Improving sEMG signals recognition accuracy using 3D compression and broad learning system

Lei Zhang, Hui Zhao, Sha Qi, Huarui Bian, Yunhang Wu, Shaobo Gu, Manfredo Atzori, Henning Müller, Yabin Cao

Abstract: Recognition of surface electromyogram (sEMG) signals plays a vital role in prosthetic control and human-machine interaction. However, achieving high recognition accuracy for forearm movements is challenging due to the inherent volatility of sEMG signals. This paper proposes an innovative approach that leverages 3D-color-grayscale conversion and the Broad Learning System (BLS) for classifying 52 movements. This approach involves compressing the data through 3D views into grayscale images and applying the BLS method to reduce classification time while improving accuracy (achieving 96.2% accuracy based on NinaPro DB1). Among them, the use of 3D-color-grayscale conversion can convert 3D sEMG signals into a format that is easier to process and store through multi-level data conversion, thereby achieving efficient encoding in experiments; using BLS can improve computational efficiency by mapping input data for feature extraction, generating enhancement nodes from these features, and connecting all features and enhancement nodes directly to the output layer to form the network structure. This novel approach can potentially assist patients with conditions such as myasthenia gravis and hemiplegia in effectively controlling exoskeletons for rehabilitation training.

External IDs:dblp:journals/sivp/ZhangZQBWGAMC25