Enhancing squat movement classification performance with a gated long-short term memory with transformer network...

Area of Science:

• Biomechanics
• Sports Science
• Machine Learning

Background:

• Bodyweight squats are fundamental sports training exercises.
• Accurate classification of aberrant squat movements is crucial for safe and effective training.
• Existing methods may lack the precision needed for real-time feedback.

Purpose of the Study:

• To present a novel gated long-short term memory with transformer network (GLTN) model for classifying bodyweight squat movements.
• To evaluate the performance of the GLTN model in distinguishing between acceptable and aberrant squat forms.
• To offer a feasible wearable solution for monitoring squat technique in sports training.

Main Methods:

• Developed a novel gated long-short term memory with transformer network (GLTN) model.
• Collected data from 22 healthy young male participants performing 9 squat variations (1 acceptable, 8 aberrant).
• Utilized four customized inertial measurement units (IMUs) placed on the thorax, waist, right thigh, and right shank at a 200 Hz sampling rate.

Main Results:

• The GLTN model achieved high classification performance: 96.34% accuracy, 96.31% precision, 96.45% recall, and 96.32% F-score.
• Outperformed state-of-the-art deep learning models in squat movement classification.
• Demonstrated the model's effectiveness in identifying aberrant squat movements.

Conclusions:

• The proposed GLTN model offers a feasible wearable solution for monitoring aberrant squat movements.
• This technology can facilitate performance enhancement and injury risk assessment during sports training.
• Coaches and practitioners should consider individual needs and goals when implementing this model.