Research On Emotion Recognition Algorithm Based On Near Infrared Spectroscopy

In the past century,mental illness has seriously jeopardized human health,and more and more researchers have begun to work on the diagnosis and prevention of mental illness.Emotion is one of the outward features of mental state,emotion recognition is a product of the development of machine learning technology.With the development of intelligent systems that sensing emotions,more and more special equipment can be used for the prediction and diagnosis of mental illness,providing technical support for early diagnosis and treatment of mental illness.Functional nearinfrared spectroscopy(fNIRS),as a non-invasive physiological signal,is widely used due to its strong anti-interference ability,safety and convenience.In this paper,the emotion recognition algorithm based on near-infrared signal is studied.The specific research contents are as follows:(1)Considering the low time resolution of near-infrared signal,an emotional evoked experiment based on video stimulation was designed,and the fNIRS acquisition instrument was used to collect the near-infrared signals induced by the emotions in recent years.(2)The collected signal contains noise such as the physiological activity signal and the power frequency interference of the instrument and denoising is required.Most of the noise can be removed by filtering with bandpass filtering,but some noise and useful brain information frequency aliasing need to be processed by the denoising algorithm.In order to remove part of the noise that is aliased with the useful brain information,this paper proposes a WPD-ICA filtering method,which combines the wavelet packet decomposition with the ICA algorithm to complete the signal processing.The WPD-ICA algorithm is compared with the EEMD-ICA method,and the results show that the WPD-ICA algorithm has higher SNR and can solve the rate aliasing problem.(3)Three types of classifiers,KNN,Naive Bayes and Support Vector Machine are studied.According to the technical characteristics of fNIRS,three kinds of linear features and three kinds of nonlinear features are extracted as the input of the classifier.The classification accuracy of three classifiers is discussed.The parameter optimization method of support vector machine classifier is discussed and the grid optimization algorithm and particle swarm optimization algorithm are compared.the results show that the classification accuracy is higher when particle swarm optimization is used.(4)Combining the brain topographic map and the 3D activation state map of the subjects,the brain activation state of the task state was studied.In this paper,the related work of emotion recognition algorithm based on fNIRS is completed and some achievements have been achieved.