CC BY
0ПРИБОРОСТРОЕНИЕ, МЕТРОЛОГИЯ И ИНФОРМАЦИОННО-ИЗМЕРИТЕЛЬНЫЕ
ВЕСТНИК ТСГУ. 2023. № 3 (70)
ПРИБОРЫ И СИСТЕМЫ
YflK 004.9:61
Yu Lei, Liu Tengfei, Yan Zhen, Zou Ji
DETECTION OF ACTIVE SEGMENTS OF EMG SIGNAL BASED ON TIME-DOMAIN VARIANCE COMBINED WITH SHORT-TERM ENERGY
Yu Lei - Master Degree Candidate, School of Electronic and Information Engineering, Changchun University, Changchun, (China); Liu Tengfei - Master Degree Candidate, School of Electronic and Information Engineering, Changchun University, Changchun, (China); Yan Zhen - Master Degree Candidate, School of Electronic and Information Engineering, Changchun University, Changchun, (China); Zou Ji - Phd., Professor, Department of Automation and System Engineering, School of Electronic and Information Engineering, Changchun, (China)
Collecting the electromyographic (EMG) signals of the upper limbs of the forearm to realize hand movement recognition is a technical difficulty in the application of electromyographic signals. In order to solve this difficulty, a dual-threshold endpoint detection algorithm with time-domain variance combined with short-term energy setting is proposed to perform endpoint detection on surface EMG signals. The algorithm is based on short-term energy and zero-crossing rate speech detection. First, the minimum energy threshold and frame data variance are designed. Then the frame length and frame shift to divide the signal into multiple small windows is used. The variance value after the time domain feature extraction is compared with the set threshold value to obtain the effective activity segment. The simulation experiment results show that the algorithm is effective in realizing the detection and interception of the start and end points of the action when processing single-channel or multi-channel surface EMG signals. Compared with the single-threshold method, it has a better detection effect.
Keywords: hand movement recognition, time-domain variance, short-term energy, surface EMG signal, active segment.
Introduction
Motor neurons in the spinal cordare the source of EMG signals. The nerves and muscles involved in each movement produce electrical activity, which can be measured by nerve point diagram or electromyography respectively [1].
© Yu Lei, Liu Tengfei, Yan Zhen, Zou Ji, 2023
BECTHHK TOry. 2023. № 3 (70)
The purpose of detecting the active segment of EMG signal is to obtain the start and end points of muscle contraction from the collected continuous muscle electric signal, so as to distinguish between muscle action and resting time. The research work in this paper is in offline mode, using the surface EMG system TB081 in the EMG laboratory to collect the four commonly used gesture signals, and through preprocessing, feature value extraction, classification and recognition, etc., the comparison is achieved. Good experimental effect [2]. The EMG signal has the characteristics of weakness, nonlinearity and susceptibility to external noise interference. In order to change the above problems, it is necessary to detect the starting point of the EMG signal movement. Endpoint detection is mainly to detect the starting point and ending point of the periodic force contraction. Therefore, an endpoint detection method based on the variance value as the analysis index combined with short-term energy is proposed [3-5]. By comparing the detection methods based on TKE operator and short-term energy threshold processing, the effectiveness of the improved algorithm is proved.
1. Endpoint detection principle
Traditional endpoint detection is based on short-term energy and zero-crossing rate as the threshold. The specific steps are as follows:
Signal framing and windowing.
Assume x(n) It is a sampled time-domain signal. Since the speech signal is a non-stationary and time-varying signal, it needs to be framed Xi (n) For the speech signal of the i-th frame after being framed and windowed, the windowing function is defined as:
Xi(n) = w(n)x[(i-1)Finc + n],
1< n < W length. (1)
In formula (1): Wiength is the frame length; Fine is the frame shift; N is the total number of samples of the EMG signal.
Calculate the data energy of the i-th frame,
Let the data energy of the i-th frame be energy_Frame:
wlength
energy frame = ^ xi (n) . (2)
i= 1
Calculate the zero-crossing rate,
Assuming that the zero-crossing rate of the speech signal of the i-th frame is ZC, it is defined as:
DETECTION OF ACTIVE SEGMENTS OF EMG SIGNAL BASED ON TIME-DOMAIN VARIANCE COMBINED WITH SHORT-TERM ENERGY
BECIHHK Tory. 2023. № 3 (70)
1 wlength
- Z sign[Xi{i +1 )J- sign[xt(i)J
2 i=1
ZC=-
(3)
2. EMG signal processing
In this paper, the E.M.G. System TB081 experimental equipment is used to collect the EMG signals of the four muscles of the upper limbs, and a disposable electrode patch is used to complete the collection on the surface of the skin. The set time parameter is 2.5 s, the number of sampling points is 5000, and the sampling frequency is 2000 Hz. Three electrodes are placed on each muscle and spaced 2 cm apart to form a complete circuit and minimize the impact of noise. A total of 4 volunteers were selected during the experiment and none of them had muscle damage. The information of the experimenter is shown in:
Table 1
Experimenter information
Number Age height Whetherstrained
male 2 24±1 178±2 no
Female 2 23±2 162±2 no
3. Band Variance Endpoint Detection Method
Since the speech signal has a large difference in the spectral characteristics of a speech signal and a non-speech signal, there is a larger amplitude at the resonance peak, and there is only background noise in the non-speech segment, and its amplitude is small. Therefore, some scholars have proposed a voice endpoint detection method based on "band variance" [6].
The EMG signal is similar to the voice signal to a certain extent, so the frequency band variance detection method is used for EMG signal endpoint detection. Since the EMG signal is nonlinear and time-varying, the short-term variance is mainly used in the actual calculation. This method of detecting the start and end endpoints with the short-term frequency band variance as a parameter is called the frequency band variance detection method [7], frequency band variance The steps of the detection method are as follows:
Find the signal spectrum of the i-th frame xi(k), Assume x(n) is the i-th frame signal after frame length and frame shift.The frequency spectrum can be obtained by FFT transformation:
hw. r 2Hkn
xii)= Z Xi)exP ——-j—
i=l v L
Xi (k)={x (0,X- (2),Xi (3)....x, (wlength)} . (4) At this time, the frequency band variance of the i-th frame of speech signal:
BECTHHK TOry. 2023. № 3 (70)
I wlength ,
A —-7 Z (Ix(k)|"Ei . (5)
wlength 1 k-1
In formula (5) is the integrated EMG value of the z-th frame EMG signal.
From equation (5), the data energy of each frame is directly proportional to the degree of signal fluctuation. The greater the energy, the more obvious the ups and downs, Therefore, the greater the value of Dz . Conversely, the smaller the energy, the smoother the fluctuation and the smaller the Dz value.
3.1 Algorithm improvement
Compared with traditional methods based on short-term energy, TKE operator and band variance endpoint detection, an endpoint detection method based on the combination of short-term energy threshold and variance is proposed.
The improved algorithm design steps are as follows:
Normalize the EMG signal data amplitude to [-1, 1] to speed up data convergence;
Frame the EMG signal and add windows. EMG signal xn the signal after the z'-th frame is obtained after frame and windowing is xz ;
Calculate the time-domain variance, and get the variance value of each frame of data from equation (5).
1 wlength ■j
A -— Z(lik)|"Ei) . (6)
wlength k-1
Calculate the data energy of each frame:
1 N
energy _ frame(i) --Z xi . (7)
wlength i-1
3.2 Endpoint detection algorithm flow design
Endpoint detection is to automatically detect the start and end points of a segment of EMG signal, and use the double-threshold threshold comparison method to do start and end point detection. The double-threshold threshold method is based on the minimum short-term energy and variance value as the analysis index.
First, set the short-term energy threshold energy_f according to experience, and set the upper and lower limit variances (sum_s1, sum_s2). If the EMG variance value Var_s is between the lower limit sum_s1 and the upper limit sum_s2, it means that the signal is in the active segment and the variance at this time is output Upper and lower limits, short-term energy value, start and end time t, and number of sampling points L; otherwise, it is not in the effective active section, the active section ends. The design process is shown in fig.1.
DETECTION OF ACTIVE SEGMENTS OF EMG SIGNAL BASED ON TIME-DOMAIN VARIANCE COMBINED WITH SHORT-TERM ENERGY
BECTHHK Tory. 2023. № 3 (70)
Fzg. 1. Endpoint detection process design
4. Experimental process and result analysis
The original data of the EMG signal in the experiment were collected based on the E.M.G.System TB081 experimental equipment. One segment of the signal is randomly selected as the experimental processing object. In the experiment, the sampling frequency is 2000 Hz, frame length wiength is set to 200, and the frame shift Fznc is 50.
4.1 Comparative analysis of different detection algorithms
Kaiser derived the TKE operator to calculate the energy of the audio signal. Due to the similarity between the EMG signal and the audio signal [8], it is widely used
BECTHHK Tory. 2023. № 3 (70)
in detecting the start and end points of the EMG signal action. The TKE operator detection algorithm steps are as follows:
<pA[x(n)]= x2 (n)-x(n +1)x(n -1) . (8)
If x(n), n=1,2,3,4,5.....N is a discrete time sequence of a given length N
x(n) = A cos[w0 (n) + 0~\ . (9)
So there is
¥a[_x(n)] = A2sin2[wo(n)] . (10)
From formula (10), it is known that the TKE operator changes in proportion to the signal amplitude and frequency. When the muscle moves, the amplitude will increase as the muscle contraction intensifies. Therefore, when using the TKE operator to do the start and end of the detection action, a threshold T is often used to judge it:
T = uo + jSo. (11)
Fig. 2 show the output endpoint detection time-domain diagram when different values of j are selected. The abscissa is the start and end time t, and the ordinate is the signal amplitude ^V.
2 4 -1-1-'-1-1-1-
1 12 14 16 18 2 2 2 24
t/S
a) j=3
DETECTION OF ACTIVE SEGMENTS OF EMG SIGNAL BASED ON TIME-DOMAIN VARIANCE COMBINED WITH SHORT-TERM ENERGY
BECIHHK Tory. 2023. № 3 (70)
b) j=7
24 -'-'-1-'-1-'-'-1-
1 5 1 6 1 7 1 8 1 9 2 2.1 2 2 2 3 24 t/S
b) 7 = 15
Fzg. 2. Output endpoint detection time-domain diagram
From the above figure, the larger the j value, the smaller the starting and ending time range determined by the threshold T, resulting in excessive loss of effective information in the signal; if the j value is selected too small, the effective segment that is intercepted will contain too much background noise. It can be seen from the figure that the amplitude of the EMG signal does not change much, all between 0.25 ... 0.40 mV, so the effective activity area can only be extracted when the j value is small; and the algorithm cannot separate the background noise and the effective segment of the activity signal Separate.
BECTHHK TOry. 2023. № 3 (70)
4.2 Algorithm verification based on short-term energy and frequency band variance
In order to overcome the influence of different degrees of noise in the traditional detection algorithm, fig. 3 shows the time-domain waveform after the filter processing, and fig. 4 shows the amplitude spectrum after the filter processing. Fig. 5 is a window diagram of the variance threshold value. During the frame shift process, the upper and lower limit variance thresholds are used to perform segmented detection on fig. 6 to obtain the time-domain waveform diagram 7 after the detection. After the algorithm is processed, the output start and end time and the number of samples L are plotted as shown in Table 2.
timet(s)
Fig. 3. Time domain diagram denoising
Fig. 4. Spectrum diagram before and after denoising
DETECTION OF ACTIVE SEGMENTS OF EMG SIGNAL BASED ON TIME-DOMAIN VARIANCE COMBINED WITH SHORT-TERM ENERGY
BECIHHK TOry. 2023. № 3 (70)
20 30 40 50 number of samples/1 Fig. 5. VAR threshold window diagram
Fig. 6. Time domain diagram of original signal
BECTHHK TOry. 2023. № 3 (70)
0.3 04 0 5 0 6 0.7 0 8 0 9 1
t/S
Fig. 7. Time domain diagram of threshold detection
Output parameters
Table 2
energyj sum s1 sum s2 t1 t2 L
0.0025 0.0186 0.0347 0.3100 1.0720 1525
It can be seen from fig. 3 and 4 that the designed Butterworth low-pass filter and high-pass filter preprocess the original EMG signal, which can effectively remove the influence of low-frequency noise and power frequency noise on the signal. Fig. 5 is the variance value after time domain feature extraction. After the segmentation processing of fig. 6 by the frame length wiength , the detected time domain map shown in fig. 7 is obtained. For the same signal segment, compared with the traditional TKE operator method, the improved algorithm has a better detection effect on the active segment of muscle activity.
5. Conclusion
This paper proposes that it is effective and feasible to use time-domain variance combined with short-term energy to set multi-threshold thresholds to determine the starting and ending points of electrical signals. The above analysis and experiments show that this method effectively establishes the internal connection between the EMG signal and the forearm movement, accurately determines the start and end time of the movement, and provides a theoretical basis for subsequent gesture movement recognition.
DETECTION OF ACTIVE SEGMENTS OF EMG SIGNAL BASED ON TIME-DOMAIN VARIANCE COMBINED WITH SHORT-TERM ENERGY
BECTHHK Tory. 2023. № 3 (70)
6.Acknowledgments
This work was supported by the project of Jilin Provincial Science and Technology Department (20210402081GH), the Project of Jilin Provincial Development and Reform Commission (2023C042-4), the Innovation and Entrepreneurship Talent Funding Project of Jilin Province (2023RY17).
References
1. Esquivel-Frausto M.E., Guerrero J.A., Macias-Diaz J.E. Activity pattern detection in electroneurographic and electromyogram signals through a heteroscedastic change-pointmethod // Mathematical Biosciences. 2010. 224(2). P.109-117.
2. Hand movement recognition method based on EMG signal level classification / Zhao Mandan, Li Dongxu, Fan Caizhi, Meng Yunhe // Beijing Biomedical Engineering. 2014. 33(05). P. 490-496.
3. Di Fabio, Richard P. Reliability of Computerized Surface Electromyography for Determining the Onset of Muscle Activity // Physical Therapy. 1987. 67 (1). P. 43-48.
4. Endpoint detection of distributed fiber sensing systems based on STF Tal-gorithm / Kun Liu, Pengfei Ma, Jianchang An, Zhichen Li, Junfeng Jiang, Pengchen Li, Liwang Zhang, Tiegen Liu // Optics&Laser Technology. 2019. Vol.114. P. 122126.
5. Wang J.Q., An Y. F., Zhu Z. H. Continuous speech segmentation method based on energy envelope // Digital technology and applications. 2013. Vol. 9. P.108-109.
6. Endpoint detection based on spectral subtraction and uniform sub-band frequency band variance method / Wang Wei, Hu Guiming, Yang Li et al. // Electroa-coustic Technology. 2016. Vol. 5. P. 40-43.
7. Chen Haoze, Zhang Zhijie. Speech endpoint detection method based on the combination of energy and frequency band variance // Science Technology and Engineering. 2019. Vol. 26.
8. Yan Zhengxiang. Human upper limb movement recognition and muscle fatigue detection based on sEMG signal. Beijing: Beijing University of Technology, 2018.
ВЕСТНИКТОГУ. 2023. № 3 (70)
Заглавие: Обнаружение активных сегментов ЭМГ-сигнала на основе дисперсии во временной области в сочетании с кратковременной энергией
Авторы:
Ю Лей - Чанчуньский университет (КНР); Лю Тенфэй - Чанчуньский университет (КНР); Ян Жен - Чанчуньский университет (КНР); Зоу Жи - Чанчуньский университет (КНР)
Аннотация: Сбор электромиографических сигналов верхних конечностей предплечья для осуществления распознавания движений руки представляет собой техническую трудность при применении электромиографических сигналов. Чтобы решить эту трудность, предлагается алгоритм определения конечной точки с двойным порогом с дисперсией во временной области в сочетании с кратковременной настройкой энергии для выполнения определения конечной точки по сигналам поверхностной ЭМГ. Алгоритм основан на обнаружении речи с кратковременной энергией и скоростью пересечения нуля. Сначала разрабатывается минимальный порог энергопотребления и дисперсия данных кадра. Затем используйте длину кадра и сдвиг кадра, чтобы разделить сигнал на несколько небольших окон. Значение отклонения после извлечения объекта во временной области сравнивается с установленным пороговым значением для получения эффективного сегмента активности. Результаты имитационного эксперимента показывают, что алгоритм эффективен в реализации обнаружения и перехвата начальной и конечной точек действия при обработке однока-нальных или многоканальных сигналы поверхностной ЭМГ. По сравнению с однопороговым методом, он обладает лучшим эффектом обнаружения.
Ключевые слова: Распознавание движений рук, дисперсия во временной области, кратковременная энергия, поверхностный ЭМГ-сигнал, активный сегмент.
