A decision-making mechanism based on EMG signals and adaptive neural fuzzy inference system (ANFIS) for hand gesture prediction

Artificial intelligence (AI)-based technologies assist users in applying the intended action when upper extremity movement cannot be fully provided. Electromyography (EMG), a depiction of muscle activity, offers various advantages when employed with AI-based systems like virtual reality applications and prosthetics controls. In this paper, a fuzzy logic (FL)-based decision-making mechanism is presented in order to provide effective control and improve the prediction performance of the stated systems. In this regard, EMG signals were collected from 30 participants when imitating different seven hand gestures. After the necessary preprocessing and segmentation processes, the Empirical Mode Decomposition (EMD) method which is the first stage of the Hilbert-Huang Transform (HHT) was applied and Intrinsic Mode Functions (IMF) were obtained. High-resolution time-frequency (TF) images were obtained by applying HHT to the IMFs determined by a statistical selection method. Various distinctive features were extracted from the visualized TF images based on the joint representation of the time and frequency domain. The Adaptive Neuro-Fuzzy Inference System (ANFIS) was then fed these features, which used two alternative clustering approaches. For seven hand gesture classifications, the average accuracy scores for the Subtractive Clustering (SC) and Fuzzy C-mean (FCM) clustering methods were obtained as 93.88% and 92.10%, respectively. The proposed feature extraction method based on TF representation combined with FL techniques yielded encouraging results for the classification of nonstationary and nonlinear biological signals such as EMG.