate the noise variance σ in the noisy signal and find the value of the critical threshold λ;
[0041] The VisuShrink threshold is taken as the judgment threshold, the wavelet coefficients larger than the threshold are retained, and the wavelet coefficients smaller than the threshold are discarded to achieve signal denoising, which has strong applicability. The calculation formula of the λ value is:
[0042]
[0043] Where N is the signal length, w 1,k It represents the high frequency coefficient after the first wavelet transform, and 0.6745 is the adjustment coefficient of the noise standard deviation.
[0044] (4) Improving the threshold function to obtain an adjustable threshold function;
[0045] The hard threshold function formula is:
[0046]
[0047] The soft threshold function formula is:
[0048]
[0049] in, represents the wavelet coefficient after threshold processing, that is, the estimated wavelet coefficient, w j,k represents the kth wavelet coefficient on the jth scale before threshold processing, and λ represents the critical threshold.
[0050] An improved threshold function is disclosed in the prior art, which improves the shortcomings of the semi-soft threshold function that does not overcome the soft and hard threshold functions and the Garrote threshold function that causes the loss of useful signals. The function is:
[0051]
[0052] This threshold function improves the shortcomings of the above-mentioned soft and hard threshold functions, which have constant deviations and discontinuities and cause large signal distortion. However, this threshold function has low flexibility and cannot meet the requirements of soft and hard threshold functions in special cases.
[0053] In view of the problem that the constant deviation of the soft threshold function and the discontinuity of the hard threshold function lead to local oscillation and edge blur in the denoised signal, a new threshold function is obtained through improvement to overcome the defects of the soft and hard threshold functions. The threshold estimator with exponential parameters introduced in this embodiment satisfies high-order differentiability. The proposed new threshold function is an adjustable threshold function, and its mathematical model is as follows:
[0054]
[0055] Among them, a and b are adjustment parameters of the adjustable threshold function, a∈[0,1], b>0.
[0056] like Figure 1 As shown in the figure, when a=0 or b=1, the adjustable threshold function can be converted into a soft threshold function. When b→0, the adjustable threshold function can be approximately converted into a hard threshold function. The adjustable threshold function can be converted into a hard threshold function or a soft threshold function to process the signal when taking a special value. In the application, the values of a and b can be set to denoise different data systems; the adjustable threshold function is continuous within the definition domain and is highly differentiable, which makes up for the deficiency of the hard threshold function that produces discontinuities at ±λ, and there will be no additional oscillation points, which can better retain the energy of the decomposed wavelet coefficients; the asymptote of the adjustable threshold function is It can well preserve the local characteristics of the signal and reduce the influence of uncontrollable noise. j,k As | increases, the signal gradually increases and the noise gradually decreases. With w j,k The deviation between them gradually decreases, overcoming the problems of the soft threshold function and effectively reducing the loss of low-frequency signal information. However, it is necessary to prevent it from being reduced to 0, turning the function into a hard threshold function.
[0057] Taking λ as the threshold, the wavelet coefficient w is j,k Perform filtering to obtain the estimated wavelet coefficients
[0058] (5) The estimated wavelet coefficients obtained in step (4) The wavelet approximation value obtained after decomposition in step (2) is reconstructed to complete the denoising process and obtain the processed signal.
[0059] (6) The denoising effect of the improved method was measured; the wavelet threshold denoising algorithm based on the adjustable threshold function was used to denoise the Gaussian white noise signal, and the denoising ability was tested using SNR and Pearson correlation coefficient. The SNR and Pearson correlation coefficient are as follows:
[0060]
[0061]
[0062] Among them, y(n) is the signal sequence reconstructed by the threshold function, x(n) is the original signal sequence, and N is the data length.
[0063] The application effect of the wavelet denoising method in this embodiment is described in detail below in combination with simulation.
[0064] In order to evaluate the performance of the wavelet denoising method in this embodiment, the simulation experiment uses a sinusoidal signal containing Gaussian white noise. The ideal sinusoidal signal amplitude is 1, the wavelet basis function is coif3, and the decomposition level is 5.
[0065] Simulation experiment 1: Comparison of noisy signal denoising results using different methods
[0066] In order to verify the noise suppression ability of the wavelet denoising method in this embodiment on the signal containing Gaussian white noise, the wavelet denoising method based on the soft threshold function and the wavelet denoising method based on the hard threshold function are used as reference methods to compare and analyze the effectiveness and superiority of the method of the present invention.
[0067] In the experiment, the SNR of the signal containing Gaussian white noise is 3dB, and the a and b of the adjustable threshold function are 0.5 and 3 respectively.
