Based On Wearable Heart Rate Monitoring And Single-Lead ECG Unbalance Data Emotion Recognition Research

Emotions have an important impact on people’s daily lives and work.Based on physiological signal monitoring and analysis,they can objectively assess individual emotional states and play an important role in health monitoring and human-computer interaction.Existing wearable devices already have the function of collecting heart rate and single-lead ECG signals.It is of great significance to explore emotion recognition based on wearable heart rate signal monitoring and single-lead ECG signals.At present,the following limitations exist in the research of emotion recognition based on heart rate and single-lead ECG signals:(1)Existing wearable devices and their algorithms only monitor the wearer’s heart rate data and make simple statistics,which cannot provide accurate emotion recognition;(2)Emotion recognition research based on single-lead ECG signals rarely considers the ability of instantaneous heart rate features extracted from ECG to express emotions,and the algorithm is more complex,without considering the limitations of single-lead wearable device computing capabilities;(3)Most of the current emotion recognition research has set prerequisites to make the data set in a relatively balanced state,less considering the imbalance of data samples caused by the difference in the number and duration of different emotions,which will limit the practicality of the algorithm.In response to the above problems,this article conducted the following research:1.A method for emotion recognition using heart rate data collected by wearable bracelet is proposed.Design an experimental paradigm of neutral + "target" emotions,and use the three stimulus materials of neutral,sad,and happy in the Chinese emotional image material library to establish a heart rate data set of 25 subjects.The target emotions are normalized using the neutral emotion correspondence data as the baseline,and the highest recognition rate of the four types exceeds 80%.This method verifies the effectiveness of the "neutral + target" experimental paradigm for identifying emotions based on heart rate,which can be applied to wearable consumer electronic devices for evaluating the wearer’s emotional state.2.In the emotion recognition based on single-lead ECG signals,an emotion recognition method based on wearable heart rate signals is introduced.While analyzing the heart ratevariability characteristics,the instantaneous heart rate characteristics are added together to serve as a physiological index that characterizes the changes in emotions.The RF-Light GBM sentiment classification algorithm framework is designed.In the two classification recognition and fusion recognition of positive and negative emotions,the recognition accuracy rate of happiness and sadness reaches 88.75%,and the accuracy rate of the five classifications is66.75%.Compared with related literature algorithms,the algorithm process proposed in this paper is simpler and the recognition accuracy is higher.3.A sentiment classification optimization algorithm based on unbalanced data is proposed.Aiming at the problem of data imbalance in the original ECG data set,the SMOTE+ ENN algorithm is used to sample the data to solve the problem of excessive distribution differences between the original data classes.The accuracy rate of happiness and sadness recognition is 93.3%,and the highest accuracy rate of five categories 88.08%.This solution can improve the accuracy of emotion recognition,and is closer to the phenomenon of individual emotion differences in actual application scenarios.The algorithm used in this paper comprehensively uses the characteristics of single-lead ECG and heart rate to describe emotions,and optimizes the unbalanced data in combination with the actual application scenarios,which provides a new idea for the realization of emotion recognition on the wearable device.