Search for a command to run...
Static stretching is a fundamental conditioning practice; however, objectively and quantitatively evaluating its physiological effects remains a significant challenge. Conventional surface electromyography analysis has predominantly relied on single-channel derived indices, such as Root Mean Square and Median Frequency. However, relying solely on single scalar metrics is insufficient to comprehensively capture local variations in muscle activity and alterations in complex neuromuscular control, suggesting that the establishment of more multifaceted evaluation methodologies is valuable. In this study, we propose a novel stretching evaluation method based on spatiotemporal pattern recognition using 16-channel surface electromyography and deep learning, specifically aimed at application in consumer electronic devices such as wearable devices for personal fitness management. Specifically, scalograms generated by Continuous Wavelet Transform and information regarding the sensor's spatial arrangement are input into a Convolutional Neural Network to construct a model that classifies the degree of stretching. Experimental results with five healthy adults demonstrated that the proposed model achieved high classification performance, with an F1-score of 88.2%. Furthermore, visualization analysis of the feature space using t-SNE confirmed that data from an intermediate stretching duration, which was excluded from training, was continuously mapped to the intermediate region between the two classes. This suggests that the proposed model learns not only discrete class classification but also the continuous transition dynamics—reflecting the depth of physiological change—of the muscle state associated with stretching intervention. This method holds potential for application as an objective biomarker for defining optimal stretching durations tailored to individual muscle characteristics for next-generation personal training assistants.