Bicycling Phase Recognition for Lower Limb Amputees Using Support Vector Machine Optimized by Particle Swarm...

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Machine Learning in Prosthetics

Background:

• Lower limb amputees face challenges in prosthetic control during activities like bicycling.
• Accurate phase recognition is crucial for effective prosthetic knee joint control.
• Existing methods may struggle with noisy sensor data from wearable devices.

Purpose of the Study:

• To develop a novel and accurate method for recognizing bicycling phases in lower limb amputees.
• To enhance prosthetic knee joint control for improved bicycling functionality.
• To optimize machine learning models for real-time prosthetic applications.

Main Methods:

• Utilized wireless wearable accelerometers and a knee joint angle sensor on the prosthesis.
• Implemented a soft-hard threshold filter to mitigate noise in sensor data.
• Extracted multi-dimensional feature vectors for training a Support Vector Machine (SVM).
• Optimized the SVM using Particle Swarm Optimization (PSO) for enhanced classification accuracy.

Main Results:

• The proposed PSO-SVM model achieved a 93% recognition accuracy on testing data.
• This accuracy significantly surpasses that of traditional Backpropagation (BP), SVM, and PSO-BP models.
• The filtering technique effectively reduced noise, enabling robust feature extraction.

Conclusions:

• The PSO-SVM method offers a highly accurate and effective solution for bicycling phase recognition in lower limb amputees.
• This approach holds significant potential for advancing prosthetic knee joint control systems.
• The integration of PSO optimization with SVM provides superior performance compared to other tested models.