A data augmentation and channel selection technique for grading human emotions on DEAP dataset. (January 2023)

• Record Type: Journal Article
• Authors: Singh, Uttam; Shaw, Rabi; Patra, Bidyut Kr.
• Is Part Of: Biomedical signal processing and control. Volume 79(2023)Part 1
• Publication Date: 2023-01
• Subjects: Deep learning -- Electroencephalography (EEG) -- Empirical Mode Decomposition (EMD) -- Genetic Algorithm (GA) -- Grey Wolf Optimization (GWO) -- Long Short Term Memory (LSTM); Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104060
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Emotion Recognition is one of the most important research area in the domain of Brain Computer Interactions. Human activities are influenced by emotions. Emotion recognition is carried out using gesture recognition, facial expressions etc. These methods are inconvenient and require quick feedback from users. Recently, Electroencephalogram technology has been found to be very efficient for emotion recognition task. Multi-Channel EEG headset is found to be an effective technology for BCI. However, it generates huge channel data and data obtained from many channels do not play effective role in identification of emotional state. In this paper, we use publicly available DEAP dataset as a source of EEG signals. Two major issues regrading the EEG data analysis is being addressed in this research. The first issue is the availability of the small number of samples. To address this issue, we exploit signal processing techniques. Multi-channel EEG with a large sampling frequency produces huge data per channel. All the channels are not important for emotion analysis and hence to address this second issue we explore metaheuristic algorithms to obtain an optimal subset of channels for emotion classification. Obtained results are very promising with 92.5% and 81.25% for two class classification in valence and arousal emotions and our proposed work can be used for practical applications of emotion classification. Highlights: Data augmentation to create more samples of data using signal processing methods. Explore statistical and frequency modelling-based methods to reduce channel data. Exploit a Metaheuristic algorithm to solve the problem of channel selection. To classify the emotional intensities using the LSTM network.