Skip to main content

Table 2 Major memory usage comparison between the new retraining method and the previous retraining method

From: A real-time, practical sensor fault-tolerant module for robust EMG pattern recognition

 

New fast retraining

Previous retraining

AB subjects (772 windows for each class, 7 classes, 6 channels, 4 features per channel)

\( {\tilde{\mu}}_g \):(6 × 4) × 4 bytes = 96 bytes;

Total size of the feature matrix:

\( \tilde{\varSigma} \) : (6 × 4) × (6 × 4) × 4 bytes = 2304 bytes;

(6 4) × 772 × 7 × 4 bytes = 518784 bytes

Total: 96 × 7 + 2304 = 2976 bytes = 2.9 Kbytes

= 506.6 Kbytes

TR1 subject (772 windows for each class, 5 classes, 6 channels, 4 features per channel)

\( {\tilde{\mu}}_g \):(6 × 4) × 4 bytes = 96 bytes;

Total size of the feature matrix:

\( \tilde{\varSigma} \) : (6 × 4) × (6 × 4) × 4 bytes =2304 bytes;

(6 × 4) × 772 × 5 × 4 bytes = 370560 bytes

Total: 96 × 5 + 2304 = 2784 bytes = 2.7 Kbytes

= 361.9 Kbytes