Research On Motion Recognition Technology Based On Improved Support Vector Machine And Channel State Information

At present,human state recognition is widely used in industry and academia.Electroencephalography(EEG)truly reflects the state of an individual and is produced by the central nervous system.The physiological state of the human body is also closely related to the central nervous system.Therefore,the identification of EEG signals has strong objectivity,and has gradually become a fatigue detection and emotion detection method.One of the most reliable biological signals in the identification field.Recognition of human body state based on EEG signals is of great significance for fields such as driving and high-altitude work,so EEG signal recognition of human body state has gradually become a research hotspot.The EEG signal is a non-stationary biological signal that is highly susceptible to interference.The original EEG signal collected by the sensor is mixed with a lot of noise,so the preprocessing of the EEG becomes extremely important.In the traditional human state recognition,many use machine learning or deep learning as the classifier,but the settings of the initial parameters of the classifier and other factors will affect the classification effect.In traditional research,multi-channel EEG signals are mainly studied,and multi-channel EEG acquisition equipment is not suitable for wearing.In order to obtain a more accurate human state recognition effect,the research contents of this paper are as follows:(1)This paper introduces a method for denoising EEG signals based on Wavelet Transform(WT),but this method cannot remove EEG artifacts.Therefore,this paper proposes a time-frequency denoising method,which removes the signal at the same time.High-frequency noise and cofrequency eye artifact in.In the fatigue state recognition,in view of the problem that the randomly set BP neural network initialization parameters will reduce the classification accuracy,this paper proposes to integrate the Genetic Algorithm(GA)and the BP neural network to construct a GA-BP classification model.The optimal neural network initialization parameters improve the recognition accuracy.(2)In emotional state recognition,this paper considers the powerful feature extraction ability of Convolutional Neural Networks(CNN)and its advantages in image recognition and classification,and proposes to classify the EEG spectrum of different states through CNN.Figure to realize emotion recognition.Considering that the classification of the Soft Max function in CNN will lead to problems such as insufficient generalization ability and over-fitting of image recognition,the support vector machine(SVM)is used to replace the Soft Max function,and a CNN-SVM classifier model is constructed to realize the recognition of emotions,the results show that the classification effect of the CNN-SVM model is better than that of the simple CNN classification.(3)In this paper,a portable fatigue recognition system is designed and used as a module in the smart helmet.The experimental results show that the system can accurately recognize fatigue.