Gesture Recognition And Interactive Control Based On Surface Electromyography

As a special physiological signal,surface electromyography(s EMG)can effectively reflect the behavior intention of the body.With its natural,direct,and non-invasive characteristics,s EMG has increasingly become a research hotspot in human-computer interaction.As a non-keyboard input technology,the s EMG-based interaction can provide users with a good interactive experience and be widely used in rehabilitation therapy,prosthetic control,robot control,helping the elderly and disabled,and other fields.This dissertation studies a series of problems in the process of gesture interaction based on s EMG,such as high-dimensional feature selection,classifier selection,real-time control,and multi-mode gesture recognition.The main work is as follows:1.To solve the problem of high dimensional features and classifier selection in s EMG pattern recognition,a particle swarm optimization algorithm based on adaptive inertia weight is adopted to optimize the features to reduce the feature dimension,and several classifiers are optimized to achieve higher accuracy of gesture recognition.Firstly,filter the offline signals collected by the subjects,select the sliding window size and extract the features.Then,the features are optimized by the particle swarm optimization algorithm with adaptive inertia weight to reduce the high-dimensional feature set.Finally,use the optimized classifier to recognize the gesture.The experimental results show that this method can effectively improve the classification accuracy of gestures.2.Aiming at the problem of human-computer interaction control based on s EMG signal,the gesture recognition and interactive control system based on s EMG is designed.The system contains the functions of data acquisition,preprocessing,high-dimensional feature extraction and optimization,classifier optimization,and interactive control,and can provide personalized pre-training for different subjects.The optimized features and classifiers are obtained by pre-training,which are used for the real-time interactive control of the manipulator and can effectively ensure control accuracy.3.To deal with the problem of low accuracy in multi-mode gesture recognition,a multi-scale convolution neural network(CNN)based on an attention mechanism is proposed to classify gestures.Firstly,utilize a multi-scale CNN to extract more representative features.Secondly,introduce an attention mechanism to better focus on the key information and screen redundant information,effectively improving the accuracy of gesture recognition.Related experiments show that the multi-scale CNN based on the attention mechanism can effectively improve the recognition accuracy of multi-mode gestures.