Research article

Surface electromyography signal denoising via EEMD and improved wavelet thresholds

  • Received: 02 August 2020 Accepted: 08 October 2020 Published: 16 October 2020
  • The acquisition of good surface electromyography (sEMG) is an important prerequisite for correct and timely control of prosthetic limb movements. sEMG is nonlinear, nonstationary, and vulnerable against noise and a new sEMG denoising method using ensemble empirical mode decomposition (EEMD) and wavelet threshold is hence proposed to remove the random noise from the sEMG signal. With this method, the noised sEMG signal is first decomposed into several intrinsic mode functions (IMFs) by EEMD. The first IMF is mostly noise, coupled with a small useful component which is extracted using a wavelet transform based method by defining a peak-to-sum ratio and a noise-independent extracting threshold function. Other IMFs are processed using an improved wavelet threshold denoising method, where a noise variance estimation algorithm and an improved wavelet threshold function are combined. Key to the threshold denoising method, a threshold function is used to retain the required wavelet coefficients. Our denoising algorithm is tested for different sEMG signals produced by different muscles and motions. Experimental results show that the proposed new method performs better than other methods including the conventional wavelet threshold method and the EMD method, which guaranteed its usability in prosthetic limb control.

    Citation: Ziyang Sun, Xugang Xi, Changmin Yuan, Yong Yang, Xian Hua. Surface electromyography signal denoising via EEMD and improved wavelet thresholds[J]. Mathematical Biosciences and Engineering, 2020, 17(6): 6945-6962. doi: 10.3934/mbe.2020359

    Related Papers:

    [1] Ravichandra Madanu, Farhan Rahman, Maysam F. Abbod, Shou-Zen Fan, Jiann-Shing Shieh . Depth of anesthesia prediction via EEG signals using convolutional neural network and ensemble empirical mode decomposition. Mathematical Biosciences and Engineering, 2021, 18(5): 5047-5068. doi: 10.3934/mbe.2021257
    [2] Tao Yang, Qicheng Yang, Yibo Zhou, Chuanbiao Wen . Glucose trend prediction model based on improved wavelet transform and gated recurrent unit. Mathematical Biosciences and Engineering, 2023, 20(9): 17037-17056. doi: 10.3934/mbe.2023760
    [3] Konki Sravan Kumar, Daehyun Lee, Ankhzaya Jamsrandoj, Necla Nisa Soylu, Dawoon Jung, Jinwook Kim, Kyung Ryoul Mun . sEMG-based Sarcopenia risk classification using empirical mode decomposition and machine learning algorithms. Mathematical Biosciences and Engineering, 2024, 21(2): 2901-2921. doi: 10.3934/mbe.2024129
    [4] Viliam Ďuriš, Vladimir I. Semenov, Sergey G. Chumarov . Wavelets and digital filters designed and synthesized in the time and frequency domains. Mathematical Biosciences and Engineering, 2022, 19(3): 3056-3068. doi: 10.3934/mbe.2022141
    [5] Rakesh Pilkar, Erik M. Bollt, Charles Robinson . Empirical mode decomposition/Hilbert transform analysis of postural responses to small amplitude anterior-posterior sinusoidal translations of varying frequencies. Mathematical Biosciences and Engineering, 2011, 8(4): 1085-1097. doi: 10.3934/mbe.2011.8.1085
    [6] Xiaoguang Liu, Mingjin Zhang, Jiawei Wang, Xiaodong Wang, Tie Liang, Jun Li, Peng Xiong, Xiuling Liu . Gesture recognition of continuous wavelet transform and deep convolution attention network. Mathematical Biosciences and Engineering, 2023, 20(6): 11139-11154. doi: 10.3934/mbe.2023493
    [7] Xuyang Xie, Zichun Yang, Lei Zhang, Guoqing Zeng, Xuefeng Wang, Peng Zhang, Guobing Chen . An improved Autogram and MOMEDA method to detect weak compound fault in rolling bearings. Mathematical Biosciences and Engineering, 2022, 19(10): 10424-10444. doi: 10.3934/mbe.2022488
    [8] Lihe Liang, Jinying Cui, Juanjuan Zhao, Yan Qiang, Qianqian Yang . Ultra-short-term forecasting model of power load based on fusion of power spectral density and Morlet wavelet. Mathematical Biosciences and Engineering, 2024, 21(2): 3391-3421. doi: 10.3934/mbe.2024150
    [9] Huanhai Yang, Shue Liu . A prediction model of aquaculture water quality based on multiscale decomposition. Mathematical Biosciences and Engineering, 2021, 18(6): 7561-7579. doi: 10.3934/mbe.2021374
    [10] Yujie Kang, Wenjie Li, Jidong Lv, Ling Zou, Haifeng Shi, Wenjia Liu . Exploring brain dysfunction in IBD: A study of EEG-fMRI source imaging based on empirical mode diagram decomposition. Mathematical Biosciences and Engineering, 2025, 22(4): 962-987. doi: 10.3934/mbe.2025035
  • The acquisition of good surface electromyography (sEMG) is an important prerequisite for correct and timely control of prosthetic limb movements. sEMG is nonlinear, nonstationary, and vulnerable against noise and a new sEMG denoising method using ensemble empirical mode decomposition (EEMD) and wavelet threshold is hence proposed to remove the random noise from the sEMG signal. With this method, the noised sEMG signal is first decomposed into several intrinsic mode functions (IMFs) by EEMD. The first IMF is mostly noise, coupled with a small useful component which is extracted using a wavelet transform based method by defining a peak-to-sum ratio and a noise-independent extracting threshold function. Other IMFs are processed using an improved wavelet threshold denoising method, where a noise variance estimation algorithm and an improved wavelet threshold function are combined. Key to the threshold denoising method, a threshold function is used to retain the required wavelet coefficients. Our denoising algorithm is tested for different sEMG signals produced by different muscles and motions. Experimental results show that the proposed new method performs better than other methods including the conventional wavelet threshold method and the EMD method, which guaranteed its usability in prosthetic limb control.


    The arm is an important part of the human body, and human participation in various social activities is inseparable from the arm. However, war, disease and accidents have caused many people to lose a hand or an entire arm, causing a variety of inconveniences in their lives. As a result, the number of amputees fitted with prosthetic hands or arms has increased over the past years. Therefore, the design and manufacture of a humanoid arm with various functions of human arm and the realization of accurate and stable control of humanoid arm have become the hot spot of rehabilitation medical treatment [1,2].

    During the contraction of skeletal muscle, a series of biochemical changes take place inside the muscle fibers, and at the same time, action potentials are generated. All the action potentials generated by muscle fibers are superimposed on the electrodes attached to the skin, which is the surface electromyography (sEMG) signal. Thus, the sEMG signal can be used as a control signal for the humanoid arm. Compared with Electroencephalogram (EEG) signal and invasive sEMG signal, sEMG signal is the best control signal for artificial arm, because it is easy, cheap and harmless to control prosthetic hand [3,4,5,6,7]. Although surface sEMG signal acquisition has non-invasive advantages, it will be interfered by skin impedance, electromagnetic interference, surrounding environment and many other factors due to the direct use of electrodes to collect signals at the corresponding muscle positions. In addition, surface sEMG signal itself is weak, with low signal-to-noise ratio (SNR) and poor stability. Under ideal experimental conditions, most of the noise can be eliminated. However, human movement is carried out in various unpredictable environments, so the sEMG signals will be mixed with various inevitable noises and artifacts, which will make the control effect of the prosthesis not meet the expectation. Therefore, it is necessary to use an effective algorithm to de-noising the sEMG signals [8,9].

