Surface Electromyogram Signal Classification Method For Forestry Machinery Gesture Control

With the rapid development of forestry machinery,the way of human operations in forests has experienced great changes and applying gesture recognition to the control of forestry machinery will be more convenient,more efficient and more humanized.Therefore,it is a research hotspot to recognize gestures accurately through analyzing the surface electromyogram(sEMG)signal by using signal processing technology.However,the recognition rate of gestures through sEMG signals from several subjects at the same time is not high enough since the traditional gesture recognition methods extract the few eharacteristies or use the classifier with low generalization performanee.To solve this problem,this paper utilizes the feature fusion vectors from 27 kinds of features and proposed a classification method based on directed acyclic graph(DAG)and support vector machine(SVM)for hand gesture recognition.The main research contents and conclusions of this paper are as follows:1.In this thesis,four-channel sEMG signals of six gestures,which included fist,finger spread,palm supination,Palm pronation,palm lateral supination and palm lateral pronation,were collected from four target arm muscles,which included palmaris longus muscle,brachioradialis muscle,extensor digitorum muscle and extensor carpi ulnaris muscle,of 15 healthy subjects and each subject did each gesture for ten times.Then the raw sEMG signals were preprocessed,including denoising,amplification and detection of active segment;2.Time-domain analysis,frequency-domain analysis method,time-and frequency-analysis method and nonlinear dynamic method were utilized to calculate the characteristics of the sEMG signal,and 27 kinds of features were obtained.Compared with the feature simple combination,the recognition result of feature vectors which were processed by feature extraction algorithms was better.Six kinds of classification algorithms were used to identify and test the data processed by four kinds of feature extraction methods.The result showed that the recognition performance of the data processed by linear discriminant analysis was the best,and the optimal dimensions were 5?8;3.An improved recognition method of DAGSVMerr algorithm based on SVM and DAG was proposed.This algorithm calculated the value of separation measure by using erroneous recognition rates from pre recognition.Compared with one-against-the rest,one-against-one and other two DAG-SVM classifiers calculating the value of separation measure based on Euclidean distance,results show that the recognition performance of DAGSVMerr algorithm were better than others and the average recognition rate reached 99.4%.Then,compared with the five commonly used classification algorithms,the experimental results show that the DAGSVMerr algorithm had the highest recognition rate.