EEG-Activity-Recognition

TSAE-Net: Temporal-Spatial Activity Encoder for EEG-based Psychological State Recognition This repository contains the official PyTorch implementation of the paper: Development of Deep Learning Models to Assess Psychological States from EEG During Physical ActivitiesYihang Deng, Yilin Ren, Qiao Changze📅 June 2025 🔍 Overview This project introduces TSAE-Net, a novel deep learning framework for accurately recognizing psychological states from EEG signals during physical activities. Traditional methods suffer from poor temporal alignment, noise sensitivity, and limited multimodal integration. TSAE-Net addresses these issues through two major components: Temporal-Spatial Activity Encoder (TSAE)Captures temporal and spatial dependencies using convolutional layers, attention mechanisms, and dynamic temporal pooling. Dynamic Activity Fusion Strategy (DAFS)Enhances robustness with multimodal fusion, temporal alignment via DTW, and contextual refinement of overlapping signals. 📎 A live demo and documentation can be found in the TSAE-Net repository. 🧠 Key Features End-to-end deep learning pipeline for EEG-based psychological state classification. Support for multiple physiological datasets: Sleep-EDF, CWL EEG/fMRI, Physionet MI, MODA. Adaptive Modality Fusion using attention mechanisms. Real-time capable Temporal Realignment of Asynchronous Signals (TRAS). State-of-the-art performance across multiple bio-signal datasets. 🧪 Datasets Dataset Description Sleep-EDF EEG-based sleep stage classification CWL EEG/fMRI Multimodal fusion of EEG and fMRI signals Physionet MI Motor imagery EEG for BCI applications MODA Multi-objective cognitive signal decoding