    Wavelet transform has become a common method for nonstationary signal processing [10] and exerts a significant effect. The most common denoising method is wavelet threshold method, including hard and soft thresholds [11]. After processing the wavelet coefficients through the hard threshold method, the wavelet coefficients become discontinuous. Therefore, the original signal may oscillate when it is reconstructed using the processed coefficients. The wavelet coefficients after soft threshold processing are however continuous, and hence the deviation between the processed wavelet coefficients and the original ones will affect the similarity between the reconstructed signal and the original one. In the literature [12,13], an improved wavelet threshold function is presented. When two parameters of the threshold function are adjusted, the processed threshold can be adjusted continuously between soft and hard thresholds. Thus, the performance is altered. Srivastava et al. [14] proposed a new wavelet denoising method for selecting the decomposition levels and noise thresholds. Their threshold function is different from the traditional denoising method. The useful signal can also be easily extracted from the signal with a high magnitude of noise. It has been successful for the sEMG signal denoising method based on wavelet on pre-processing stage of movement recognition of the upper and lower limbs [14,15,16]. Raurale et al. [17] proposed a wrist motion recognition system for embedded platforms to control prostheses and gesture devices. Their research promotes real-time control on embedded devices, but the quality of the signal is not considered in the experiment. Mastinu et al. [18] have come up with a way to control a prosthetic hand, but they can only control the movement of the hand, not the entire arm, which is limited

    The empirical mode decomposition (EMD) method, which has numerous advantages for nonstationary signal analysis [19,20,21], was first introduced by Huang et al. [22], and the EMD method can be used for sEMG noise reduction. However, the EMD method has a disadvantage, that is, the so-called mode mixing effect. Wu and Huang proposed an ensemble empirical mode decomposition (EEMD) algorithm to overcome mode mixing [23,24]. Torres et al. [24] proposed the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) algorithm to improve EEMD. CEEMDAN obtains IMF modes by calculating the unique residue after adding adaptive white Gaussian noise in every stage of EMD. The CEEMDAN algorithm can effectively overcome the mode-mixing problem and reduce the reconstruction error, which has been studied extensively from various fields [25,26,27]. EMD based method may be used alone for sEMG signal denoising [28,29], e.g., EEMD or CEEMDAN decomposition are applied and then certain part of the IMF with high noise is discarded. In such a process some useful signal components may be lost, especially when the signal contains more sharp components like sEMG signal.

    In this study, a novel sEMG signal denoising algorithm is proposed based on the EEMD, an improved wavelet threshold, and the extracted IMF1, which combine the advantages of wavelet and EMD. The methodology is validated through experiments on different sEMG signals, and results show that the random noise in the sEMG signal can be effectively denoised, so as to make prosthetic limb control more accurate, fast and robust.

    The remainder of this paper is organized as follows. In section 2, we present the proposed new method in detail. In section 3, we discuss the experimental results. In section 4, we draw the conclusion.

    The current de-noising algorithm may not completely de-noising or remove some useful signals, which makes it difficult for the control of the prosthesis to meet people's expectations. We propose a new sEMG denoising method based on EEMD and wavelet transform, whose flow diagram is shown in Figure 1. Using the method the noised sEMG signal is first decomposed into several intrinsic mode functions (IMFs), and each IMF is analyzed by the autocorrelation method to determine the high-frequency IMFs containing random noise [30]. Second, the useful signal from IMF1 is extracted by a new processing method based on wavelet transform [13], and the remainder of the IMFs containing random noise is processed using an improved wavelet threshold denoising method [12]. Finally, the processed high-frequency IMFs and low-frequency IMFs are used to reconstruct the denoised sEMG signals.

    Figure 1.  Flow diagram of the new method.

    The steps of the denoising algorithm are as follows:

    (1) Decompose the noised sEMG signal using the EEMD method and obtain n IMFs.

    (2) Calculate the autocorrelation function of each IMF and their normalization results.

    (3) Calculate the variance of each normalized autocorrelation function to measure the noise component of each IMF.

    (4) Find the boundary IMF ck+1(t) through the variance. If the variance is less than a certain threshold, the corresponding IMF is considered to be a high-frequency IMF with a high noise component, whereas other IMFs are considered to be low-frequency IMFs with a low noise component.

    (5) Process c1(t) alone using wavelet extraction to obtain c'1(t). Process the remaining high-frequency IMFs using an improved wavelet threshold denoising method to obtain c'2(t)c'k(t).

    (6) Reconstruct the original signal, s'(t)=c'1(t)+ki=2c'i(t)+nj=k+1cj(t)+r(t), where s'(t) is the denoised signal.

    EMD divides signals into several IMFs and a residual [22], as follows:

    s(t)=ni=1ci(t)+rn(t) (2.1)

    The EEMD method is proposed based on EMD and white noise to solve the modal mixing problem of the EMD. The details are shown in Figure 2, where N is the time to add the noise.

    Figure 2.  Flow diagram of the EEMD.

    The autocorrelation function is used to measure the similarity between signals x(t) and x(t+τ). The normalized autocorrelation function is defined as follows:

    Rx(τ)=E[x(t)x(t+τ)]E[x(t)x(t)] (2.2)

    where Rx(τ) represents the normalized autocorrelation function and E() denotes the mean value.

    The autocorrelation function diagram of random noise is a sharp pulse. However, the diagram of a noised EMG signal has a certain width. The sEMG signal is decomposed into several IMFs by the EEMD. If the IMF contains more random noise, then the middle part of the autocorrelation function diagram narrows. Therefore, in this study, we determine the noise component of each IMF by the autocorrelation function. Then, we obtain the boundary IMF, which determines the IMFs that should be processed. Finally, we conduct further denoising of the IMF with high noise component.

    In traditional signal denoising using the EEMD method, the first IMF is considered the noise. However, it is realized that the first IMF contains certain amount of useful signal [31]. In particular, some peak positions of the signal often have a small component in the first IMF. The sEMG signals often have some abrupt peaks. Therefore, the extraction of useful signals from IMF1 can reduce signal distortion. In this study, a processing method [13] based on wavelet transform is used to extract the useful signal from IMF1. The details are as follows:

    (1) Take the kth level discrete wavelet transform (DWT) of the IMF1, and define the “peak-to-sum ratio” of the detail coefficients, as follows:

    Sj=max(|wj|)Nji=1|wj,i| (2.3)

    where wj is the wavelet coefficient of the level j. If Sj0.2<Sj+1, then j is selected as the decomposition level for the wavelet transform of IMF1. For the parameter settings in this subsection please refer to Ref. [13].

    (2) Select two thresholds λL and λH to deal with the wavelet coefficients of each level, as follows:

    λj,L=μjκj,Lσj (2.4)
    λj,H=μj+κj,Hσj (2.5)

    where λj,L and λj,H are the upper and lower thresholds of the decomposition level j, respectively, μj and σj are the mean and standard deviation of the wavelet coefficients at decomposition level j, respectively, and κj,L, κj,H are the adjustable parameters for each threshold calculated as follows:

    ① If Sj0.01,

    κj,L=max(|wj<0|)μjσj (2.6)
    κj,H=max(|wj>0|)μjσj (2.7)

    ② If 0.01<Sj0.2, we use the peak positive and negative coefficient values in the following manner:

    κj,L=(Sr,LSj,LSr,L)max(|wj<0|)μjσj (2.8)
    κj,H=(Sr,HSj,HSr,H)max(|wj>0|)μjσj (2.9)

    where Sr,L and Sr,H are defined as Sr,LSk,L+Sk+1,L2 and Sr,HSk,H+Sk+1,H2, respectively, which mean the reference peak-to-sum coefficient values. Sj,L and Sj,H are the peak-to-sum ratios of the positive and negative coefficient values, respectively.

