Design of Ensemble Stacked Auto-Encoder for Classification of Horse Gaits with MEMS Inertial Sensor Technology

Area of Science:

• Biomechanics
• Machine Learning
• Sports Science

Background:

• Accurate classification of horse gaits is crucial for effective equestrian training and self-coaching.
• Existing methods may lack the precision required for detailed gait analysis.

Purpose of the Study:

• To develop and evaluate an ensemble stacked auto-encoder (ESAE) model for classifying horse gaits.
• To leverage wavelet packets from horse rider motion data for enhanced gait recognition.
• To provide a robust system for equestrian self-coaching.

Main Methods:

• Utilized an ensemble stacked auto-encoder (ESAE) model trained with feedforward, backpropagation, and gradient descent.
• Extracted features using wavelet packets from motion data captured by 16 inertial sensors worn by an expert rider.
• Employed cross-entropy and mean squared error for performance evaluation, alongside statistical values and ensemble modeling.

Main Results:

• The proposed ESAE model achieved high performance in classifying horse gaits.
• The method demonstrated superior accuracy compared to conventional algorithms, including support vector machines (SVM).
• Wavelet packet analysis combined with ensemble learning significantly improved classification efficacy.

Conclusions:

• The developed ESAE model offers a promising approach for accurate horse gait classification in equestrian self-coaching.
• The integration of wavelet packets and ensemble methods enhances the robustness and performance of gait analysis.
• This technology has the potential to revolutionize equestrian training by providing objective feedback.