Feature evaluation and pattern recognition of lower limb muscle EMG during postural balance control

We recognized EMG signal patterns of lower limb muscles by using neural networks and performed feature evaluation during the recovery of postural balance of human body. Surface electrodes were attached to lower limb and EMG signals were collected during the balance recovery process from a perturbation without permitting compensatory stepping. A waist pulling system was used to apply transient perturbations in five horizontal directions. The EMG signals of fifty repetitions of five motions were analyzed for ten subjects. Twenty features were extracted from EMG signals of one event. Feature evaluation was also performed by using DB (Davies-Bouldin) index. By using neural networks, EMG signals were classified into five categories, such as forward perturbation, backward perturbation, lateral perturbation and two oblique perturbations. As results, motions were recognized with mean success rates of 75 percent. With the neural networks classifier of this study, the EMG patterns of lower limb muscles during the recovery of postural balance can be classified with high accuracy of recognition.