    (3) After determining the threshold of each layer, hard thresholding is processed in the following manner:

    ˜wj,i={0     λj,Lwj,iλj,Hwj,i   otherwise (2.10)

    (4) Reconstruct the wavelet coefficients after hard thresholding to obtain the new IMF1, which is the useful signal component extracted from IMF1.

    An improved wavelet threshold denoising method is used to remove the noise from high-frequency IMFs.

    When analyzing the original signal using wavelet transform, selecting the appropriate wavelet function is vital. The maximum entropy method is commonly used. First, the signal is decomposed by wavelet transform. Then, the entropy of the wavelet coefficients is calculated. Finally, the wavelet function which obtain the maximum entropy is selected as the optimal wavelet function. The entropy is calculated in the following manner: First, the wavelet coefficient {wj}(j=1,2,,N) is normalized, as follows:

    pj=|wj|/Nj=1|wj| (2.11)

    Then, the entropy of the wavelet coefficient is calculated as follows:

    En=Ni=1pilog pi (2.12)

    For the EMG signal, through numerous experiments, the use of the sym8 wavelet can often derive the maximum entropy and the best denoising effect. Therefore, we will use the sym8 wavelet as the denoising wavelet basis function in subsequent experiments.

    A decomposition level that results in the best denoising effect exists. By testing the white noise characteristics of the obtained wavelet coefficients sequence, we can determine the decomposition level adaptively [32]. The specific process is as follows:

    (1) Take one level wavelet transform for the pending signal.

    (2) Test the white noise characteristics of the high frequency wavelet coefficients obtained by step (1). If the coefficients belong to white noise, then we continue decompose to obtain the second-level high frequency wavelet coefficients and the white noise characteristics is tested. This process is repeated until the wavelet coefficients of one level do not belong to white noise.

    (3) According to Step (2), if Step (2) decomposes n times, then the selected decomposition level is n − 1.

    A novel improved two-parameter threshold function [12] is used in this study to overcome the disadvantages of the soft and hard threshold functions. By adjusting two parameters of the threshold function, we can change the denoising performance by adjusting the threshold value through different parameters in different circumstances.

    The improved threshold function is defined as follows:

    ˜wj={wj+λλ2β+1                         wj<λsign(wj)α(2β+1)λ2β|dj|2β+1     |wj|λwjλ+λ2β+1                         wj>λ (2.13)

    where α and β are variable parameters and β is an integer [12],

    α={1  β \gt 00  β = 0 (2.14)

    In this experiment, the original sEMG signal was obtained from the standard database, BioPatRec open source platform [33]. The sampling frequency is 2000 Hz, bandwidth at 20–400Hz. The sampling duration is 3–4 s. Part of the signal generated by one motion was selected, and white noise was added to obtain the noised sEMG signal. The experiment was recorded from four bipolar electrodes, distributed on average about a third of the way up the forearm, at a sampling frequency of 2 KHZ and a resolution of 14 bits. The average age of the subjects was 33.9 ± 13 years, and 60% of them were male. Six motion categories (hand opening/closing, wrist flexion and extension, forearm inward and outward rotation) were investigated using four disposable pairs of electrodes, using the same method as offline evaluation. Subjects were visually guided through the biomedical recording user interface. Each movement was repeated three times for three seconds, with the same length of rest between each contraction. The intensity of muscle contraction is required to be around 70–80% of the maximum voluntary contraction, which can be visually verified by the total sEMG strength during muscle contraction.

    In this section, first, the sEMG signal produced by the flexor carpi radialis when bending the elbow is selected as the research object. Then, 10 dB Gaussian white noise was added to the sEMG signal to obtain a noised sEMG signal.

    The noised sEMG signal is decomposed into several IMFs and one residue by the EEMD method to reduce the noise. Figure 3 shows the noised sEMG signal and the first five IMFs.

    Figure 3.  First five IMFs of the sEMG signal after being decomposed by the EEMD: (a) noised sEMG signal, (b) IMF1, (c) IMF2, (d) IMF3, (e) IMF4, and (f) IMF5.

    The normalized autocorrelation function of each IMF and their variance are calculated. Figure 4 shows the normalized autocorrelation function of the first five IMFs. Table 1 shows the variance of each normalized autocorrelation function. According to a large number of experimental results, the threshold of the variance is set to 0.005. The variance of the first three IMFs’ normalized autocorrelation function are less than 0.005, whereas that of the others are greater than 0.005. From the properties of the autocorrelation function and the variance threshold method, we need a de-noising algorithm to make the original signal undergo EEMD decomposition, so that the variance of the first three IMF components is also less than 0.05. Because of the high noise component of the high frequency IMF will cause a large deviation in the control of the prosthesis.

    Figure 4.  Normalized autocorrelation function of the noisy sEMG signal and the first five IMFs: (a) noised sEMG signal, (b) IMF1, (c) IMF2, (d) IMF3, (e) IMF4, and (f) IMF5.
    Table 1.  Variance of each normalized autocorrelation function.
    IMF IMF1 IMF2 IMF3 IMF4 IMF5 IMF6 IMF7 IMF8 IMF9 IMF10
    Variance 0.0006 0.0016 0.0038 0.0145 0.0094 0.0264 0.0359 0.0453 0.0504 0.1694

     | Show Table
    DownLoad: CSV

    During signal denoising by the traditional EMD or EEMD method, the first IMF is usually regarded as noise and discarded. Indeed, the noise will be eliminated. However, in this manner, the useful signal component in IMF1 is also discarded. After reconstructing the signal, many parts are distorted. If some useful components are treated as noise removal, the details of some people's movements may not be collected, which is one of the reasons why the current de-noising algorithm is not effective in practical application.

    The first IMF is shown in Figure 3(b). Clearly, most of IMF1 are random noise, but some are still useful signals. Given that the sEMG signals often have some peaks, some useful components from the mutated peaks exist in IMF1.

    A wavelet extraction is used to improve the denoising effect of the sEMG signal, and the useful signal component extracted from IMF1 is shown in Figure 5.

    Figure 5.  Extraction of the useful signal component from IMF1: (a) IMF1 and (b) extracted component.

    Figure 6 shows the denoising effect with IMF1 discarded and IMF1 processed. The denoised sEMG signal and original sEMG signal are part of the complete signal from sampling points 1,000 to 1,400. As shown in Figure 6(a), the denoised sEMG signal obtained using the traditional EEMD method is compared with the original sEMG signal, and some peak parts of the sEMG signal are lost.

    Figure 6.  (a) Traditional denoising effect with the IMF1 discarded and (b) improved denoising effect with the IMF1 processed.

    On the contrary, Figure 6(b) illustrates that the peak parts of the sEMG signal are not lost when the useful signal component extracted from the IMF1 is added before reconstructing the signal. This allows us to eliminate unwanted noise while preserving the useful details of human motion signals, thus greatly improving the accuracy and robustness of prosthetic control.

