Motion Artifacts Removal And Multi-Task Recognition Of FNIRS Signal

With the continuous development of artificial intelligence,brain-computer interface(BCI)has emerged as an emerging technology in neuroscience.It can bypass neuromuscular pathways when transmitting brain information and realize direct communication between the brain and external devices.Military,aerospace,daily life,clinical treatment and other fields have achieved rapid development,and show huge application potential and development prospects.At present,the pattern recognition research in the BCI system mainly focuses on identifying two mental tasks of the same type,and there are development bottlenecks such as fewer classification categories,low classification accuracy,and poor real-time performance.Functional near-infrared spectroscopy(fNIRS)is an emerging neuroimaging technology,which has developed rapidly due to its large spatial resolution,small device size,low cost,and wearability.This article starts with the preprocessing,feature extraction,feature selection and classification of fNIRS-BCI signal,and aims to optimize the multi-task recognition model to obtain higher classification accuracy and information transmission rate.The specific research content is as follows:(1)In the preprocessing stage,the step-by-step fusion algorithm Spline-Polynomial fitting-SSA(SPS)and its improved algorithm TDDR-Polynomial fitting-SSA(TPS)are proposed to effectively remove the motion artifacts in the fNIRS signal: baseline shifts,slow drifts,spikes.Because the spline interpolation can effectively remove the baseline shifts,the polynomial fitting has a good effect on the slow drifts removal,and the singular spectrum analysis(SSA)has excellent performance in removing spikes,the three algorithms are merged to propose the SPS algorithm.However,the SPS algorithm needs to manually adjust the parameters when removing the baseline shifts.Therefore,the TPS algorithm is proposed to improve the degree of automation of the algorithm,that is,the TDDR algorithm is used to remove the baseline shifts to realize the improvement of the SPS algorithm.The experimental results on simulation data and real data show that the performance of the algorithm in this paper is better than that of the comparison algorithm in removing motion artifacts,and the TPS algorithm has the best effect,reducing the impact of motion artifacts on classification performance.(2)On the basis of TPS preprocessing,a multi-task recognition scheme based on the fusion of linear and non-linear features of motor imagery(MI),mental arithmetic(MA),and idle state(IS)is proposed.In the feature extraction stage,by studying the impact of two time window lengths on the classification results,the optimal time window of 5～20 s is determined.In the feature selection stage,the performance of the multi-task fNIRS-BCI system in this paper is further improved by applying the principal component analysis(PCA)algorithm;In the classification stage,by comparing the performance of different classifiers,it is found that SVM can obtain higher accuracy and faster information transmission rate than LDA,which provides a new idea for enriching the functions of the BCI system.