Ranking hand movements for myoelectric pattern recognition considering forearm muscle structure

Abstract

Previous pattern recognition algorithms using surface electromyography (sEMG) have been developed for subsets of predefined hand movements without considering muscle structure. In order to decode hand movements, it is important to know which movements are appropriate for PR due to the different independence of movements between individuals and the high correlated characteristics of sEMG patterns between movements. This paper proposes a method to personally rank the order of hand movements from subsets (31 finger flexion, 31 finger extension, and 4 wrist movements in this paper). The movements were sorted into a ranked order with respect to the locations of the electrodes on the proximal forearm and the distal forearm. We evaluated the classification error as the number of desired movements (N _m) changed. The maximum N _m with an error lower than 10% was 20 for the proximal forearm and 10 for the distal forearm from ranked movements of individuals. Our method could help to identify the optimized order of hand movements considering the personal characteristics of each individual.

Appendix

Mathematical definitions of time-domain features which were used in this study are as follows [26]. x _i (k) is the kth signal sample, i is the ith window, N is the number of samples in the window, and xth is the threshold value.

1.1 Mean absolute value (MAV)

$${\text{MAV}}_{i} = \frac{1}{N}\sum\limits_{k = 1}^{N} {\left| {x_{i} \left( k \right)} \right|} ,$$

(2)

1.2 Waveform length (WL)

$${\text{WL}}_{i} = \sum\limits_{k = 1}^{N - 1} {\left( {\left| {x_{i} \left( k \right) - x_{i} \left( {k + 1} \right)} \right|} \right)} .$$

(3)

WL is a combined measure of waveform amplitude, frequency, and duration.

1.3 Zero crossing (ZC)

$${\text{ZC}}_{i} = \sum\limits_{k = 1}^{N} {f\left( k \right)} ,$$

(4)

where

$$f\left( x \right) = \left\{ {\begin{array}{*{20}l} {1,} \hfill & {{\text{if}}\;x_{i} \left( k \right) \times x_{i} \left( {k + 1} \right) < 0\;{\text{and}}\;\left| {x_{i} \left( k \right) - x_{i} \left( {k + 1} \right)} \right| > x{\text{th}}} \hfill \\ {0,} \hfill & {{\text{otherwise}}.} \hfill \\ \end{array} } \right.$$

ZC represents the number of points in the window where the sign of a function changes (e.g., from positive to negative). This feature is an estimate of the properties in the frequency domain.

1.4 Slope sign change (SSC)

$${\text{SSC}}_{i} = \sum\limits_{k = 2}^{N - 1} {f\left[ {\left( {x_{i} \left( k \right) - x_{i} \left( {k - 1} \right)} \right) \times \left( {x_{i} \left( k \right) - x_{i} \left( {k + 1} \right)} \right)} \right]} ,$$

(5)

where

$$f\left( x \right) = \left\{ {\begin{array}{*{20}l} {1,} \hfill & {{\text{if}}\;x > x{\text{th}}} \hfill \\ {0,} \hfill & {{\text{otherwise}}.} \hfill \\ \end{array} } \right.$$

This feature is similar to ZC regarding the frequency properties.