Classification of surface EMG signals for early signs of prolonged fatigue

Sports training are very important to athlete in maintaining and improving their performance. During training, adequate rest is essential to allow recuperation and build body strength. Inadequate rest may expose the body to prolonged fatigue (PF). This condition needs to be managed accordingly to avoid chronic fatigue syndrome. Currently, the non-invasive assessment in identifying PF are training log record, questionnaire and Borg scale. Recent findings indicate that there are strong characteristics on surface electromyography (EMG) under PF conditions such as glycogen breakdown, existence of lactate and soreness. This study extends the investigation of PF signs, especially on the inceptions of PF. An experiment has been conducted on twenty participants to investigate the behavior of surface EMG during five days of intensive training that was based on Bruce Protocol treadmill test. The intention was to induce PF on biceps femoris (BF), rectus femoris (RF), vastus lateralis (VL) and vastus medialis (VM). Besides surface EMG signals, physiological measurements were also collected from the participant. Physiological results demonstrate that the earliest PF signs developed were soreness, lethargy and performance decrement. For the surface EMG, they went through three main processes: de-noising, feature extraction and classification. De-noising technique through stationary wavelet transform (SWT) was employed in enhancing quality of surface EMG signals. During de-noising process, new method in estimating threshold (Th) was proposed. The method demonstrated to have higher performance in term on noise removal and accuracy in PF classification, compared to conventional Th methods. Nine features extracted from the de-noised surface EMG signals. There were changes in median frequency (ΔFmed), mean frequency (ΔFmean), mean absolute value (ΔMAV), root mean square (ΔRMS), and five features from wavelet indices (ΔWI). Daily fatigue mappings indicate that the emergence of PF can be traced based on extracted features. The mappings indicate that ΔFmed and ΔFmean tend to increase under PF condition for all four muscles BF, RF, VL and VM. Additionally, under PF condition, the mapping indicates an increase in ΔRMS and ΔMAV but decrease in ΔWI for RF muscle. In the classification stage, Naïve Bayes (NB) and Support Vector Machine (SVM) demonstrate accuracy with 98% and 97% respectively, in distinguish PF on RF, 94% and 96% respectively on BF, both 95% on VL, and 98% and 96% respectively on VM. Thus, this study successfully demonstrates that surface EMG can be used in identifying the inception of PF. The findings presented are significant in sports field to prevent higher degree of PF.

Preview Text FK 2018 32 - IR.pdf Download (437kB) | Preview

Actions (login required)

View Item