Research On Emotion Recognition For Wearable Devices

Emotions are especially important in human daily life,and its fierce and fast response to the outside world can directly affect the psychological and physiological status of individuals.Due to the rapid development of computer science and the expansion of the desired market,the identification of emotional states has gradually become a research hotspot in today's society,and there is a very broad application prospect and floor scene in many fields.Because physiological signals are directly generated by the nervous system and control,the research on the emotional state in recent years has also gradually developed.At the same time,at the same time,physiological signals must have professional equipment acquisition and testing this requirement in spatial requirements and convenience limitations.With the gradual popularity of wearable equipment such as bracelets in recent years,portable,reliable and practical new features that can be attached to wearable devices are also increasing.In this context,this paper proposes an emotional identification scheme for a wearable device based on the corresponding physiological signal characteristics.First,from the different discussion and classification of mood at home and abroad,analyze the problems faced herein and determine the selected emotional classification model,consider the hardware conditions and physiological signals of existing wearable devices.Specific characteristics,the physiological signals of the participating emotion identification of this paper are determined,namely the skin electrical signal and the pulse signal.Second,focusing on the emotional induction scheme for self-bucing emotional databases,where data acquisition tools are existing wearable devices,emotionally induced materials from the Chinese emotional photo gallery and international emotional sound library.The material will stimulate the positive,negative,neutral and deviation of the subject.The experimental personnel operates software assist operations through the E-Prime psychology experiment,completing 20 volunteers' skin electrical signals and the data acquisition of pulse signals,and pre-propeling the resulting physiological signals to use wavelet transform or smooth filtering.Processing,standardization processing,and extraction of time domain and frequency domain features form a clean and effective self-built emotional database.Again,the selection of the classification recognition model is determined based on the small sample data characteristics of the emotional physiological signal,that is,the support vector machine is supported.A method of introducing a nuclear function is selected for the nonlinear characteristics of physiological signals,and a physiological signal based on a support vector machine emotional identification model based on a radial base Gaussian nuclear function is designed.The comparison of various parameter optimization methods is proposed to determine the importance of penalty factors and nuclear parameters in the model core function in the model core function.Finally,the application and effect analysis of the support vector machine emotional identification model is performed on the common DEAP emotional data set and self-built database of the academic community.Among them,the classification results from the DEAP data set can be found that the average recognition accuracy of multi-physiological signals is 70%,which is significantly better than 54% of single-skinned signals and 58% of the single pulse signal.The average identification accuracy of the average recognition model can reach 63% after applying the emotional identification model to the self-built database,which proves the feasibility of the model.At the same time,based on the emotional recognition model,the decision boundary function that can be applied to the wearable device is constructed and the identification effect is tested.The test effect is 64% accuracy,the effectiveness of the test effect can reach 59%,which proves that the wearable device is based on the feasible identification scheme based on the corresponding physiological signal characteristics.