Classification of EMG during walking using principal component analysis and learning vector quantization for biometrics study

Abstract

Electromyography (EMG) signal classification for biometrics is a new field in biomedical engineering. EMG is an electrical activity that occurs in the muscle layer during active motion. Since the way people walking is defined by the structure of individual muscles and bones, we hypothesized that the way of walking is unique and must be able to be used in biometrcis data. In this study, we classified the EMG data of 8 lower limb muscles during normal walking test (Rectus Femoris, Vastus Medialis, Vastus Lateralis, Bicep Femoris, Semitendinosus, Gastrocnemius Medialis, Gastrocnemius Lateralis and Tibialis Anterior). Six healthy volunteer were involving in this study by walking in gaitlab with 8 EMG electrodes attached on their muscles. Each volunteer performed 3 walking trial, so in total 18 EMG datasets were analized for classification. Principal Component Analysis was used to extract the features of EMG data of all 8 muscles during walking. Learning Vector Quantization (LVQ) was used to classify the EMG data based on subject. Training and testing method in LVQ networks used the Leave-One-Out Cross Validation (LOOCV) method. The accuracy of the system in classifying the EMG data based on subject is 88.8%. In conclusion, EMG data during walking of 8 lower limb muscles was quiet unique to be implemented in biometrics application.