    The remainder of the high-frequency IMF, IMF2, and IMF3 are processed by the improved wavelet threshold denoising method. Figure 7 shows the original IMF2 and IMF3 and the processed IMF2 and IMF3.

    Figure 7.  (a) Original IMF2, (b) processed IMF2, (c) original IMF3, and (d) processed IMF3.

    We reconstruct the original signal. Figure 8 shows the original sEMG signal, noised sEMG signal, and denoised sEMG signal.

    Figure 8.  (a) Original sEMG signal, (b) noised sEMG signal, and (c) denoised sEMG signal.

    Figures 9(a) to 9(h) show the denoising results of the noised sEMG signal using different denoising methods. EEMDWT means the proposed new method based on EEMD and wavelet transform. EMDWT means the method based on EMD and wavelet transform. CEEMDAN & WT means the method based on CEEMDAN and wavelet transform.

    Figure 9.  Denoising results of part of the noised sEMG signal using different denoising methods: (a) part of original sEMG signal, (b) hard threshold, (c) soft threshold, (d) improved threshold, (e) EMD, (f) EMDWT, (g) EEMD, (h) EEMDWT, (i) CEEMDAN, (j) CEEMDAN & WT.

    First, 5, 10, 15, and 20 dB Gaussian white noise were added to the noise-free sEMG signal to illustrate the advantages of the proposed algorithm quantitatively. Then, the traditional method and the proposed improved method are used to denoise the signal. Signal-to-noise ratio (SNR), Peak Signal-to-Noise Ratio (PSNR) and root mean square (RMSE) will be used as the indices of the denoising effect. SNR defines the signal energy with respect to the energy of the error. PSNR is an engineering term for the ratio between the maximum possible power of a signal and the power of corrupting noise that affects the fidelity of its representation. RMSE defines the energy of the error signal during denoising. SNR, PSNR and RMSE are defined as follows:

    SNR = 10lgNi=1x2iNi=1(xiˆxi)2 (3.1)
    RMSE=1NNi=1(xiˆxi)2 (3.2)
    PSNR=10log10(MAX2xiMSE)=20log10(MAXxiRMSE) (3.3)

    where xi is the original sEMG signal, ˆxi is the denoised sEMG signal, N is the signal length, MSE=RMSE2 and MAXxi is the maximum of xi.

    Table 2 presents the performance indices computed under different noise intensities by various denoising methods. The SNR, PSNR and RMSE values are the average value of ten experiments.

    Table 2.  Performance indices using different methods for the semg signal produced by the flexor carpi radialis when bending the elbow (SNR: dB, PSNR: dB, RMSE: 10–5).
    Noised signal SNR5.000 PSNR24.62 RMSE2.762 SNR10.00 PSNR29.54 RMSE1.567 SNR15.00 PSNR34.42 RMSE0.894 SNR20.00 PSNR39.49 RMSE0.498
    Hard threshold 10.06 28.87 2.065 12.48 31.91 1.563 15.69 34.36 1.080 17.85 36.37 0.842
    Soft threshold 9.182 28.39 2.284 11.76 31.25 1.697 13.70 33.60 1.358 15.65 35.53 1.085
    Improved threshold 10.32 29.21 2.003 14.85 32.62 1.190 18.44 35.47 0.786 21.91 37.74 0.528
    EMD 7.539 26.99 2.759 10.17 31.28 2.037 8.29 28.89 2.531 8.843 27.79 2.375
    EMDWT 10.21 28.50 2.029 14.23 33.92 1.277 17.12 36.83 0.916 19.14 39.97 0.726
    EEMD 7.866 28.17 2.658 9.009 30.20 2.330 9.25 27.52 2.267 8.837 27.12 2.377
    EEMDWT 10.79 30.15 1.898 14.95 34.07 1.176 18.86 38.59 0.750 23.58 42.76 0.435
    CEEMDAN 8.667 28.05 2.544 12.79 32.18 1.547 12.32 31.82 1.411 11.04 30.19 1.403
    CEEMDAN & WT 10.66 29.30 1.932 14.79 33.87 1.170 18.64 37.93 0.859 23.47 42.19 0.477

     | Show Table
    DownLoad: CSV

    Table 2 indicates that the improved threshold methods had higher SNR, PSNR and lower RMSE than the traditional threshold methods. In all methods, the proposed new method EEMDWT had the highest SNR, PSNR and lowest RMSE, which illustrates the superiority of the proposed new method over the other methods to denoise the sEMG signal.

    In the experiment, we also selected the sEMG signal produced by the flexor carpi radialis when bending the wrist to verify the superiority of the proposed new method for the sEMG signal produced by different motions. The selected sEMG signal was processed. Table 3 shows the performance indices computed under different noise intensities by various denoising methods.

    Table 3.  Performance indices using different methods for the semg signal produced by the flexor carpi radialis when bending the wrist (SNR: dB, PSNR: dB, RMSE: 10–5).
    Noised signal SNR5.000 PSNR19.64 RMSE2.634 SNR10.00 PSNR24.52 RMSE1.502 SNR15.00 PSNR29.36 RMSE0.860 SNR20.00 PSNR34.70 RMSE0.465
    Hard threshold 11.71 26.04 1.244 14.77 28.86 0.874 16.27 30.72 0.736 18.71 32.63 0.556
    Soft threshold 11.69 26.14 1.246 14.63 29.08 0.889 15.97 30.41 0.762 17.79 32.23 0.618
    Improved threshold 11.92 26.37 1.214 16.16 29.60 0.746 17.59 31.12 0.632 23.37 33.16 0.325
    EMD 10.09 22.94 1.500 13.63 28.07 0.998 18.21 32.49 0.589 23.84 38.29 0.308
    EMDWT 11.37 25.37 1.293 15.02 30.43 0.849 18.64 33.74 0.560 24.66 38.65 0.280
    EEMD 11.42 23.74 1.287 14.52 28.93 0.900 19.09 33.51 0.532 24.80 39.33 0.276
    EEMDWT 12.40 26.77 1.149 17.02 31.77 0.675 20.71 35.11 0.442 26.21 40.17 0.234
    CEEMDAN 9.171 23.56 1.666 14.41 28.81 0.911 18.89 33.27 0.544 24.22 38.74 0.295
    CEEMDAN & WT 11.99 24.38 1.212 16.50 29.41 0.703 20.33 33.65 0.477 25.15 39.67 0.250

     | Show Table
    DownLoad: CSV

    As shown in Table 3, the proposed new method has the highest SNR, PSNR and lowest RMSE. The denoising effect of the traditional hard and soft wavelet threshold methods is poor. Tables 2 and 3 show the denoising effect of the sEMG signal produced by two motions on the same muscle. We observe that the denoising effect of traditional EMD, EEMD and CEEMDAN are unstable for different sEMG signals. As for CEEMDAN & WT, the performance may sometimes be superb and sometimes poor, leading to a lower average value than EEMDWT. This instability of CEEMDAN & WT makes us choose EEMD in the proposed method.

    In the experiment, the sEMG signals produced by the soleus when rotating and lifting the foot were selected to verify the superiority of the proposed new method for sEMG signals produced by different muscles. Tables 4 and 5 show the results. The proposed new method always had a significant denoising effect compared with the other methods.

