Time Frequency Analysis and Deep Feature Extraction for Hand Gesture Classification

Hand gesture categorization is a crucial field of study for hand gesture-based systems that provide significant advantages in human-computer interaction (HCI). HCI-based systems are coupled with surface electromyography (sEMG) data source to obtain extremely clear and accurate movement information. This paper proposes a deep learning-based method based on sEMG signals for classifying human hand gestures. The sEMG signals are displayed in the form of time–frequency images by applying continuous wavelet transform. Time frequency images are considered as input to convolutional neural network. The pre trained model such as VGG16, AlexNet, and Resnet-18 are utilized for deep feature extraction. And classification of hand gestures using sEMG signal is done with support vector machine classifier (SVM). The experimental results are compared to different findings from the literature and improved classification accuracy of 92.4% for 7 gestures and 93.8% for 6 gestures is achieved.