A new data augmentation convolutional neural network for human emotion recognition based on ECG signals. (May 2022)

• Record Type: Journal Article
• Authors: Nita, Sihem; Bitam, Salim; Heidet, Matthieu; Mellouk, Abdelhamid
• Is Part Of: Biomedical signal processing and control. Volume 75(2022)
• Publication Date: 2022-05
• Subjects: Human emotion detection -- Electrocardiogram (ECG) -- Heart Rate Variability (HRV) -- Data augmentation -- Convolutional Neural Network (CNN); 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.2022.103580
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Human emotion detection. Convolutional neural networks (CNN) for emotion recognition. Electrocardiogram (ECG). Heart Rate Variability (HRV) for emotion detection. Data augmentation for CNN. Abstract: Nowadays, human emotion recognition based on electrocardiogram (ECG) signal is considered as a hot topic applied in many sensitive domains such as healthcare, social security, and transportation systems. In the literature, various machine learning algorithms were proposed to this purpose however, the recognition accuracy of these techniques is hampered by the hardness of acquiring huge and balanced number of ECG dataset samples, which is considered as a major challenge in this topic. Therefore, we propose in this paper, a new data augmentation convolutional neural network (CNN) for human emotion recognition based on ECG signal. Specifically, we suggest to enrich the ECG dataset by a significant number of representative ECG samples, generated according to randomize, concatenate and resample realistic ECG episodes process. Hence, a new seven-layer CNN classifier is suggested, consisting of seven layers to detect human emotions in terms of valence, arousal, and dominance levels. Experiments that have been carried out using our proposal for Data Augmentation Convolutional Neural Network strategy on benchmark DREAMER database resulted in an accuracy rate of 95.16% to detect valence, 85.56% for arousal and 77.54% for dominance.