EMG-Based Automatic Gesture Recognition Using Lipschitz-Regularized Neural Networks

Single-channel surface electromyography signal classification with variational mode decomposition and entropy feature for lower limb movements recognition

We address adversarial attacks on the actuators at the joints of legged robots trained by deep reinforcement learning. The vulnerability to the joint attacks can significantly impact the safety and robustness of legged robots. In this study, we demonstrate that the adversarial perturbations to the torque control signals of the actuators can significantly reduce the rewards and cause walking instability in robots. To find the adversarial torque perturbations, we develop black-box adversarial attacks, where, the adversary cannot access the neural networks trained by deep reinforcement learning. The black box attack can be applied to legged robots regardless of the architecture and algorithms of deep reinforcement learning. We employ three search methods for the black-box adversarial attacks: random search, differential evolution, and numerical gradient descent methods. In experiments with the quadruped robot Ant-v2 and the bipedal robot Humanoid-v2, in OpenAI Gym environments, we find that differential evolution can efficiently find the strongest torque perturbations among the three methods. In addition, we realize that the quadruped robot Ant-v2 is vulnerable to the adversarial perturbations, whereas the bipedal robot Humanoid-v2 is robust to the perturbations. Consequently, the joint attacks can be used for proactive diagnosis of robot walking instability.

Surface electromyographic (sEMG) signal contains abundant information such as joint torque and joint motion, which is widely used in human-computer interactive intelligent rehabilitation equipment. In this work, the ankle torque of lower limb is taken as the research object, and the feature parameters of sEMG which represent the fatigue state are analysed. Advance prediction of fatigue features for specific time periods was performed using a normalized minimum average square (NLMS) filter. While the modified cerebellar model neural network (WFCMNN) is used to classify fatigue, which can be divided into three states, namely no fatigue, transition to fatigue, and fatigue. The results show that the accuracy of classification is 96.429%, which is better than other advanced models. At the same time, sEMG signal is used to predict fatigue in advance, which can solve the problem of differences between different individuals. Such strategy is helpful for doctors and physiotherapists to carry out rehabilitation treatment for patients, as a pre judgment and diagnosis index.

In this work, a system for monitoring human postures, seizures, and falls from bed using received signal strength indicator (RSSI) and surface electromyography (sEMG) signals is studied through experiments. In this proposed system, a person who is located inside a wireless link is monitored by considering the change in measured RSSI signals as the 2.4 GHz IEEE 802.15.4 signals received at a receiver. Human motions in bed that affect RSSI levels can be captured. Thus, with this technique, it does not raise a privacy concern compared with vision-based technology. Additionally, sEMG signals associated with muscle movements from human postures are recorded from the human body’s abdominal muscles. Eight different activities, including normal and critical events, are tested and evaluated. Experimental results indicate that the proposed system could automatically monitor different human postures in real-time. RSSI and sEMG signals correlated to each posture have their own patterns. Furthermore, the relationship between human behaviors and RSSI and sEMG levels is summarized.

The efficacy of using surface electromyography (sEMG) signals to understand the biomechanics of movement has been an active area of research. The need for a reliable standard to monitor/predict rehabilitative outcomes and an efficient control to drive orthotics and exoskeletons has motivated this work to use sEMG as an estimate for joint angle prediction. This work involves acquisition of 8 channel sEMG signals from lower extremity during knee articulation, involving flexion and extension for 20 able bodied subjects. Simultaneous knee joint angle measurement from goniometer is used to train a Back Propagation Neural Network (BPNN) to estimate knee Range of motion (ROM) with Root Mean Square values of 8 sEMG signals as input. Knee joint angles at full ROM for flexion $(90.9\pm 2.4)$ and extension $(19.1 \pm 2.6)$ were obtained. Estimation accuracy based on average Mean square error for flexion (0.146 $\pm 0.197$ ) and extension $(0.098\pm 0.129)$ ) with correlation factor (r) between actual and estimated values were found to be 0.93 and 0.95 respectively. Estimated flexion and extension angles were used to actuate the knee joint of a Virtual Human model. The results suggest that a high degree of correlation can be achieved between sEMG and kinematic measurements.