    Table 4.  Performance indices using different methods for the semg signal produced by the soleus when rotating the foot (SNR: dB, PSNR: dB, RMSE: 10–5).
    Noised signal SNR5.000 PSNR15.85 RMSE0.854 SNR10.00 PSNR20.91 RMSE0.477 SNR15.00 PSNR26.03 RMSE0.265 SNR20.00 PSNR30.75 RMSE0.154
    Hard threshold 11.91 22.76 0.385 14.82 25.37 0.276 16.61 27.23 0.224 20.76 29.06 0.139
    Soft threshold 11.96 22.81 0.383 14.83 25.69 0.275 16.44 27.29 0.229 19.26 28.98 0.165
    Improved threshold 11.97 22.82 0.383 15.19 26.04 0.264 17.16 28.01 0.211 23.10 29.78 0.106
    EMD 8.124 18.95 0.596 13.49 24.34 0.321 18.32 29.17 0.184 23.76 34.12 0.099
    EMDWT 11.01 21.50 0.427 15.01 26.58 0.270 19.84 31.13 0.155 24.22 34.70 0.093
    EEMD 8.73 19.60 0.555 14.44 25.37 0.288 19.39 30.31 0.163 24.44 35.10 0.091
    EEMDWT 12.11 22.73 0.377 17.13 28.05 0.211 21.29 32.17 0.131 25.74 36.03 0.078
    CEEMDAN 8.656 19.48 0.560 13.97 25.15 0.304 19.11 29.79 0.168 23.89 34.70 0.097
    CEEMDAN & WT 12.01 20.23 0.379 16.79 25.68 0.220 20.77 30.09 0.144 24.95 35.60 0.088

     | Show Table
    DownLoad: CSV
    Table 5.  Performance indices using different methods for the semg signal produced by the soleus when lifting the foot (SNR: dB, PSNR: dB, RMSE: 10–5).
    Noised signal SNR5.000 PSNR14.46 RMSE0.422 SNR10.00 PSNR19.20 RMSE0.244 SNR15.00 PSNR24.50 RMSE0.133 SNR20.00 PSNR29.03 RMSE0.079
    Hard threshold 11.30 20.70 0.206 12.87 25.04 0.172 18.27 30.36 0.092 23.34 35.58 0.051
    Soft threshold 11.30 20.69 0.206 12.87 25.03 0.172 18.27 30.36 0.092 23.34 35.73 0.051
    Improved threshold 12.70 20.59 0.175 13.76 24.95 0.155 20.89 30.28 0.068 26.58 35.63 0.035
    EMD 8.389 17.78 0.287 13.34 22.73 0.163 18.37 27.76 0.091 24.00 32.74 0.048
    EMDWT 10.48 22.19 0.226 14.90 26.56 0.136 20.49 31.87 0.071 25.59 37.20 0.040
    EEMD 9.359 18.73 0.257 14.15 23.52 0.148 19.59 28.90 0.079 24.73 33.66 0.044
    EEMDWT 13.87 23.23 0.153 16.71 27.58 0.110 22.14 32.93 0.059 27.72 37.66 0.031
    CEEMDAN 9.24 18.34 0.261 14.11 23.44 0.149 19.41 28.69 0.081 23.75 33.17 0.049
    CEEMDAN & WT 13.76 22.22 0.155 15.78 27.36 0.127 20.92 32.39 0.070 27.02 36.73 0.034

     | Show Table
    DownLoad: CSV

    In summary, this study aimed to analyze the method of denoising the sEMG signal and propose a new effective method to improve the robustness and accuracy of prosthetic control. As shown in Tables 2 to 5, the improved threshold method always has a better effect than the traditional hard and soft threshold denoising methods. Therefore, in the proposed new method, the improved threshold method is used to remove the noise in the high-frequency IMFs. In the traditional EMD based methods, the denoising effect is unstable. The effect may sometimes be good, but it may sometimes be bad, which depends on the sEMG signal. In three combination methods, EMDWT, EEMDWT and CEEMDAN & WT, the IMFs are processed in the same manner, but the decomposition method is different. The combined method EMDWT based on EMD has a good denoising effect. However, the denoising effect is not good enough because of the drawback of mode mixing. The combined method CEEMDAN & WT based on CEEMDAN also has a good denoising effect. However, the denoising effect is not good enough because of its instability. Therefore, the proposed method selects the EEMD method as the decomposition method, which can overcome mode mixing and have a stable denoising effect. As shown in each table, the proposed methods all have the best denoising effect. The results illustrate that the proposed algorithm can improve the robustness of emg to changing noises. In addition, in the case of low or no noise, there is no defect of information loss, which plays a significant role in promoting the field of prosthetic limb control and medical rehabilitation.

    Prosthetic limb control is currently the focus of rehabilitation in the field of medical rehabilitation and the signal to control the prosthesis must be of high quality. Therefore, how to obtain pure EMG signals is an important issue. Since its birth, EMD has been widely used in various fields of signal processing and analysis. However, the modal decomposition method has some shortcomings in practical applications, such as modal aliasing, which greatly hinders its further application in practice. There are two reasons for modal aliasing: on the one hand, it is related to the EMD algorithm itself; on the other hand, the frequency characteristics of the original signal will also affect the decomposition result, and even produce modal aliasing. For example, white noise residuals and false modes after EEMD decomposition, and as the EEMD method becomes more and more widely used, the defects of the EEMD method gradually become prominent. However, EEMD cannot effectively solve the modal aliasing phenomenon, and effective IMF requires manual screening. Other scholars have proposed improved EEMD methods based on the EEMD method, such as SVM-based EEMD improvement methods, complementary generalized empirical mode decomposition (CEEMD) and other auxiliary methods in noise, and complementary set overall empirical mode decomposition (CEEMD) The generalized empirical mode decomposition (ceemdan) of the complete auxiliary noise set restores the integrity of the emd decomposition, but these methods require many iterations, long calculation time, and low decomposition efficiency. In addition to EEMD, wavelet transform plays an important role in non-stationary signal processing. The most commonly used denoising method is wavelet threshold method, including hard threshold method and soft threshold method. The wavelet coefficients become discontinuous after hard thresholding. Therefore, when reconstructing the original signal with the processed wavelet coefficients, oscillations may occur in some places. Although the wavelet coefficients obtained after soft threshold processing are continuous, the deviation between the processed wavelet coefficients and the original coefficients will affect the degree of approximation between the reconstructed signal and the original signal.

    In order to solve the problems of existing denoising algorithms, this paper proposes a new sEMG signal denoising algorithm based on EEMD. This algorithm is an improved wavelet threshold algorithm and extracts useful signal components from IMF1. The noisy surface EMG signal is decomposed into a series of internal mode functions. Through autocorrelation analysis, the high frequency internal mode function containing random noise is determined. A new method is used to process the first IMF with the most random noise to extract useful signal components. For other internal model functions containing random noise, an improved wavelet threshold is used to denoise. Finally, reconstruct the remaining internal mode functions to obtain clean surface EMG signals. The surface EMG signals generated by the flexor carpi radialis and soleus muscles during elbow flexion, wrist rotation, and foot lift were verified experimentally. As mentioned earlier, we use SNR, PSNR and RMSE as quality estimators for denoising. The results show that the method has high signal-to-noise ratio, peak signal-to-noise ratio and the lowest root mean square error, and can stably remove random noise in surface EMG signals, and the effect is significant. This work lays the foundation for the subsequent study of EMG signal feature extraction and action recognition, and ensures its usability in prosthetic control.

    In this study, we found that CEEMDAN & WT can sometimes achieve good results, but its performance lacks stability. Making this method feasible will be one of our future research directions. If this method can achieve a stable denoising effect, it will greatly promote the process of prosthesis control and medical rehabilitation. At present, the algorithm has a good denoising effect under the interference of white noise, but no experiments have been carried out on motion artifact noise and electromagnetic interference noise. In future work, we will try to extract these noise signals and further verify the effect of our algorithm.

