Improved Classification Accuracy of Hand Movements Using Softmax Classifier and Kalman Filter.

Area of Science:

• Biomedical Engineering
• Rehabilitation Engineering
• Signal Processing

Background:

• Effective control of myoelectric lower arm prostheses relies on accurate interpretation of surface Electromyography (sEMG) signals.
• Current methods face challenges in precisely identifying intended hand movements from noisy sEMG data.
• Advancements in machine learning and signal processing are crucial for improving prosthesis functionality.

Purpose of the Study:

• To enhance the accuracy of hand gesture classification from sEMG data for myoelectric lower arm prosthesis control.
• To evaluate the efficacy of combining feature arrays, a Kalman filter (KF), and a Softmax classifier for improved classification.
• To validate the proposed method using a standard dataset and multiple participants.

Main Methods:

• Utilized the BioPatRec database containing sEMG data from 17 participants performing ten distinct hand movements.
• Implemented a classification system incorporating feature arrays and a Softmax classifier.
• Applied a Kalman filter (KF) to pre-process and smooth the training sEMG data before classification.

Main Results:

• The Softmax classifier achieved an initial accuracy of 95.3% without Kalman filter pre-processing.
• Incorporating the Kalman filter (KF) to smooth the training data resulted in a significant accuracy improvement to 99.3%.
• The proposed method demonstrated high reliability in classifying ten different hand movements.

Conclusions:

• The integration of feature arrays, Kalman filtering, and a Softmax classifier offers a highly accurate method for sEMG-based hand gesture recognition.
• Kalman filter pre-processing is a key factor in achieving near-perfect classification accuracy for myoelectric control.
• This approach holds significant potential for advancing the performance and usability of lower arm prostheses.