Negative Emotion Recognition Based On Physiological Signals

Affective computing is the key technologies to achieve high-level human-computer interaction. Emotion recognition is an integral part of affective computing, it related to cognitive science, psychology, physiology, behavior science, sensor technology, computer science and so on, has a broad application prospect. In this paper, the emotion recognition theory is applied to the evaluation of negative emotion, and proposed a negative emotion recognition research based on physiological signals.First, this paper described the shortage of two classical affective emotional database. On that base, proposed a negative emotion recognition research experiment based on physiological signals. In this experiment, the subjects are16of graduate students in grade one to grade three of Biomedical engineering in Yanshan University (8male and8female). They were respectively collected ECG and EMG under the stimulation of the noise and light music.This experiment collect a total of32sets of data, including16sets of negative emotion data as the datasource, and the16sets of no-emotion data as reference.This experiment obtain a large amount of ECG and EMG signals, but not all of them contains emotional information,so we need to use feature extraction to select effective feature and remove redundant and ineffective feature. For Continuous wavelet transform is utilized to extract10feature (R-R interval, heart rate, PQS wavelength, QRS wavelength etc) from ECG,and extract14feature from EMG.For the24feature vectors which extracted from ECG and EMG signals, this paper used support vector machine (SVM) method to recognize the classes of the feature vectors. For support vector machine using radial basis function (RBF) kernel, this paper proposed two SVM method to find the optimal value of the trade-off parameter C and the kernel parameter σ:cross-validation method and genetic algorithm. In the classification results of ten times, the average correct classification rate of SVM based on genetic algorithm is increased by0.43%in the training set than the cross-validation method, and increased by4.45%in the test set. In addition, the average correct classification rate of SVM classifiers under these two optimization methods are all greater than81%. So the proposed method displayed better classification performance. The result verifies the feasibility and effectivity of the designed method.