    This work was supported by the National Natural Science Foundation of China (Nos. 61971169, and 60903084), and Jinhua Science and Technology Bureau (Nos. 2019–3–020 and 2020–3–036).

    No potential conflict of interest was reported by the authors.



    [1] Carlo J. De Luca, Physiology and mathematics of myoelectric signals, IEEE. Trans. Biomed. Eng., 6 (1979), 313-325.
    [2] J. Maier, A. Naber, M. Ortiz-Catalan, Improved prosthetic control based on myoelectric pattern recognition via wavelet-based de-noising, IEEE Trans. Neur. Syst. Reh. Eng., 26 (2017), 506-514.
    [3] L. Liu, X. Chen, Z. Lu, S. Cao, D. Wu, X. Zhang, Development of an EMG-ACC-based upper limb rehabilitation training system, IEEE Trans. Neur. Syst. Reh. Eng., 25 (2016), 244-253.
    [4] L. L. Chuang, Y. L. Chen, C. C. Chen, Y. C. Li, A. M. Wong, A. L. Hsu, et al., Effect of EMG-triggered neuromuscular electrical stimulation with bilateral arm training on hemiplegic shoulder pain and arm function after stroke: a randomized controlled trial, J. Neuroeng. Rehabil., 14 (2017), 122. doi: 10.1186/s12984-017-0332-0
    [5] K. Veer, Development of sensor system with measurement of surface electromyogram signal for clinical use, Optik, 127 (2016), 352-356. doi: 10.1016/j.ijleo.2015.10.072
    [6] C. Castellini, A. E. Fiorilla, G. Sandini, Multi-subject/daily-life activity EMG-based control of mechanical hands, J. Neuroeng. Rehabil., 6 (2009), 1-11. doi: 10.1186/1743-0003-6-1
    [7] R. E. Johnson, K. P. Kording, L. J. Hargrove, J. W. Sensinger, EMG versus torque control of human-machine systems: Equalizing control signal variability does not equalize error or uncertainty, IEEE Trans. Neur. Syst. Reh. Eng., 25 (2016), 660-667.
    [8] F. Zhang, P. Li, Z. G. Hou, Z. Lu, Y. Chen, Q. Li, et al., sEMG-based continuous estimation of joint angles of human legs by using BP neural network, Neurocomputing, 78 (2012), 139-148. doi: 10.1016/j.neucom.2011.05.033
    [9] G. A. Lovell, P. D. Blanch, C. J. Barnes, EMG of the hip adductor muscles in six clinical examination tests, Phys. Ther. Sport., 13 (2012), 134-140. doi: 10.1016/j.ptsp.2011.08.004
    [10] A. Phinyomark, P. Phukpattaranont, C. Limsakul, Fractal analysis features for weak and single-channel upper-limb EMG signals, Expert. Syst. Appl., 39 (2012), 11156-11163. doi: 10.1016/j.eswa.2012.03.039
    [11] S. Mallat, A wavelet tour of signal processing, Academic Press, 1999.
    [12] D. L. Donoho, De-noising by soft-thresholding, IEEE Trans. Inf. Theory, 41 (1995), 613-627. doi: 10.1109/18.382009
    [13] H. Liu, W. Wang, C. Xiang, L. Han, H. Nie, A de-noising method using the improved wavelet threshold function based on noise variance estimation, Mech. Syst. Signal Process., 99 (2018), 30-46. doi: 10.1016/j.ymssp.2017.05.034
    [14] M. Srivastava, C. L. Anderson, J. H. Freed, A new wavelet denoising method for selecting decomposition levels and noise thresholds, IEEE Access, 4 (2016), 3862-3877. doi: 10.1109/ACCESS.2016.2587581
    [15] M. S. Hussain, M. B. I. Reaz, F. Mohd‐Yasin, M. I. Ibrahimy, Electromyography signal analysis using wavelet transform and higher order statistics to determine muscle contraction, Expert Syst., 26 (2009), 35-48. doi: 10.1111/j.1468-0394.2008.00483.x
    [16] M. Khezri, M. Jahed, Surface electromyogram signal estimation based on wavelet thresholding technique, In 30th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, (2008), 4752-4755.
    [17] S. Raurale, J. McAllister, J. M. del Rincon, Emg wrist-hand motion recognition system for real-time embedded platform, In ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, (2019), 1523-1527.
    [18] E. Mastinu, F. Clemente, P. Sassu, O. Aszmann, R. Brånemark, B. Håkansson, et al., Grip control and motor coordination with implanted and surface electrodes while grasping with an osseointegrated prosthetic hand, J. Neuroeng. Rehabilitation, 16 (2019), 1-10. doi: 10.1186/s12984-018-0454-z
    [19] G. Wei, F. Tian, G. Tang, C. Wang, A wavelet-based method to predict muscle forces from surface electromyography signals in weightlifting, J. Bionic. Eng., 9 (2012), 48-58. doi: 10.1016/S1672-6529(11)60096-6
    [20] J. Xu, Z. Wang, C. Tan, L. Si, X. Liu, A novel denoising method for an acoustic-based system through empirical mode decomposition and an improved fruit fly optimization algorithm, Appl. Sci., 7 (2017), 215. doi: 10.3390/app7030215
    [21] Y. Lv, R. Yuan, G. Song, Multivariate empirical mode decomposition and its application to fault diagnosis of rolling bearing, Mech. Syst. Signal Process., 81 (2016). 219-234.
    [22] X. Zhao, M. Li, G. Song, J. Xu, Hierarchical ensemble-based data fusion for structural health monitoring, Smart Mater. Struct., 19 (2010), 045009. doi: 10.1088/0964-1726/19/4/045009
    [23] N. E. Huang, Z. Shen, S. R. Long, M. C. Wu, H. H. Shih, Q. Zheng, et al., The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis, Proc. Math. Phys. Eng. Sci., 454 (1998), 903-995. doi: 10.1098/rspa.1998.0193
    [24] Z. Wu, N. E. Huang, Ensemble empirical mode decomposition: a noise-assisted data analysis method, Adv. Adap. Data. Anal., 1 (2009), 1-41. doi: 10.1142/S1793536909000047
    [25] M. E. Torres, M. A. Colominas, G. Schlotthauer, P. Flandrin, A complete ensemble empirical mode decomposition with adaptive noise, In 2011 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, (2011), 4144-4147.
    [26] B. He, Y. P. Bai, MEMS hydrophone signal denoising based on wavelet packet and CEEMDAN, Math. Pract. Theory, 46 (2016), 139-147.
    [27] Y. Xu, M. Luo, T. Li, G. Song, ECG signal de-noising and baseline wander correction based on CEEMDAN and wavelet threshold, Sensors, 17 (2017), 2754. doi: 10.3390/s17122754
    [28] Y. Li, Y. Li, X. Chen, J. Yu, H. Yang, L. Wang, A new underwater acoustic signal denoising technique based on CEEMDAN, mutual information, permutation entropy, and wavelet threshold denoising, Entropy, 20 (2018), 563. doi: 10.3390/e20080563
    [29] A. O. Andrade, S. Nasuto, P. Kyberd, C. M. Sweeney-Reed, F. R. Van Kanijn, EMG signal filtering based on empirical mode decomposition, Biomed. Signal Process Control, 1 (2006), 44-55.
    [30] X. Zhang, P. Zhou, Filtering of surface EMG using ensemble empirical mode decomposition, Med. Eng. Phys., 35 (2013), 537-542. doi: 10.1016/j.medengphy.2012.10.009
    [31] W. Jiao, Z. Li, D. Wang, A method for wavelet threshold denoising of seismic data based on CEEMD, Geophys. Prospect. Pet., 53 (2014), 164-172.
    [32] J. X. Zhang, Q. H. Zhong, Y. P. Dai, The determination of the threshold and the decomposition order in threshold de-noising method based on wavelet transform, In Proceedings of the CSEE, 24 (2004), 118-122.
    [33] W. B. Wang, X. D. Zhang, X. L. Wang, Empirical mode decomposition de-noising method based on principal component analysis, Acta. Elec. Sin., 41 (2013), 1425-1430.
    [34] M. Ortiz-Catalan, R. Brånemark, B. Håkansson, BioPatRec: A modular research platform for the control of artificial limbs based on pattern recognition algorithms, Source Code Biol. Med., 8 (2013), 11.
  • This article has been cited by:

