Hybrid attention-based deep learning network for emotion recognition by ECG signal

Emotions play an important role in our daily activities, decision-making, and artificial intelligence needs to identify emotions to interact constructively with its audience. In this paper, an intelligent method for two-dimensional emotion recognition is proposed. The ECG signal available in the DREAMER database has been used to recognize emotions because of the high correlation of this signal with emotions and easy recording. First step for valence and arousal recognition, the ECG signal is entered into the deep learning network, which is a combination of CNN and LSTM. CNN performs feature extraction and LSTM performs data classification. The attention mechanism aims to optimize the weights and improve the performance of the network, overseeing the proposed deep learning network. Using the proposed method, valence and emanation were identified with ۹۵% and ۹۴% accuracy, respectively. The proposed hybrid network is very suitable for high-dimensional data, and the use of the attention mechanism helps to improve the performance of the network by preventing overfit and getting stuck in local optimal.

کلیدواژه ها:

emotion ، CNN ، LSTM ، eCG ، Attention mechanism