Time-Frequency Representation and Convolutional Neural Network-Based Motor Imagery Recognition

View at publisher Summarise with VIRAT
Authors: Vikram Singh Kardam, Sachin Taran, Anukul Pandey
Cited by: 0
References: 50

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Motor imagery (MI)–based brain–computer interfaces rely on the analysis of electroencephalogram (EEG) signals for communication and neurorehabilitation. The non-stationary nature and low signal-to-noise ratio of EEG pose challenges for conventional feature extraction techniques such as STFT, CSP, and wavelets. This paper presents a compact and effective framework that employs a smoothed pseudo–Wigner–Ville distribution (SPWVD) to obtain high-resolution time–frequency representations, followed by deep convolutional neural networks for automated feature learning. Three well-known pretrained models—VGG16, ResNet50, and AlexNet—are evaluated along with a lightweight configurable CNN designed for EEG. Experiments performed on BCI Competition IV dataset 2b demonstrate that SPWVD-based representations significantly enhance separability of MI tasks. The configurable CNN achieves the highest accuracy of 93.00%, outperforming pretrained models while requiring fewer parameters, indicating strong potential for real-time MI-BCI applications.

Fields of Study & Subjects

EEG and Brain-Computer Interfaces Emotion and Mood Recognition Muscle activation and electromyography studies

Publication details

Journal / source
International Conference on Consumer Electronics
Pages
1–5
Publication date
3 February 2026
Language
EN