Skip to content
Cai Jing, Yuan Shouguo, Li Rui, Xu Menghui
1. Jilin University, School of Instrument Science and Electrical Engineering, Changchun, Jilin 130061.
Abstract:
Emotions are closely related to human behavior, family, and society. Emotions not only reflect various feelings, thoughts, and behaviors of humans but also represent psychological and physiological responses generated by various external stimuli. Therefore, accurate recognition of emotions is crucial in many fields. Changes in emotions can lead to changes in electroencephalogram (EEG) signals, and conversely, these changes also reflect emotional states. Based on the DEAP database, this study extracts time-domain and frequency-domain features from EEG signals and uses PCA (Principal Component Analysis) for dimensionality reduction. A weighted KNN algorithm is employed for 5-fold cross-validation training, achieving an accuracy rate of 80% for recognizing four emotional states: excited, relaxed, depressed, and angry.
Emotions are a feedback mechanism to objective phenomena and a form of emotional expression, thus emotion recognition is widely applied in fields such as artificial intelligence, psychology, affective computing, computer vision, and medical treatment. Physiological signals are generated by the activity of the autonomic nervous system in the body, which cannot be controlled by human will or disguised, and can objectively reflect the physiological and psychological activity states of the body, making them a reliable basis for accurately judging emotional states. With the development of science and technology, significant research achievements have been made in emotion recognition based on physiological signals (EEG, ECG, pulse, respiration, skin temperature, EMG, skin conductance). Literature indicates that EEG signals, which are most closely related to brain activity, can most authentically reflect human emotional states.
In recent years, emotion recognition based on EEG signals has become a hot topic in the fields of emotion research and human-computer interaction. Pane et al. proposed a strategy combining emotional lateralization and holistic learning, achieving a classification accuracy of 75.6% using random forests on the DEAP dataset. Verma et al. conducted multimodal emotion recognition based on the DEAP database using Support Vector Machines (SVM). Kolodyazhniy analyzed data from 34 participants using the K-nearest neighbors algorithm and cross-validation, achieving a maximum accuracy of 73.2% for recognizing fear, sadness, and neutral emotional states. However, these methods recognize few types of emotions and have low accuracy rates. In this paper, we propose a method using the weighted KNN algorithm and 5-fold cross-validation based on the DEAP database to accurately recognize the four emotions: excited, relaxed, depressed, and angry, achieving an accuracy rate of up to 80%.
Source: Journal of Electronic Technology Application, September Issue
Click belowto read the original article and download the PDF of the paper
