Research And Application Of Depression Recognition Method Based On Multimodal Sleep Physiological Signals

In recent years,the global incidence of depression has increased year by year,and the research on the recognition of depression has gradually become the focus of research in the fields of affective computing and clinical medicine.Various studies have shown that physiological signals such as EEG,ECG,and EMG can reflect a person's physiological and emotional state to varying degrees,which is helpful for clinically assisted diagnosis of depression.Physiological signals during sleep period are stable,not easily disturbed and easy to collect.Compared with physiological signals during waking period,it can effectively improve the accuracy of depression recognition.In addition,the depression recognition model based on a single modality has certain limitations and cannot meet the needs of clinical depression diagnosis.Therefore,based on the data of EEG and ECG signals during sleep,this paper conducted researches and analysis on depressed people and normal people,constructed a multimodal depression recognition model based on EEG and ECG signals during sleep,and designed and developed depression recognition system.The main research contents and results of this paper are shown as follows:1.According to the difference of EEG and ECG signals between depressed people and normal people,the paper collected the EEG and ECG data of 18 depressed patients and 18 normal people during sleep.Then,based on the collected data,the paper extracted and studied the characteristics of EEG and ECG signals during sleep.Finally,the feature matrices of the two physiological signals with the best distinguishing ability were selected by the method of feature selection,which laid a foundation for the construction of depression recognition model.2.Through the research and analysis of traditional feature layer fusion technology,this paper proposed a feature layer fusion strategy based on the improved weighting method,and compared this strategy with the traditional feature layer fusion strategy.The experimental results show that based on the improved weighting method Fusion strategy had better results.In addition,this paper explored the differences in experimental results of various classifiers on local physiological data sets,and finally selected support vector machine as the classifier of depression recognition model in this paper,and built a depression recognition model based on multimodal sleep physiological signals.3.Based on the depression recognition model constructed in this paper,a PC-side depression recognition system was designed and implemented.The system mainly includes data visualization module,information management and storage module and depression assessment module.It realizes the functions of data visualization,data preprocessing,feature extraction,feature storage,single-mode depression assessment and fusion assessment,which improves the value and practicability of the depression recognition model in the application.The paper explored a method for identifying depression based on physiological sleep signals.By analyzing and studying the difference between physiological sleep signals of depressed people and normal people,a set of depression recognition models based on physiological sleep signals was proposed,and based on this recognition model,developed a recognition model for depression,which improved the practicality of the depression recognition model.