Study on the methods of feature extraction based on electromyographic signal classification.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Rehabilitation Technology

Background:

• Electromyography (EMG) signals are crucial for understanding muscle activity, particularly in advanced applications such as bionic hand control.
• EMG signal processing involves acquisition, pre-processing, feature extraction, and classification, with feature extraction being critical due to signal complexity.
• Selecting the most representative signal channels is essential for efficient and accurate EMG analysis.

Purpose of the Study:

• To develop and evaluate a novel feature extraction method for identifying the most representative two-channel EMG signals from an eight-channel acquisition.
• To compare the efficacy of the proposed method against traditional techniques like Principal Component Analysis (PCA) and Support Vector Machine (SVM) feature elimination.
• To validate the performance of the selected EMG channels using K-Nearest Neighbor (KNN), Random Forest (RF), and SVM classification algorithms.

Main Methods:

• Proposed a novel correlation heat map method for feature extraction to select two key EMG channels from eight.
• Employed traditional methods including Principal Component Analysis (PCA) and Support Vector Machine (SVM) feature elimination for comparison.
• Utilized K-Nearest Neighbor (KNN), Random Forest (RF), and Support Vector Machine (SVM) classifiers to evaluate the extracted features.

Main Results:

• The proposed correlation heat map method demonstrated superior performance in selecting representative EMG channels.
• Classification accuracy achieved using the proposed method surpassed that of the traditional PCA and SVM feature elimination techniques.
• The study confirmed the effectiveness of the selected two-channel signals across multiple classification algorithms.

Conclusions:

• The correlation heat map offers an effective approach for feature extraction in multi-channel EMG data.
• This method improves the efficiency and accuracy of EMG signal analysis, particularly for complex applications like prosthetic control.
• The findings suggest that reducing channel dimensionality using the proposed method can lead to enhanced classification performance.