Smooth Multiscale Convolutional Attention Transformer Network for Continuous Motion Estimation of Hand Knuckle Angle using Surface EMG Signals

Abstract

As a key technology in the field of human-computer interaction and intelligent perception, continuous hand motion estimation has made significant progress in recent years driven by deep learning. Its core goal is to accurately capture the dynamic trajectory of hand joints through the modeling of high-dimensional time-series data, so as to realize natural and low-latency human-machine cooperative operation. Although current deep learning algorithms have achieved more satisfactory prediction results in the field of continuous motion estimation, they still need to be improved in terms of prediction accuracy, compatible action diversity and prediction robustness.In order to solve the problems of few compatible actions and poor robustness of existing deep learning models, based on the research in the previous part, this part proposes a smooth multiscale convolutional attention Transformer network that improves the generalization and the number of compatible actions for continuous motion estimation. The model consists of a serial connection between a multiscale convolutional channel attention network and an improved Transformer network, which integrates the advantages of the two network architectures and is capable of extracting features in multiple scale dimensions of the sEMG, in addition to the addition of a smoothing algorithm to further improve the accuracy and noise immunity of the model. This section focuses on the design of the experimental scheme and the analysis of the results of the smoothed multiscale convolutional attention Transformer network. Firstly, the components of the network are introduced: the multiscale convolutional attention module, the improved Transformer network module, and the smoothing algorithm module. The proposed network is then evaluated in the Ninapro dataset and compared with several existing state-of-the-art algorithms to demonstrate the effectiveness of the proposed method.