RETRACTED ARTICLE: Innovative use of optical sensors for real-time imaging and diagnosis using enhanced verge denoising and LSTM for sports medicine

• Published in: Optical and quantum electronics Vol. 56; no. 4
• Main Authors: Zhang, Xiaodan, xie, Dongwei, Gao, Xiang
• Format: Journal Article
• Language: English
• Published: New York Springer US 18.02.2024
• Subjects: Characterization and Evaluation of Materials; Computer Communication Networks; Electrical Engineering; Lasers; Optical Devices; Optics; Photonics; Physics; Physics and Astronomy; Wavelet packet verge denoising; Optical sensors; Wearable devices; Muscle fatigue recognition; Sports medicine
• DOI: 10.1007/s11082-023-06094-9
• ISSN: 1572-817X

In this article, researchers describe a novel method for identifying muscular tiredness in human subjects by combining surface electromyography (sur-EMG) signals with long short-term memory (LSTM) networks. To improve the quality of the sur-EMG data, the experiment begins with the shaving of unwanted leg hair and wiping the skin with medicinal alcohol. Sur-EMG sensors are placed on specific leg muscles, and the participants perform an incremental cycling protocol. The core contribution of this research is an enhanced wavelet packet verge denoising algorithm that effectively filters noise from sur-EMG signals. This algorithm uses a continuous and differentiable verge function to optimize denoising. A Shannon entropy-based method determines the optimal number of decomposition layers, ensuring the highest denoising efficacy. Feature extraction techniques, such as root mean square (RMS), integrated electromyogram (IEMG), median frequency (MF), and mean power frequency (MPF), are employed to capture the temporal and frequency domain characteristics of the sur-EMG signals. A muscle fatigue recognition model is built using long short-term memory (LSTM) networks. The sur-EMG signals are sorted into fatigued and non-fatigued categories using this model. The performance of the LSTM model is compared to support vector machine (SVM) and convolutional neural network (CNN) models. The accuracy of the proposed strategy is shown to be superior by the evaluation results compared to other models, achieving an accuracy rate of 95.27%. The study emphasizes the importance of feature selection and dataset choice in developing robust muscle fatigue recognition models. These findings have promising implications for the field of muscle fatigue assessment and sports science.