Project Abstract

Emotion recognition (ER) has evolved to be a multi-disciplinary field of research, with the inclusion of computer scientists brining generalisability and reusability of machine learning (ML) models artifacts and results, automation of laborious legacy procedures like deep learning for feature extraction, and has seen increasing accuracy of classification i.e. k-means, LTSM. Current EEG ER research uses medical grade equipment that, whilst being successful in thematically related industry i.e. use in psychological and medical industry, is constrained in consumer industry deployment and thus the accompanying benefits; smart watches evidence the potential market capable of providing funding for further research. This is a result of the cost and thus inaccessibility, and unfitting form factor for consumers that medical grade imposes. Additionally, current research is wasteful and lacks generalisability; that is, in the literature, the majority of methodologies involve creating bespoke classifiers, and the reducing generalisability results from participant specific tuning of classifiers. In this paper, the viability of transfer learning (TL) for ER using a consumer-grade EEG headband is tested, and the performance of such compared with traditional ER classifiers i.e. k-means, warranted by TL?��s reusability and superior capability of deep learning to analyse time frequency spatial domain with less resources, unburdened by participant specific tailoring. The data used is from the Emognition Dataset acquired from Havard Dataverse; EEG from a consumer-grade Muse 2 headband. Various tools were used inclusive of Matlab, EEGlab, MNE library, and AlexNet Model. A python script read JSON EEG data then pre-processing using ICA, LPF, and epoching. Wavelet transforms are inputted to TL then tested against a traditional classifier. This method should increase generalisability and reusability, and prove as accurate as medical grade EEG headsets and more than the classical classification methods fostering more interest in wearable ER research and Inaugurating ideation for consumer market deployment.

Keywords: Emotion Recognition, Deep Learning, Wearables

 Conference Details

Session: Poster Session A at Poster Stand 67

Location: Sir Stanley Clarke Auditorium at Tuesday 7th 13:30 – 17:00

Markers: Alma Rahat, Hans Ren

Course: BSc Computer Science, 3rd Year

Future Plans: I’m looking for work