Research On The Algorithm Of Human Body Posture Recognition Based On CNN

The use of radar to classify human activities has been applied in many ways in security,defense and rescue operations.Doppler modulation caused by human motion can identify specific human activities through radar signals.This modulation is often referred to as the "micro-Doppler" signal.However,for the traditional radar human motion recognition technology,there is a single data acquisition frequency band,and the time-frequency analysis method is not considered to avoid the polarization effect.Therefore,multi-band data acquisition and joint time-frequency of data are performed in the radar human echo.Analysis to avoid polarization effect is particularly important.In view of the cumbersome problem of traditional artificial extraction features,this paper uses convolutional neural network(CNN)to identify and classify radar echo spectra,thereby further improving recognition efficiency and accuracy.In view of the above problems,the main research work carried out in this paper is as follows:(1)For the differences in the movement patterns of various parts of the human body activities(bending,walking,crawling,sitting,walking and not swinging arm),the multi-scattering center model is used to model the human body and analyze the human body steps.State echo model and method for extracting echo characteristics of stepped variable frequency radar.(2)Aiming at the traditional single-band acquisition method of radar human echo signals,a step-by-step frequency radar is used for periodic continuous frequency sweep to obtain multi-band data.Since the radar echo signal caused by human activity is a non-stationary signal,the lightning echo signal is processed by three time-frequency analysis methods of short-time Fourier transform,Wigner-Ville distribution and Hilbert's yellow distribution,and converted into time.Spectral representation domain.(3)Convolutional neural networks are used for the cumbersome problem of the identification of time-spectrum maps.According to the principle of migration learning,the convolutional neural network structure for single time-frequency analysis and joint time-frequency analysis is built on the basis of Inception-v3 model.By comparing single time-frequency analysis and single time-frequency analysis and joint time in multi-band and multi-band The results of the frequency analysis method prove that the multi-band and joint time-frequency analysis method can effectively avoid the polarization effect,thereby improving the recognition efficiency and accuracy.