A Multi-Class Muscle Fatigue Recognition Model for Athletes Using KPCA Dimensionality Reduction and SVM Classification on Multi-Channel sEMG Signals

Abstract

Muscle fatigue is an inevitable phenomenon in athletic training, and its accurate assessment is crucial for preventing injuries. To address the issues of high redundancy in multi-channel surface electromyography (sEMG) features and the limitations of linear dimensionality reduction methods, this study proposes a muscle fatigue recognition model based on Kernel Principal Component Analysis (KPCA) and Support Vector Machine (SVM). sEMG signals were collected from 12 subjects performing sustained contractions. A comprehensive set of 48 time-domain, frequency-domain, and nonlinear features was extracted. KPCA was employed for nonlinear dimensionality reduction before classification. The resulting features were fed into an SVM classifier to distinguish between three fatigue states: relaxed, transitional fatigue, and fatigued. The model was evaluated using 10-fold cross-validation. Results demonstrated that the KPCA-SVM combination achieved the highest performance, with an average recognition accuracy of 91.5%, precision of 0.91, recall of 0.93, and F1-score of 0.92, outperforming other combinations of dimensionality reduction methods (MI, PCA) and classifiers (FLDA, KNN). This method provides an effective tool for the objective assessment of muscle fatigue in athletes.