Leveraging graph neural networks and gate recurrent units for accurate and transparent prediction of baseball pitching speed

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: March 5, 2025

Long short-term memory (LSTM) networks are widely used in biomechanical data analysis but have the significant limitations interpretability and decision transparency. Combining graph neural (GNN) with gate recurrent units (GRU) may offer a better solution. This study proposes validates hybrid GNN-GRU model for predicting baseball pitching speed enhancing its using layer-wise relevance propagation (LRP). C3D from 53 athletes were downloaded public dataset. Kinematic features of 9 joints during process calculated Visual3D, resulting total 208 valid pitches. The feature input into both LSTM models, hyperparameters tuned particle swarm optimization. LRP was employed to obtain contribution rate changes kinematic prediction results throughout cycle. accuracy models evaluated mean absolute error (MAE), squared (MSE), R-squared (R2). showed that there statistical differences MAE R2 metrics between on test set. (P = 0.000, Z - 5.170, Cohen's d 1.514) 2.981, 2.314) significantly lower than those model. Compared LSTM, achieved potentially more susceptible influence variability. Moreover, GNN-GRU-based demonstrated

Language: Английский

---

New Framework for Human Activity Recognition for Wearable Gait Rehabilitation Systems

Applied System Innovation, Journal Year: 2025, Volume and Issue: 8(2), P. 53 - 53

Published: April 15, 2025

This paper presents a novel Human Activity Recognition (HAR) framework using wearable sensors, specifically targeting applications in gait rehabilitation and assistive robots. The new methodology includes the usage of an open-source dataset. dataset surface electromyography (sEMG) inertial measurement units (IMUs) signals for lower limb 22 healthy subjects. Several activities daily living (ADLs) were included, such as walking, stairs up/down ramp walking. A signal conditioning, denoising, filtering, feature extraction activity classification is proposed. After testing several conditioning approaches, Wavelet transform (WT), Principal Component Analysis (PCA) Empirical Mode Decomposition (EMD), autocepstrum analysis (ACA)-based approach chosen. Such complex effective enables supervised classifiers like K-nearest neighbor (KNN), neural networks (NN) random forest (RF). classifier has shown best results with accuracy 97.63% EMG extracted from soleus muscle. Additionally, RF IMU 98.52%. These emphasize potential HAR systems rehabilitation, paving way real-time implementation devices.

Language: Английский

---

Image encoding and wearable sensors-based locomotion mode recognition using convolutional recurrent neural networks

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 100, P. 107068 - 107068

Published: Oct. 19, 2024

Language: Английский

---

A Data-Driven Approach to Estimate Human Center of Mass State During Perturbed Locomotion Using Simulated Wearable Sensors

Annals of Biomedical Engineering, Journal Year: 2024, Volume and Issue: 52(8), P. 2013 - 2023

Published: April 1, 2024

Language: Английский

---

Estimation of electrical muscle activity during gait using inertial measurement units with convolution attention neural network and small-scale dataset

Journal of Biomechanics, Journal Year: 2024, Volume and Issue: 167, P. 112093 - 112093

Published: April 1, 2024

Language: Английский

---