[0068] Depend on Figure 3 (b) It can be seen that after the wavelet denoising algorithm based on the hard threshold function processes the signal, the "Gibbs phenomenon" occurs, there are many abnormal spikes and glitches, the noise removal effect is poor, and the waveform distortion is large; Figure 3 (c) It can be seen that after the wavelet denoising algorithm based on the soft threshold function processes the signal, the abnormal peaks of the signal waveform are reduced, but they cannot be completely removed. Small burrs appear near x=100, and the signal amplitude change range between 200 and 300 samples is significantly reduced, and some useful signals are lost. The denoising effect needs to be improved. Figure 3(d) It can be observed that after the improved wavelet threshold denoising algorithm is processed, there is basically no spike burr in the signal waveform, and there is only a small spike at x=100. Compared with the method of filtering noise with a hard threshold function, the denoising effect of the algorithm proposed in this embodiment is significantly improved; and the waveform amplitude change of the signal after the algorithm proposed in this embodiment is relatively ideal, which improves the shortcoming that the reconstructed signal lacks a large amount of useful signals after filtering the noise part with a soft threshold function. In summary, the improved wavelet denoising algorithm based on the adjustable threshold function reduces the distortion of the signal, and the burr value of the overall signal is very rare, which is closer to the ideal signal, the information is relatively complete, and the denoising effect is significantly improved.
[0069] Simulation experiment 2: De-noising performance comparison at different noise intensities
[0070] To verify the applicability of the method of the present invention, noisy signals with different noise intensities were selected for denoising, and SNR and Pearson correlation coefficient were used for evaluation. Signals with SNRs of -3dB, 0dB, 3dB, 5dB, 8dB, and 10dB were selected for experiments.
[0071] according to Figure 4 (a) It can be seen that when the signal SNR is small, the adjustable threshold function has a similar denoising effect to the soft and hard threshold functions, but still maintains a better denoising effect. When the signal SNR increases within a certain range, it can be intuitively observed that the adjustable threshold function has a more significant denoising processing ability. Compared with the soft and hard threshold functions, the adjustable threshold function gradually has a higher leading level in SNR. And by Figure 4 (b) It can be seen that the signal processed by the wavelet denoising algorithm based on the adjustable threshold function has always maintained a higher level of approximation to the ideal signal, and has a greater advantage in denoising effect. Therefore, when processing noisy signals with different SNRs, the adjustable threshold function has a greatly improved noise reduction ability compared to the soft and hard threshold functions, proving that the wavelet denoising algorithm based on the adjustable threshold function has ideal applicability and superiority for noise intensity signals within a certain range.
[0072] Based on the above experimental results, the following conclusions are obtained:
[0073] An adjustable threshold function with continuity, flexibility and small constant deviation designed in this embodiment can effectively solve the shortcomings of soft and hard threshold functions. A wavelet denoising method based on the adjustable threshold function is further proposed, which can suppress noise on signals containing Gaussian white noise. The method is flexible and easy to implement. When the signal SNR increases within a certain range, the denoising algorithm proposed in this article has more significant advantages.
Claims
1. A wavelet denoising method based on an improved threshold function, characterized in that: The following steps are involved: (1) Input a noisy signal; (2) Select the wavelet basis function and decompose the noisy signal through the wavelet basis to obtain the wavelet decomposition coefficient w j,k ; (3) Estimate the noise variance σ in the noisy signal and find the value of the critical threshold λ; (4) Improving the threshold function to obtain an adjustable threshold function; Among them, a and b are adjustment parameters of the adjustable threshold function, λ is the critical threshold; a∈[0,1], b>0; setting a and b values is suitable for different denoising situations. When a=0 or b=1, the adjustable threshold function is converted into a soft threshold function; when b→0, the adjustable threshold function is approximately converted into a hard threshold function; The hard threshold function formula is: The soft threshold function formula is: According to the improved adjustable threshold function, the wavelet coefficient w j,k Perform filtering to obtain the estimated wavelet coefficients (5) Use inverse wavelet transform to reconstruct the signal and obtain the denoised signal.
2. The wavelet denoising method according to claim 1, characterized in that: The calculation formula of the critical threshold value λ in step (3) is: Where N is the length of the noisy signal.
3. The wavelet denoising method according to claim 2, characterized in that: The noise variance σ is estimated by: Among them, w 1,k is the high frequency coefficient after the first wavelet transform, and 0.6745 is the adjustment coefficient of the noise standard deviation.
4. The wavelet denoising method according to claim 1, characterized in that: The noisy signal is a signal containing Gaussian white noise. After the signal is denoised, the signal-to-noise ratio (SNR) and the Pearson coefficient are used to detect the denoising ability.
5. The wavelet denoising method according to claim 4, characterized in that: The signal-to-noise ratio SNR coefficient is: The Pearson coefficient is: Among them, y(n) is the signal sequence reconstructed by the threshold function, x(n) is the original signal sequence, and N is the data length.
Citation Information
Cited By
Vehicle vibration signal denoising method based on wavelet decomposition
CN116383605A
Vehicle vibration signal denoising method based on wavelet decomposition
CN116383605B