Muscle fatigue is commonly experienced in both normal and subjects with neuromuscular disorders. Surface electromyography (sEMG) signals are useful technique for analyzing muscle fatigue. sEMG signals are highly nonstationary and exhibit complex nonlinear characteristic in dynamic contractions. In this work, an attempt is made to classify sEMG signals recorded from biceps brachii muscles in nonfatigue and fatigue using multifractal features. The signals are recorded from 26 healthy normal adult subjects while performing standard experimental protocol involving dynamic contraction. The preprocessed signals are divided into six segments. The first and last segments are considered as nonfatigue and fatigue conditions respectively. The signals are then subjected to multifractal detrended moving average algorithm and eight multifractal features are extracted from both conditions. Further, information gain (IG) based ranking is used for reducing the number of features. Three different classification algorithms are employed namely, k-Nearest Neighbor algorithm (kNN), Naive Bayes (NB) and logistic regression (LR) for classification. The results show that signals exhibit multifractal characteristics and the multifractal features such as, generalized Hurst exponent, degree of multifractality and scaling exponent slope are significantly different in fatigue condition. The Hurst exponent for small fluctuation and degree of multifractality are found to be very highly significant feature. The LR and kNN classifier performance gave an accuracy of 84% and 82% respectively. This method of using multifractal features appears to be useful in classifying sEMG signals in dynamic contraction. This study can also be extended to classify fatigue condition in various neuromuscular disorders.

This paper investigates the output‐constrained control problem for physical human–robot interaction using electromyography signals. An interaction force estimation method based on surface electromyography (sEMG) signals is introduced, where random forest is provided to extract interaction force information from sEMG signals. Using sEMG, the interaction force between the robotic and human operator can be separated from that in the environment. An adaptive optimal tracking controller based on robot velocity prediction error feedback is designed, in which an asymmetric time‐varying integral barrier Lyapunov function is used to deal with dynamic spatial constraints. Finally, stability analysis is provided, and the effectiveness is verified by experiments based on a 7‐degree‐of‐freedom Franka Panda robot.

Recognizing the breathing resistances of wearing respirators from respiratory and sEMG signals with artificial neural networks

Various myoelectric prostheses controlled by electromyography (EMG) signals have been developed. However, there have been few studies that provide fast and accurate methods to predict handgrip force from EMG signals. Rapid and precise handgrip force prediction is required, especially for the real-time control system of myoelectric prostheses. In this study, extreme learning machine (ELM) is applied to predict handgrip force from surface EMG signals of forearm muscles. Furthermore, ELM is compared with support vector machine (SVM) and multiple nonlinear regression (MNLR). The below 10 % of the surface EMG and handgrip force signals were cut away, and then the root mean square feature extracted from the modified surface EMG signals was taken as input vector for these three kinds of predicting mechanisms. For the testing dataset, ELM achieved a slightly larger root mean squared error than SVM did and a smaller one than MNLR did. Meanwhile, all three methods showed high correlation coefficients. For the total processing time, ELM and MNLR consumed much less time than SVM did. Experimental results demonstrate that ELM possesses a relatively good accuracy and little consumed time, although SVM is effective for handgrip force estimation in terms of accuracy. Overall, ELM has a promising potential for predicting handgrip force rapidly and precisely.

This study explores the integration of surface electromyography (sEMG) signals for precise servo control in real-time hand movement detection. It also further utilizes the feedback system designed to measure effects of workout exercises on forearm sEMG signals and delves into how different time and types of workouts affect forearm muscle response. Medical electrodes captured sEMG signals from forearm muscles and converted them into digital signals using an Arduino UNO R3 microcontroller board and an Olimex EMG shield to monitor hand pose variations. The digital signals were then processed to calculate the root mean square (RMS) voltage, which was then used to control the position of a servo motor. The implementation of a proportional-derivative (PD) controller further enhanced the accuracy and stability of the servo movements. Then the feedback system was used to measure how sEMG signals vary due to hand opening and closing before and after a range of workout exercises with varying time and workout types, and the results were compared to reach the conclusion that both workout time and type influences how forearm muscles respond to hand closing and opening. Future work may explore the effect of more time and types of workouts on muscle response, expand the range of test subjects, and improve the accuracy and stability of the test system

Surface electromyography (sEMG) signal is widely used in lower limb exoskeleton system because it can reflect human motion intention. Aiming at the problem of low decoding accuracy of joint motion angle of lower limb exoskeleton robot, a knee motion angle prediction method based on sEMG signal is proposed. Firstly, in order to solve the problem of high time efficiency and insufficient stability of single time domain feature extraction method for sEMG signal, the eigenvalues are extracted from time domain, frequency domain and time-frequency domain respectively, and the prediction error of different channel sEMG signal feature parameter combination model is analyzed. Then, an innovative particle swarm optimization LSTM deep learning algorithm (PSO-LSTM) is proposed to establish the mapping model between sEMG signal and knee joint angle, and determine the PSO-LSTM model corresponding to the optimal feature parameter combination to realize the accurate prediction of knee joint angle. Finally, the experimental results show that the predicted results of the model are highly consistent with the real angle values, which is helpful to improve the human-computer cooperation of the lower limb exoskeleton system.KeywordssEMG signalLower extremity exoskeleton robotIntention perceptionPSO-LSTM algorithm

