Recognition of forearm muscle activity by continuous classification of multi-site mechanomyogram signals

Implementation of Hand Gesture Recognition Device Applicable to Smart Watch Based on Flexible Epidermal Tactile Sensor Array

Ever since the development of digital devices, the recognition of human gestures has played an important role in many Human-Computer interface applications. Various wearable devices have been developed, and inertial sensors, magnetic sensors, gyro sensors, electromyography, force-sensitive resistors, and other types of sensors have been used to identify gestures. However, there are different drawbacks for each sensor, which affect the detection of gestures. In this paper, we present a new gesture recognition method using a Flexible Epidermal Tactile Sensor based on strain gauges to sense deformation. Such deformations are transduced to electric signals. By measuring the electric signals, the sensor can estimate the degree of deformation, including compression, tension, and twist, caused by movements of the wrist. The proposed sensor array was demonstrated to be capable of analyzing the eight motions of the wrist, and showed robustness, stability, and repeatability throughout a range of experiments aimed at testing the sensor array. We compared the performance of the prototype device with those of previous studies, under the same experimental conditions. The result shows our recognition method significantly outperformed existing methods.

A CNN-SVM combined model for pattern recognition of knee motion using mechanomyography signals

The commonly used classifiers for pattern recognition of human motion, like backpropagation neural network (BPNN) and support vector machine (SVM), usually implement the classification by extracting some hand-crafted features from the human biological signals. These features generally require the domain knowledge for researchers to be designed and take a long time to be tested and selected for high classification performance. In contrast, convolutional neural network (CNN), which has been widely applied to computer vision, can learn to automatically extract features from the training data by means of convolution and subsampling, but CNN training usually requires large sample data and has the overfitting problem. On the other hand, SVM has good generalization ability and can solve the small sample problem. Therefore, we proposed a CNN-SVM combined model to make use of their advantages. In this paper, we detected 4-channel mechanomyography (MMG) signals from the thigh muscles and fed them in the form of time series signals to the CNN-SVM combined model for the pattern recognition of knee motion. Compared with the common classifier performing the classification with hand-crafted features, the CNN-SVM combined model could automatically extract features using CNN, and better improved the generalization ability of CNN and the classification accuracy by means of combining the SVM. This study would provide reference for human motion recognition using other time series signals and further expand the application fields of CNN.

Real-time continuous recognition of knee motion using multi-channel mechanomyography signals detected on clothes

Mechanomyography (MMG) signal has been recently investigated for pattern recognition of human motion. In theory, it is no need of direct skin contact to be detected and unaffected by changes in skin impedance. So, it is hopeful for developing wearable sensing device with clothes. However, there have been no studies so far to detect MMG signal on clothes and verify the feasibility of pattern recognition. For this study, 4-channel MMG signals were detected on clothes from the thigh muscles of 8 able-bodied participants. The support vector machines (SVM) classifier with 4 common features was used to recognize 6 knee motions and the average accuracy of nearly 88% was achieved. The accuracy can be further improved up to 91% by introducing a new proposed feature of the difference of mean absolute value (DMAV), but not by root mean square (RMS) or mean absolute value (MAV). Furthermore, the first-order Markov chain model was combined with the SVM classifier and it can avoid the misclassifications in some cases. For application to wearable power-assisted devices, this study would promote the developments of more flexible, more comfortable, and minimally obtrusive wearable sensing devices with clothes and recognition techniques of human motion intention.

A preliminary investigation on the utility of temporal features of Force Myography in the two-class problem of grasp vs. no-grasp in the presence of upper-extremity movements

BACKGROUND

In upper-extremity stroke rehabilitation applications, the potential use of Force Myography (FMG) for detecting grasping is especially relevant, as the presence of grasping may be indicative of functional activity, which is a key goal of rehabilitation. To date, most FMG research has focused on the classification of the raw FMG signal (i.e. instantaneous FMG samples) in order to determine the state of the hand. However, given the temporal nature of force generation during grasping, the use of temporal feature extraction techniques may yield increased accuracy. In this study, the effectiveness of classifying temporal features of the FMG signal for the two-class grasp detection problem of "grasp" versus "no grasp" (i.e. no object in hand) was evaluated with ten healthy participants. The experimental protocol comprised grasp and move tasks, requiring the use of six different grasp types frequently used in daily living, in conjunction with arm and hand movements. Data corresponding to arm and hand movements without grasping were also included to evaluate robustness to false positives. The temporal features evaluated were mean absolute value (MAV), root mean squared (RMS), linear fit (LF), parabolic fit (PF), and autoregressive model (AR). Off-line classification performance of the five temporal features, with a 0.5 s extraction window, were determined and compared to that of the raw FMG signal using area under the receiver operating curve (AUC).

RESULTS

The raw FMG signal yielded AUC of 0.819 ± 0.098. LF and PF resulted in the greatest increases in classification performance, and provided statistically significant increases in performance. The largest increase obtained was with PF, yielding AUC of 0.869 ± 0.061, corresponding to a 6.1% relative increase over the raw FMG signal. Despite the additional fitting term provided by PF, classification performance did not significantly improve with PF when compared to LF.

CONCLUSIONS

The results obtained indicate that temporal feature extraction techniques that derive models of the data within the window may yield modest improvements in FMG based grasp detection performance. In future studies, the use of model-based temporal features should be evaluated with FMG data from individuals with stroke, who might ultimately benefit from this technology.

Mechanomyography and muscle function assessment: a review of current state and prospects

Previous studies have explored to saturation the efficacy of the conventional signal (such as electromyogram) for muscle function assessment and found its clinical impact limited. Increasing demand for reliable muscle function assessment modalities continues to prompt further investigation into other complementary alternatives. Application of mechanomyographic signal to quantify muscle performance has been proposed due to its inherent mechanical nature and ability to assess muscle function non-invasively while preserving muscular neurophysiologic information. Mechanomyogram is gaining accelerated applications in evaluating the properties of muscle under voluntary and evoked muscle contraction with prospects in clinical practices. As a complementary modality and the mechanical counterpart to electromyogram; mechanomyogram has gained significant acceptance in analysis of isometric and dynamic muscle actions. Substantial studies have also documented the effectiveness of mechanomyographic signal to assess muscle performance but none involved comprehensive appraisal of the state of the art applications with highlights on the future prospect and potential integration into the clinical practices. Motivated by the dearth of such critical review, we assessed the literature to investigate its principle of acquisition, current applications, challenges and future directions. Based on our findings, the importance of rigorous scientific and clinical validation of the signal is highlighted. It is also evident that as a robust complement to electromyogram, mechanomyographic signal may possess unprecedented potentials and further investigation will be enlightening.