    1. Songsong Li, Ying Yang, Chen Li, Huimin He, Qi Zhang, Siqi Zhao, Research on Signal Processing Technology of Ultrasonic Non‐Destructive Testing Based on EEMD Combined with Wavelet Packet , 2023, 1931-4973, 10.1002/tee.23770
    2. William Bekerman, Madhur Srivastava, Determining Decomposition Levels for Wavelet Denoising Using Sparsity Plot, 2021, 9, 2169-3536, 110582, 10.1109/ACCESS.2021.3103497
    3. Jianguo Zhang, Man Wang, Bing Jia, Yujie Wang, Study on Optimization of Infrasound Filtering Method for Coal Sample Failure under Load, 2022, 10, 2296-6463, 10.3389/feart.2022.834234
    4. Zhangfeng Zhao, Gaohong Liu, Yueliang Wang, Jiyu Peng, Xin Qiao, Jiang Zhong, Research on Dynamic Measurement Method of Flow Rate in Tea Processing, 2022, 22, 1424-8220, 4294, 10.3390/s22114294
    5. Xiaoqian Tang, Jian Ying, Qiang Zhu, Hangjun Chen, Guoyin Yang, Like Jiang, Hao Cai, Mengting Huang, 2022, Research of noise reduction about sEMG signal for live working based on MEEMD and DFA, 978-1-6654-7732-1, 1, 10.1109/CIYCEE55749.2022.9959026
    6. Marianne Boyer, Laurent Bouyer, Jean-Sébastien Roy, Alexandre Campeau-Lecours, Reducing Noise, Artifacts and Interference in Single-Channel EMG Signals: A Review, 2023, 23, 1424-8220, 2927, 10.3390/s23062927
    7. H. Ashraf, U. Shafiq, Q. Sajjad, A. Waris, O. Gilani, M. Boutaayamou, O. Brüls, Variational mode decomposition for surface and intramuscular EMG signal denoising, 2023, 82, 17468094, 104560, 10.1016/j.bspc.2022.104560
    8. Jing Li, Junhong Wang, Ting Wang, Wanzeng Kong, Xugang Xi, Quantification of body ownership awareness induced by the visual movement illusion of the lower limbs: a study of electroencephalogram and surface electromyography, 2023, 0140-0118, 10.1007/s11517-022-02744-4
    9. Xugang Xi, Jinsuo Ding, Junhong Wang, Yun-Bo Zhao, Ting Wang, Wanzeng Kong, Jingqi Li, Analysis of Functional Corticomuscular Coupling Based on Multiscale Transfer Spectral Entropy, 2022, 26, 2168-2194, 5085, 10.1109/JBHI.2022.3193984
    10. Jinfeng Zhao, Jianxin Liu, Danilo Pelusi, The Diagnostic Value of Scanning in the Injury of Triceps Crus of Volleyball Players, 2022, 2022, 1932-8745, 1, 10.1155/2022/2203065
    11. Jin Su Kim, Min-Gu Kim, Jae Myung Kim, Sung Bum Pan, Multi-Stream CNN-Based Personal Recognition Method Using Surface Electromyogram for 5G Security, 2022, 72, 1546-2226, 2997, 10.32604/cmc.2022.026572
    12. Lu Sun, Xiuling Shan, Qihu Dong, Chong Wu, Mei Shan, Hongxia Guo, Rui Lu, Osamah Ibrahim Khalaf, Ultrasonic Elastography Combined with Human Papilloma Virus Detection Based on Intelligent Denoising Algorithm in Diagnosis of Cervical Intraepithelial Neoplasia, 2021, 2021, 1748-6718, 1, 10.1155/2021/8066133
    13. Noureddine Messaoudi, Samia Belkacem, Raïs El'hadi Bekka, Ability of spatial filters to distinguish between two MUAPs generated from MUs with different locations, sizes and fibers pennation, 2023, 18, 1748-0221, P03041, 10.1088/1748-0221/18/03/P03041
    14. Yuexing Gu, Tongyou Liu, Yuanjing Xu, Yuling Shen, Hang Ren, Jinwu Wang, 2022, A Sample Data Augmentation Method for EMG Gesture Recognition, 979-8-3503-9916-5, 442, 10.1109/IC2ECS57645.2022.10087946
    15. Xian Hua, Jing Li, Ting Wang, Junhong Wang, Shaojun Pi, Hangcheng Li, Xugang Xi, Evaluation of movement functional rehabilitation after stroke: A study via graph theory and corticomuscular coupling as potential biomarker, 2023, 20, 1551-0018, 10530, 10.3934/mbe.2023465
    16. Pratap Kumar Koppolu, Krishnan Chemmangat, Automatic selection of IMFs to denoise the sEMG signals using EMD, 2023, 73, 10506411, 102834, 10.1016/j.jelekin.2023.102834
    17. Daniele Esposito, Jessica Centracchio, Paolo Bifulco, Emilio Andreozzi, A smart approach to EMG envelope extraction and powerful denoising for human–machine interfaces, 2023, 13, 2045-2322, 10.1038/s41598-023-33319-4
    18. Youbin Huang, Jian Mao, Kai Huang, Jinming Liu, Ni Tang, Electromagnetic signal denoising model based on stacked ET layers structure, 2025, 242, 02632241, 115796, 10.1016/j.measurement.2024.115796
    19. Yuyan Zhang, Zhixia Yang, Xiaoli Du, Xiaoyuan Luo, A New Method for Denoising Underwater Acoustic Signals Based on EEMD, Correlation Coefficient, Permutation Entropy, and Wavelet Threshold Denoising, 2024, 23, 1671-9433, 222, 10.1007/s11804-024-00386-6
    20. Ali Zakaria Messaoui, Laurent Peyrodie, Mohamed Amine Alouane, François Gionco, Mohamed Guiatni, 2024, Deep Learning-Based Efficient Human Joint Movement Prediction using Surface Electromyography Data, 979-8-3503-6102-5, 1, 10.1109/ICCAD60883.2024.10553805
    21. Guo Hua-Ling, Zhenh Bin, Liu Li-Ping, Liu Hui, Study on Denoising Method of Surface Defect Signal of Rail Based on CEEMD and Wavelet Soft Threshold, 2023, 69, 1063-7710, 929, 10.1134/S1063771022600504
    22. Junpo Liu, Alexis Achurra, Chuang Zhang, Alan Bury, Xinjian Wang, A long short term memory network-based, global navigation satellite system/inertial navigation system for unmanned surface vessels, 2024, 23, 2046-4177, 316, 10.1080/20464177.2024.2334029
    23. Chen Zhang, Zijian Zhou, Shuren Zhou, 2023, sEMG Signal Denoising Based on Variational Mode Decomposition and Wavelet Thresholding, 979-8-3503-3603-0, 913, 10.1109/ICMSP58539.2023.10171095
