Fig. 3

Effect of wrist position information on linear and non-linear classification of 4 hand motions in 6 non-amputees and 2 partial-hand amputees. Each classifier was trained and tested using data from 13 wrist positions. Each classifier was trained with either EMG features alone (TDAR) or EMG features combined with wrist position features (TDAR + POS). Results are shown for (a) non-amputees, (b) Partial-hand subject 1 and (c) Partial-hand subject 2. The percent change in classification error when wrist position data was combined with EMG features is shown in figures. D-F. A positive change represents an improvement in performance with the addition of wrist position features. Results are shown for (d) non-amputees, (e) Partial-hand subject 1 and (f) Partial-hand subject 2. Error bars represent standard errors. LDA: Linear discriminant analysis; QDA: Quadratic discriminant analysis; LNN: Neural Network with linear activation function; MLPANN: Neural Network with non-linear activation functions