Fig 1.
Overview of the proposed model.
The model expresses an artificial EMG signal zt at t, based on a process involving white Gaussian noise passed through a shaping filter H and variance
. Variance
is the value at t of a random variable σ2 having a distribution determined by a commanded muscle force component of variance
and signal-dependent noise ε according to the commanded muscle force
.
Fig 2.
The subjects were seated with the right upper arm pointing downward, the right forearm bent forward to the horizontal, and the palm turned upward. EMG signals were recorded from a pair of electrodes attached to the biceps brachii while the subjects were weighted with a load on the right wrist and maintained the right elbow on a desk.
Fig 3.
EMG signals were recorded using six electrodes (L = 6: Ch. 1: extensor carpi ulnaris; Ch. 2: flexor digitorum profundus; Ch. 3: extensor digitorum; Ch. 4: flexor carpi ulnaris; Ch. 5: triceps brachii; Ch. 6: biceps brachii) at a sampling frequency of 1000 Hz.
Fig 4.
Screenshot of the EMG measurement system.
The bar graph shows the muscle activation level of the agonist muscle.
Fig 5.
Examples of measured and artificial EMG signals for each load weight.
(a) Measured EMG signals. (b) Artificial EMG signals generated from the measured EMG signals under each load weight based on the proposed model. (c) Artificial EMG signals generated from the measured EMG signals under a 1000 g load based on the proposed model with the variance modulation.
Fig 6.
Average absolute percentage error in average amplitude for each load weight.
(a) Influence of the recording source of the preset parameters. (b) Comparison of each generation method. Error bars represent the standard deviations for all subjects.
Fig 7.
Correlation coefficient in power spectrum for each load weight.
(a) Influence of the recording source of the preset parameters. (b) Comparison of each generation method. Error bars represent the standard deviations for all subjects. All correlation coefficients had p < 0.001.
Fig 8.
Root mean square error (RMSE) in kurtosis for each load weight.
(a) Influence of the recording source of the preset parameters. (b) Comparison of each generation method. Error bars represent the standard deviations for all subjects.
Fig 9.
Examples of artificial and measured EMG signals with a muscle activation level of 80%MVC.
(a) Artificially generated EMG signals for each channel. (b) Measured EMG signals for each channel. The artificial and measured EMG signals are used for a part of the test and the training data in motion classification, respectively.
Fig 10.
Average classification rates of each method.
(a) Muscle activation level of 40%MVC. (b) Muscle activation level of 80%MVC. Error bars in the results of Subjects A–D represent the standard deviations for all trials and those in the average of all subjects represent the standard deviations for all subjects.