    24. Kuangxi Su, Chengli Zheng, Xing Yu, Portfolio allocation with CEEMDAN denoising algorithm, 2023, 27, 1432-7643, 15955, 10.1007/s00500-023-08883-6
    25. Yubo Sun, Zhenhui Guo, Yuxin Feng, Ningbo Yu, 2023, A Novel Surface Electromyography Signals Denoising Method for Gait Phase Classification, 978-988-75815-4-3, 3465, 10.23919/CCC58697.2023.10241069
    26. Xudong Chen, Hongdi Guo, Shaowei Hu, Chongshi Gu, Na Lu, Jinjun Guo, Xing Liu, Tzu-Kang Lin, Dynamic Cluster Zoning of Arch Dam Deformation Considering Changing Working Conditions, 2024, 2024, 1545-2255, 10.1155/2024/8813251
    27. Han Zhao, Zhenwei Wang, Yu Yang, Ziyu Zhang, 2024, FastICA-based noise reduction algorithm for surface electromyographic signals, 979-8-3503-9437-5, 1290, 10.1109/NNICE61279.2024.10498893
    28. Xiaoxiang Gao, Xiangjun Chen, Muyang Lin, Wentong Yue, Hongjie Hu, Siyu Qin, Fangao Zhang, Zhiyuan Lou, Lu Yin, Hao Huang, Sai Zhou, Yizhou Bian, Xinyi Yang, Yangzhi Zhu, Jing Mu, Xinyu Wang, Geonho Park, Chengchangfeng Lu, Ruotao Wang, Ray S. Wu, Joseph Wang, Jinghong Li, Sheng Xu, A wearable echomyography system based on a single transducer, 2024, 2520-1131, 10.1038/s41928-024-01271-4
    29. Xin Li, Jingran Bu, Zhen Yang, Hao Li, Hui Zuo, Yuning Wang, Jing Zhou, A Denoising Method for Loaded Coal-Rock Charge Signals Based on a Joint Algorithm of IWT and ICEEMDAN, 2024, 12, 2169-3536, 85178, 10.1109/ACCESS.2024.3414590
    30. Zemin Zhou, Yanrui Qu, Boqing Zhu, Bingbing Zhang, Detection of Typical Transient Signals in Water by XGBoost Classifier Based on Shape Statistical Features: Application to the Call of Southern Right Whale, 2024, 12, 2077-1312, 1596, 10.3390/jmse12091596
    31. Qi Liu, Zhifan Zhao, Huaishu Hou, Jinhao Li, Shuaijun Xia, Research on signal denoising algorithm based on ICEEMDAN eddy current detection, 2024, 19, 1748-0221, P09026, 10.1088/1748-0221/19/09/P09026
    32. Liu Cui, Zhisen Si, Kai Zhao, Shuangkui Wang, Denoising method for colonic pressure signals based on improved wavelet threshold, 2024, 10, 2057-1976, 065047, 10.1088/2057-1976/ad81fc
    33. Jianpeng Tang, Xugang Xi, Ting Wang, Junhong Wang, Lihua Li, Zhong Lü, Analysis of corticomuscular-cortical functional network based on time-delayed maximal information spectral coefficient, 2023, 20, 1741-2560, 056017, 10.1088/1741-2552/acf7f7
    34. Huilin Chen, Li Zhang, Danyang Li, Jingao Xu, Weiqi Yang, Zheng Yang, REFLoc: A Resilient Evolutionary Fusion Framework for Robust Indoor Localization, 2024, 73, 0018-9456, 1, 10.1109/TIM.2024.3403196
    35. Peng Pang, Xiwen Wei, Xinnian Yang, Real-time Monitoring Algorithm of Muscle State Based on sEMG Signal, 2023, 2560, 1742-6588, 012003, 10.1088/1742-6596/2560/1/012003
    36. Jantine J Wisse, Peter Somhorst, Joris Behr, Arthur R van Nieuw Amerongen, Diederik Gommers, Annemijn H Jonkman, Improved filtering methods to suppress cardiovascular contamination in electrical impedance tomography recordings, 2024, 45, 0967-3334, 055010, 10.1088/1361-6579/ad46e3
    37. Jianpeng Tang, Xugang Xi, Ting Wang, Junhong Wang, Lihua Li, Zhong Lü, Influence of neuromuscular electrical stimulation pulse waveform on corticomuscular coupling and the brain functional connectivity network, 2024, 88, 17468094, 105627, 10.1016/j.bspc.2023.105627
    38. Yongkun Zhao, Shibo Jing, Haijun Wu, Honghan Li, Masahiro Todoh, E-TRGAN: A Novel Transformer Generative Adversarial Network for High-Density Surface Electromyography Signal Reconstruction, 2024, 73, 0018-9456, 1, 10.1109/TIM.2024.3472778
    39. Xin Huang, Weiwei Qian, Peng Zhang, Zhongtian Ding, Shunming Li, A hybrid transformer masked time-domain denoising network for vibration signals, 2025, 36, 0957-0233, 016193, 10.1088/1361-6501/ad99f2
    40. Pawel Trajdos, A noise-tolerant dual multi-classifier system with fuzzy model applied to the sEMG-based control of a bionic upper limb prosthesis, 2025, 104, 17468094, 107441, 10.1016/j.bspc.2024.107441
    41. Mohamed Ait Yous, Said Agounad, Siham Elbaz, Detection, identification and removing of artifacts from sEMG signals: Current studies and future challenges, 2025, 186, 00104825, 109651, 10.1016/j.compbiomed.2025.109651
    42. Jin Su Kim, Sungbum Pan, Electromyogram Based Personal Recognition Using Attention Mechanism for IoT Security, 2023, 77, 1546-2226, 1663, 10.32604/cmc.2023.043998
    43. Jianpeng Tang, Xugang Xi, Ting Wang, Lihua Li, Jian Yang, Evaluation of the impacts of neuromuscular electrical stimulation based on cortico-muscular-cortical functional network, 2025, 265, 01692607, 108735, 10.1016/j.cmpb.2025.108735
    44. Kuan-Chen Wang, Kai-Chun Liu, Ping-Cheng Yeh, Sheng-Yu Peng, Yu Tsao, TrustEMG-Net: Using Representation-Masking Transformer With U-Net for Surface Electromyography Enhancement, 2025, 29, 2168-2194, 2506, 10.1109/JBHI.2024.3475817
    45. Weidong Zhu, Xugang Xi, Ting Wang, Shunlan Du, Lihua Li, Corticomuscular Coupling Analysis of Dynamic Balance During Eccentric and Concentric Muscle Contractions, 2025, 61, 0953-816X, 10.1111/ejn.70120
    46. Zonglu Zhang, 2025, Chapter 9, 978-3-031-88293-7, 82, 10.1007/978-3-031-88294-4_9
  • Reader Comments
  • © 2020 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(5633) PDF downloads(289) Cited by(46)

Figures and Tables

Figures(9)  /  Tables(5)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog