The Research Of Fall Prediction Method Based On Surface Electromyography

In recent years,the aging of society has become more and more common,and the health of the elderly has gradually become a hot topic of social concern.Fall is a high-incidence event among the elderly,which seriously threatens the physical and mental health.After a fall,if it can be detected in time and send out a distress signal,it can speed up the pace of medical rescue arrival and win valuable time for the elderly.For this reason,researchers at home and abroad have done a lot of related research.But falls have already happened and injuries have been caused.Therefore,another research direction,fall prediction,has gradually developed.The core problem of fall prediction is to detect falls during the period from body imbalance to collision between body and ground after falls occur,so as to gain time for opening protective devices,avoid or mitigate the damage caused by collisions,and distinguish them from activities of daily life.Surface electromyography(sEMG)signal is an important electrophysiological signal,which reflects the intention and state of muscle activity during human movement,and contains a lot of information related to limb movement.In this study,sEMG signal was used as signal source,and a fall prediction method based on sEMG signal was proposed.The main work was as follows:(1)Studied the roles of lower limb muscle blocks in human activities,and determined the main muscles affecting walking stability,to provide the basis for signal acquisition position.Successfully collected the lower limb sEMG signals of 20 subjects in falls and activities of daily life to complete data acquisition.To improve the recognition accuracy,the original sEMG signals were pre-processed,noise reduction and segmentation to provide reliable information for subsequent fall prediction.(2)Support Vector Machine(SVM)was used to predict falls.The pre-processed sEMG signals were extracted feature values,constructed feature vectors.Then designed and trained SVM to obtain the classification model and detected the test data.Sliding time window technology was used to simulate the online situation.In order to improve the performance of SVM,cross validation method was used to optimize penalty parameter C and kernel function g.The sensitivity of the recognition results was 83%,the specificity was 92%,the accuracy was 90%,and the average lead time was 220.62 ms.(3)Convolutional Neural Network(CNN)was used to predict falls.Pattern recognition using SVM requires feature extraction first.The accuracy of recognition depends largely on the extracted features.CNN can automatically extract features,eliminating the process of manually extracting features,and the extracted features are task-dependent and non-handcrafted,which have a stronger discernment.After pre-processing the sEMG signal,built CNN model,set up parameters,trained the CNN to obtain the classification model and the test data was detected and identified.Sliding time window technology was used to simulate the online situation.The sensitivity of the recognition result was 91%,the specificity was 96%,the accuracy was 94%,and the average lead time was 250.2ms.Comparing the two methods,the results showed that the evaluation indexes based on CNN method were higher than those based on SVM method,and the performance of CNN method was better.This study collected the main muscle sEMG information through design experiments,and studied the accurate and rapid fall prediction method using SVM and deep learning,which provided a feasible method for fall prevention.