Research Of Surface EMG Classification Of Hand Movement Based On AdaBoost Algorithm

Surface EMG is a biological signals collected from the surface of the musclethrough surface electrodes. It reflects the function and action status of muscle to someextent. It is a key issue that the information extracted from surface EMG accuratelyand the action recognition. In this paper, we research the feature extraction and patternrecognition of surface EMG to the recognition and classification on hand movements.This paper studies the following aspects:1. We read a lot about the papers of surface EMG, pattern recognition,classification and optimization algorithm. We study the theoretical basis and resultingmechanism of surface EMG, the research progress on the classification issue of theartificial limb and surface EMG at home and abroad in recent years.2. According to the topography, we determine the muscle selected relating to thethe hand movements as the source of surface EMG. We completed four kinds of handmovements accurately captured by MQ-8EMG logger. Because the original EMGcontains much noise, this paper designs the high-pass filter、low-pass filter, notchfilter by sofware to complete original EMG signal denoising and retain effective EMG.Because continuous EMG is inconvenient for feature extraction, we use the way ofmoving window energy to intercept the surface EMG. We complete the pretreatmentof surface EMG which will be beneficial to the pattern recognition of fourmovements.3. We extract the feature of the surface EMG signals on the time domain,frequency domain, time-frequency domain.We compare the three way to select thefeatue which can clearly distinguish the four movements as the basis of surface EMGpattern recognition. This paper selects the variance of wavelet packet coefficient asfeature vectors to distinguish the four movements by the wavelet packet analysis. Weminimize the number of feature vector, while ensuring the classification results.4. According to the common cassification algorithm of pattern recognition, we select the appropriate method for the classification of surface EMG. We research theBP neural network, SVM algorithm, AdaBoost algorithm and the improved AdaBoostalgorithm. This paper use the BP neural network, SVM algorithm, AdaBoostalgorithm as a comparative experiment. We calculate the recognition rate of the fourmovements by the three methods. We analyze the classification advantages ofAdaBoost algorithm for surface EMG and the recognition ability of AdaBoostalgorithm under the interference of fatigue EMG. Meanwhile, we compute theiterations of the classification algorithms to complete the action classification.5. We carry out the simulation experiments. The results show that AdaBoostalgorithms studied can ensure accuracy in the training process of surface EMG. At thesame time, the identification of fatigue state also has a good effect which achieves theintended purpose of this paper.