Hand Gesture Recognition Using sEMG, Time-Frequency Analysis and Artificial Intelligence

This study addresses the challenge of improving prosthetic control through the analysis of electromyographic (sEMG) signals and gesture recognition. Using the Ninapro database [1] containing EMG signals from 10 subjects performing 17 different movements, focusing on recognizing 12 specific gestures. Each gesture was repeated six times per subject, resulting in a robust dataset for analysis. The methodology employed a combination of statistical techniques, arithmetic operations and Fourier transformations to preprocess and analyze the EMG signals. Various machine learning (ML) models were evaluated, with the RandomForest classifier achieving the highest accuracy at 20.8%, while other classifiers like KNeighbors, GaussianNB, SVM(Support Vector Machines), and MLP(Multi-Layer Perceptrons) showed lower performance. To improve the results, Principal Component Analysis (PCA) was applied, but it only resulted in marginal improvements. Given the limitations of traditional ML methods, the EMG signals were converted into grayscale images and applied Fast Fourier Transform (FFT) for frequency domain analysis. A convolutional neural network (CNN) was trained on these images, achieving a high accuracy of 95% in gesture recognition. The CNN's performance underscores the potential of deep learning (DL) techniques in capturing complex patterns in sEMG signals, leading to significant improvements in real-time prosthetic control. This work contributes the foundation for future AI integration in biomedical applications, particularly in assistive technologies. Overall, this study highlights the effectiveness of artificial intelligence in assistive engineering through detailed exploration of sEMG signal processing and gesture recognition.