The successful emergence of deep learning (DL) in wireless system applications has raised concerns about new security-related challenges. One such security challenge is adversarial attacks. Although there has been much work demonstrating the susceptibility of DL-based classification tasks to adversarial attacks, regression-based problems in the context of a wireless system have not been studied so far from an attack perspective. The aim of this paper is twofold: (i) we consider a regression problem in a wireless setting and show that adversarial attacks can break the DL-based approach and (ii) we analyze the effectiveness of adversarial training as a defensive technique in adversarial settings and show that the robustness of DL-based wireless system against attacks improves significantly. Specifically, the wireless application considered in this paper is the DL-based power allocation in the downlink of a multicell massive multi-input-multi-output system, where the goal of the attack is to yield an infeasible solution by the DL model. We extend the gradient-based adversarial attacks: fast gradient sign method (FGSM), momentum iterative FGSM, and projected gradient descent method to analyze the susceptibility of the considered wireless application with and without adversarial training. We analyze the deep neural network (DNN) models performance against these attacks, where the adversarial perturbations are crafted using both the white-box and black-box attacks.

Hand gesture recognition is a challenging task due to the complexity of hand movements and to the variety among the same gesture performed by distinct subjects. Recent technologies, such as Kinect sensor, provide new opportunities, allowing to capture both RGB and depth (RGB-D) images, which offer high discriminant information for efficient hand gesture recognition. In the aspect of feature extraction, the traditional methods process the RGB and depth information independently. In this paper, we propose a robust hand gesture recognition system based on a new feature extraction method, fusing RGB images and depth information simultaneously, by using the quaternion algebra that provide a more robust and holistical representation. In fact, we introduce, for the first time, a novel type of feature extraction method, named quaternion Tchebichef moment invariants. The novelty of the proposed method in this paper lies in the direct derivation of invariants from their orthogonal moments, based on the algebraic properties of the discrete Tchebichef polynomials. The proposed approach based on quaternion algebra is suggested to process the four components holistically, for a robust and efficient hand gesture recognition system. The obtained experimental and theoretical results demonstrate that the present approach is very effective for addressing the problem of hand gesture recognition and have proved its robustness against geometrical distortion, noisy conditions and complex background compared to the state of the art, indicating that it could be highly useful for many computer vision applications.

In this work, an attempt has been made to analyze the cyclostationarity of surface electromyography (sEMG) signals recorded during dynamic contraction of biceps brachii muscle. Twenty five healthy adult volunteers have participated in this study. The recorded signals are preprocessed and segmented into three zones, namely, non-fatigue zone, first muscle discomfort zone and fatigue zone. These signals are subjected to strip spectral correlation algorithm to estimate the spectral correlation density (SCD). Cyclic domain profile (CDP) is extracted from the normalized SCD magnitude. The area of CDP is calculated for all the three zones. The results show that strip spectral correlation algorithm based SCD estimation is able to demonstrate the cyclostationary property of sEMG signal in all three zones. It is also observed that the normalized SCD magnitude spectrum is distinct for all cases. The CDP area shows the significant variation in fatigue zone. Further it appears that cyclostationarity based on SCD can be used to differentiate different neuromuscular and pathological conditions.

The analysis of surface electromyography (sEMG) signals is significant in the detection of muscle fatigue. These signals exhibit a great degree of complexity, nonlinearity, and chaos. Also, presence of high degree of fluctuations in the signal makes its analysis a difficult task. This study aims to analyze the nonlinear dynamics of muscle fatigue conditions using Fuzzy recurrence networks (FRN). Dynamic sEMG signals are measured from biceps brachii muscle of 45 normal subjects referenced to 50% of maximal voluntary contractions (MVC) for this. Recorded signals are then pre-processed and divided into ten equal parts. FRNs are transformed from the signals. The network features, namely average weighted degree (AWD) and Closeness centrality (CC) are extracted to analyze the muscle dynamics during fatiguing conditions. The decrease in these features during fatigue indicates a reduction in signal complexity and an increase in complex network stiffness. Both AWD and CC features are statistically significant with [Formula: see text]. Further, these features are classified using Naïve Bayes (NB), k nearest neighbor (kNN) and random forest (RF) algorithms. Maximum accuracy of 96.90% is achieved using kNN classifier for combined FRN features. Thus, the proposed features provide high-quality inputs to the neural networks that may be helpful in analyzing the complexity and stiffness of neuromuscular system under various myoneural conditions.