Motion Intention Recognition Based on Deep Learning Approaches for EEG Signal Analysis Considering Laban/Bartenieff Movement System

Brain-Computer Interface (BCI) based on Motor Imagery (MI) is a key technology for achieving fine-grained motor intention recognition. However, existing studies lack a systematic movement classification paradigm, resulting in insufficient decoding accuracy for complex intentions. To address this issue, this paper introduces the 'Effort' framework of the Laban/Bartenieff Movement System (LBMS), focusing on the 'Light-Strong' classification, and categorizes catching and releasing actions into four types: Light Catching (LC), Light Releasing (LR), Strong Catching (SC), and Strong Releasing (SR). EEG signals were collected from subjects via a non-invasive device through controlled experiments, with light and strong action data groups acquired. After preprocessing (including denoising, band-pass filtering, and data augmentation), Deep Learning (DL) methods were adopted for classification: Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) networks were constructed to verify the discriminability of 'Light-Strong' related EEG features; a hybrid CNN-BiLSTM-Attention model was proposed to realize multi-class action recognition, presenting a novel feature description paradigm for motor intention recognition.