Generative adversarial network-enhanced spatio-temporal graph convolution for driving fatigue monitoring in athletic training

Fatigue monitoring is essential for athletic training.This study addresses class imbalance and scarcity of severe fatigue samples by proposing a spatio-temporal graph convolutional network enhanced with a generative adversarial network.A conditional Wasserstein generative adversarial network generates realistic synthetic skeletal sequences to expand the training set.Combined with spatio-temporal feature extraction, our model achieves end-to-end fatigue level classification.Evaluations on the daily life activities dataset demonstrate superior performance, with accuracy, precision, recall, and F1-score reaching 92.5%, 91.8%, 92.2%, and 92.0%respectivelyoutperforming support vector machine by 20.2%, long short-term memory by 11.0%, baseline spatio-temporal graph convolutional network by 3.8%, and variational autoencoder-augmented models by 2.8% in accuracy.Ablation studies validate both the generative adversarial network augmentation and nonlinear labelling strategy, offering a reliable vision-based framework for fatigue monitoring.