Feature selection and dimensionality reduction: An extensive comparison in hand gesture classification by sEMG in eight channels armband approach

Analysis of Influence of Segmentation, Features, and Classification in sEMG Processing: A Case Study of Recognition of Brazilian Sign Language Alphabet

“…In total, 33 features were extracted and evaluated: 24 features from the time domain and nine from the frequency domain. These features were chosen due to their use in other feature sets in the literature [19,[53][54][55], which present their math expressions. Frequency features were extracted after the signals passed to a fast Fourier transform in each segment with 2000 points of resolution.…”

“…Some features needed parameterization and their values were based on studies that applied them for gesture recognition. For auto-regressive and cepstral coefficients, the fourth order was chosen [53,56];…”

“…Some features needed parameterization and their values were based on studies that applied them for gesture recognition. For auto-regressive and cepstral coefficients, the fourth order was chosen [53,56]; in the histogram feature, nine bins were extracted [57]. The second-moment was used in L-Scale [55]; the third order was chosen for V-order [53]; and a threshold value of 10 −2 was defined for the Myopulse percentage rate, Willison amplitude, sign slope change, and zero crossing features [58].…”

“…For auto-regressive and cepstral coefficients, the fourth order was chosen [53,56]; in the histogram feature, nine bins were extracted [57]. The second-moment was used in L-Scale [55]; the third order was chosen for V-order [53]; and a threshold value of 10 −2 was defined for the Myopulse percentage rate, Willison amplitude, sign slope change, and zero crossing features [58]. A 0.2 σ tolerance (standard deviation) and two dimensions were chosen for sample entropy [59]; the band of frequency was split at the average for the frequency ratio [53].…”

“…The second-moment was used in L-Scale [55]; the third order was chosen for V-order [53]; and a threshold value of 10 −2 was defined for the Myopulse percentage rate, Willison amplitude, sign slope change, and zero crossing features [58]. A 0.2 σ tolerance (standard deviation) and two dimensions were chosen for sample entropy [59]; the band of frequency was split at the average for the frequency ratio [53]. The performance of the features in the classification process was also compared with feature sets presented in the literature [54][55][56].…”