Study On Rehabilitation Motion Classification And Prediction Algorithm Of Dual-arm Robot Based On SEMG

According to statistics,more than 120 million people worldwide need rehabilitation treatment,while in China,the number of disabled population is about85 million,of which about 70% or more need rehabilitation training.Therefore,rehabilitation training has a broad market and far-reaching social significance.With the progress of technology,the application of robotics in the field of rehabilitation has gradually become a hot spot for research.Robots can provide precise,safe and controlled rehabilitation training,avoiding the limitations of manual operation.Since EMG signals precede limb movements,the patient’s muscle contraction signals can be captured earlier,thus enabling faster and more accurate control of rehabilitation robots,which has now become an latent means of rehabilitation training for people with disabilities.In this context,this study explores the action classification prediction algorithm based on surface EMG signals with a two-arm robot rehabilitation training system,aiming to improve the accuracy and efficiency of robot rehabilitation training.The main research contents are as follows:First,the two-armed robot was analyzed,a research plan was developed based on the requirements of action classification prediction,and the action segment dataset establishment and online action segment discrimination method were designed.In the action segment dataset establishment,an angle signal-based action segment classification method is proposed to establish the dataset by fusing the joint angle information,and the analysis is compared with multiple methods according to the accuracy magnitude.In the online action segment segmentation,the average absolute value online action segmentation method based on majority voting is proposed according to the combination of acquisition channel discrimination,and the accuracy,F1-score and time consumption are compared and analyzed with various methods.Next,the feature reduction and classification prediction algorithms are studied.In feature reduction,a feature selection algorithm based on the maximum correlation minimum redundancy criterion combined with particle swarm is proposed by combining statistical laws and population intelligence algorithms,and is compared and analyzed according to the accuracy and time consuming.In the selection of action classification algorithms,three action classification algorithms are constructed based on different classification principles and comparative analysis is performed according to the accuracy rate.In the selection of angle prediction algorithms,a long and short-term memory neural network regression algorithm was constructed.Finally,experimental verification was conducted.The real-time online experiments included action classification real-time experiment,action classification real-time stability experiment,and angle prediction real-time experiment.The two-arm robot rehabilitation action control experiment includes the online accuracy experiment of action classification,the stability experiment of action classification accuracy and the online angle prediction fitting experiment.The experimental results show that the proposed classification and prediction algorithm can effectively complete the classification and prediction of rehabilitation movements,meet the design requirements,and can provide an effective aid for rehabilitation training.