EMG features dataset for arm activity recognition

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human-Computer Interaction

Background:

• Hand gesture recognition is crucial for intuitive human-computer interaction (HCI).
• Electromyography (EMG) signals offer a promising, non-invasive method for capturing hand movements.
• Existing datasets may lack the specific features or diversity needed for robust algorithm development.

Purpose of the Study:

• To present a novel dataset for hand gesture recognition using surface electromyography (EMG) signals.
• To facilitate the development and benchmarking of machine learning algorithms for gesture classification.
• To provide a resource for research in prosthetic control and advanced HCI.

Main Methods:

• Collected surface EMG data from eight healthy subjects performing three distinct hand gestures (lifting, grabbing, flexing).
• Utilized the Delsys Trigno Wireless biofeedback system with four sensors on the dominant hand.
• Processed raw EMG signals, extracted seven time-domain features per channel, reduced dimensionality using Principal Component Analysis (PCA) to six components (95% variance).

Main Results:

• A comprehensive dataset of processed EMG signals and extracted features was generated.
• Principal Component Analysis (PCA) effectively reduced 28 features to 6 components, retaining 95% of the data variance.
• Machine learning models (Random Forest, Logistic Regression) were trained and tested for gesture classification.

Conclusions:

• The presented EMG hand gesture dataset is suitable for training and evaluating machine learning models.
• The dataset holds significant potential for advancing gesture recognition algorithms.
• Applications include improved prosthetic limb control and novel human-computer interaction interfaces.