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

With the further research of communication technology and artificial intelligence technology,human body recognition technology has a broader development prospect.Human motion recognition can obtain the user’s state and behavior for human-computer interaction,which is applicable to many industries.At present,since fine-grained signal information can be obtained in cheap Wi-Fi devices,finer motion recognition based on channel state information(CSI)has become a research hotspot.Aiming at the problem that the traditional motion recognition system is difficult to recognize similar actions,this paper carries out the research of motion recognition system based on improved support vector machine(SVM)and Wi-Fi CSI.The main contributions of this paper are as follows:(1)A multi-strategy adaptive improvement method of Differential Evolution(DE)algorithm based on parabolic function and historical evolutionary information is proposed.Four strategies are used to improve the algorithm,which make the scaling factor,crossover probability and mutation mechanism change adaptively,prevent the algorithm from falling into prematurity at the beginning of iteration,optimize the global search ability of the algorithm,thus improving the optimization performance of the algorithm.The improved algorithm is named Multi-Strategy Adaptive Differential Evolution Algorithm(MSADE).(2)A method to optimize SVM using MSADE algorithm(MSADE-SVM)is proposed.In the face of nonlinear problems in the application of examples,the performance of gaussian kernel-based SVM depends on the parameter selection of kernel parameters and error penalty factor.Penalty parameters and kernel parameters affect the recognition accuracy and learning ability of nonlinear SVM,and there is no reliable theoretical basis for the selection of these parameters.The commonly used grid cross validation search method often fails to get the best parameter combination.The parameter optimization ability of the MSADE algorithm proposed in this paper is superior to the traditional search method,which can realize the parameter combination optimization of support vector machine and then obtain the optimal training model.(3)Put forward a similar action recognition system based on improved support vector machine and Wi-Fi channel state information.First,the system uses CSI,which is measured by cheap Wi-Fi equipment,to obtain motion information.Next,the third-order Hermite-interpolation polynomial is interpolated to make the data sampled evenly in the time domain to ensure the integrity of the data.Then,Hampel filtering,wavelet threshold denoising and Butterworth low-pass filtering are used to remove the noise interference.Ande then,principal component analysis(PCA)is used to extract the motion features.Finally,MSADE-SVM is used to identify similar actions.In the off-line stage,through training the off-line sample data,the optimal training model is obtained.In the online phase,the set of CSI measurement values obtained is processed and then put into the optimal model for classification and recognition.Taking the standing biceps curl for example,the experimental results show that this system embodies the advantage of high-performance in similar action recognition and can be applied to many high-precision application scenarios in the future.