Research On Microwave Detection And Recognition Methods For Imaginary Arm Movement Brain Activity Signals

In the brain-computer interface system,how to accurately identify the movement intention in the subject’s brain is of great research significance.With the development of artificial intelligence technology,the application of the BCI system of motor imagination has become a reality,and it has been widely used in big data medicine,virtual reality,sports injury rehabilitation and stroke rehabilitation treatment.With the production and application of nano-scale microelectronics manufacturing and packaging technology,the non-contact collection device has gradually become portable and compact,which is very conducive to the development of portable intelligent medical devices,and is also a major scenario application of the Internet of Things.However,a series of problems that need to be solved,such as the low signal-to-noise ratio,weak resolution of the signal collected by non-contact method,the need to penetrate the whole brain due to the acquisition medium,and the weak localization of brain function,that is,the problem of decoding the intention of fine motion imagination,has been puzzling relevant scientific researchers.In this paper,the microwave antenna array in non-contact mode is used for detection,and the microwave transmission signal(MT)is used to represent the brain activity information of motor imagination.First of all,after obtaining the original MT signal data,use short-time Fourier transform(STFT)and select the preprocessed data in the mu frequency band and perform the integrated empirical mode decomposition(EEMD).Select the specific IMF after the EEMD to extract the multi-scale sampling entropy(MSSE).Then the data is optimized by using local linear embedding(LLE)and isometric feature mapping(ISOMAP)feature selection and dimensionality reduction.The datasets after feature engineering processing is put into shallow machine learning classifiers such as decision tree(DT),logical regression(LR),support vector machine(SVM),K nearest neighbor(KNN)and random forest(RF)for pattern recognition and evaluation,The highest accuracy rate of LLE-SVM model is 96.97%,and the average accuracy rate is 90.85%,which shows that it is feasible and effective to use multi-scale sampling entropy to represent the "left hand" and "right hand" two modes of motion imagination.Then,in order to further recognize more motion intentions in motion imagination,this article first proposes a new deep learning model called ALNN,which is a new neural network model combined with ABSTLAY and LSTM.High accuracy rate is obtained in fine motion imagination three mode identification,with the average accuracy rate as high as 62.14%.The fine motion imagination three mode is decoded well,but the accuracy rate in four mode identification is low.In order to achieve performance improvement in four-mode recognition,an attention mechanism based on key-value pairs has been added to ALNN,and the new model is ALANN.The final experimental results show that the ALANN has a maximum accuracy of89.55% and a maximum average accuracy of 79.52% in three classification recognition,which is a significant improvement compared to the ALNN model.At the same time,in the four mode recognition of fine motion imagination,the accuracy is as high as 57.69% and the maximum average accuracy is 52.34%,which can effectively identify the subdivision modes of fine motion imagination.In this thesis,some new theories and methods are proposed to decode the motion imagination intention of non-contact brain-computer equipment,and have certain effects,which can provide reference value for the development of future non-contact brain information acquisition BCI system.