Recognition Algorithms Of Hand Movements Based On Feature Dimensionality Reduction And Multiple Signal Fusion

Surface Electromyography(sEMG)is a kind of biological electrical signals with noninvasive collection.It can react to the state of human body.Although,the recognition algorithms of hand gesture on healthy subjects have reached a very high recognition rate,the accuracy of classification based on amputee subjects is still low.In order to improve the classification accuracy,we approach the recognition algorithms of hand movements based on feature dimensionality reduction and multiple signal fusion.This paper is supported by Science and technology development program of Jilin.“Mechanism,method and key technology of prosthesis control”(20150101191JC).To carry out the research on pattern recognition algorithm for multi-channel sEMG will promote the project electrical control on the development and industrialization of bionic limbs.So this article research will have a certain social significance.In this paper,research work mainly include following aspects:1)We analyzed the mechanism and features of sEMG,introduced the database we has been used.And we introduced the Clinical data of hand amputated subjects,the acquisition equipment,the placement of electrodes and the details of experiments etc.After preprocessing and filtering the data,we detected the activity period of extraction.2)Considering the real-time requirement,we employed the sliding window analysis method for accelerometer and sEMG signals.For each data block,we used time domain and time-frequency domain analysis to extract the feature.After selecting the best feature value,we put them into the four types of classifiers to recognize of the 17 kinds of hand movements.3)We conducted the locally linear embedding,nonlinear dimensionality reduction,laplacian eigenmaps and principal component analysis,respectively,to reduce the dimension of accelerometer signal.Not only we reduced the characteristic value of dimension,but also improved the classification accuracy.4)We mixed the characteristic values of the accelerometer signal and sEMG and put them into SVM classifier.Finally,the average accuracy rates of classification were up to 88.8±1.9% over 9 amputated subjects,which was an outstanding result.New evaluation index was employed to discuss the benefit of Accelerometers.In comparison with the previous literature,the results confirmed that our algorithm has good classification accuracy and strong robustness,which met the expected goals.