Research On Fine-grined Emotional Recognition Based On Brain Signals

EEG emotion recognition commonly uses Brain-Computer Interface(BCI)devices to collect physiological data and record the emotion data corresponding to the current state,but the variability of recording accuracy and the irreversibility of accuracy loss in data processing make it more difficult to apply the technology.Therefore,this project intends to explore the universality of EEG emotion recognition and realize its technical extensibility,and conduct an in-depth study on the problems of emotion analysis granularity,signal processing strategy,and data feature selection in EEG emotion recognition.To address the problem of coarse granularity in EEG emotion recognition,this project aims to propose a fine-grained Affective Computing(FGAC)model with universality through the research of emotional psychology.Firstly,by mining the public emotion expression data,extracting the high-coverage emotion categories in the life and work of the public in the current society;secondly,using Euclidean space theorem to construct the mapping rules of emotion to realize the spatial mapping of multidimensional emotion data;finally,carrying out label mapping for the quantitative data of emotion corresponding to EEG signals to obtain 20 types of fine-grained emotion labels.In view of the low signal-to-noise ratio of EEG signals,nonlinear non-smoothness,precision loss,low-band and single-channel recognition limitations,and signal overlap of similar emotions,this project proposes the Chronological Peaks Coding(CPC)model,which aims to reduce the noise data and at the same time,to explore the EEG signal from the microscopic perspective.The aim is to reduce the noise data and explore the local significant trend of each part of EEG signal from the microscopic perspective.Firstly,the EEG frequency range is decomposed into bands to obtain different EEG bands;secondly,the EEG amplitude characteristics under a single band are retained for all signal peaks under positive and negative phases;finally,the interval coding strategy is used to explore the local signal variation trends and patterns of EEG signals,which are further used to explore EEG features.For the feature selection problem,in view of the universal research using channel and feature fusion strategy mechanism,which requires high accuracy of hardware devices and is not universal.Therefore,this topic implements single-channel emotion recognition by constructing Partial Fluctuation Pattern(PFP)features.Firstly,the pattern recognition method is used to uncover the high-frequency cycle change pattern of the signal under different emotions;secondly,the multi-dimensional PFP features are constructed,including: fluctuation intensity,fluctuation number,fluctuation repetition rate,fluctuation perimeter rate,and fluctuation density;finally,support vector machine is used to verify the classification of fine-grained emotion recognition.The experimental results show that this project achieves an average recognition accuracy of 93%for 20-dimensional fine-grained emotion in a single channel and in each waveband.This study addresses the problems of coarse granularity of emotion recognition,low signal-to-noise ratio in low frequency bands,and multi-dependent channel fusion in EEG emotion recognition research.Firstly,the constructed FGAC model can realize the fine-grained mapping of multidimensional emotion data;secondly,the constructed CPC model can efficiently reduce the redundant noise while effectively preserving the significant periodic change characteristics of the signal;finally,the constructed PFP features can effectively realize the high precision emotion recognition.This topic explores the universality of EEG emotion recognition,reduces the reliance on hardware devices,refines the emotion recognition categories,and can be more conducive to applications in different fields.