Enhanced EMG‐Based Hand Gesture Recognition by Using Generalized Deep Infomax Networks

Researcher interest in developing methods to acquire and interpret biological signals is rapidly growing. This interest spans a wide range of applications, with a particular focus on prosthetic control and rehabilitation, are the complexity and variability of surface electromyography (sEMG) data make precise hand gesture recognition (HGR) from sEMG signals essential. This study integrates the advantages of the generalized linear model (GLM) with a deep infomax (DIM) structure in a deep‐learning model, proposing the generalized linear DIM (GLDIM) classifier. While mutual information maximization has not previously been applied to signal inputs, DIM primarily depends on the topology of convolutional neural networks (CNNs). In this study, we adapt DIM concepts to the frequency domain, which is regarded as a more general structure than CNNs. To address challenges related to potential sEMG signal loss, we introduce a variable alpha‐trimmed mean (v‐ATM) filter for preprocessing. Furthermore, a feature vector is created by extracting Mel‐frequency cepstral coefficient (MFCC) features from the preprocessed signal. The proposed GLDIM classifier is suitable for real‐time applications in gesture‐controlled systems, as shown by experimental results that demonstrate its ability to achieve 99.27% accuracy for the DualMyo dataset and 99.018% accuracy for the Ninapro DB5 dataset.