Through-Wall Radar Human Motion Recognition Based On Feature Enhancement And Shallow Neural Networks

The recognition of human motions based on through-wall radar has an important application value in national defense security,disaster rescue,patient monitoring and other fields,which has become a hot and difficult research spot in recent years.The signal of radar in through-wall environment is attenuated and distorted,which makes it more difficult for human motion recognition.This paper proposes a human motion recognition method based on feature enhancement and Convolutional Neural Networks(CNN).Firstly,the micro Doppler feature of human motion behind the wall is enhanced by adaptive threshold filtering and S-method,and then the CNN is used to classify the human motion data collected by the stepped-frequency through-wall radar system accurately.Besides,this paper utilize the time window segmentation and Bi-directional Long-Short Term Memory(Bi-LSTM)for human continuous motions recognition behind the wall.The main contents and work of this paper are as follows:(1)The existing methods of human motion recognition are summarized and the research progress of through-wall radar human motion recognition are expounded in detail.(2)The basic principle of human motion recognition based on stepped-frequency throughwall radar is introduced specifically.Firstly,the structure of stepped-frequency through-wall radar system,the characteristics of transmitted signal as well as the basic principles of Doppler and micro Doppler effects are introduced.Then,the preprocessing algorithms of radar echo data and commonly used time-frequency analysis methods are presented.Finally,the principles of human motion recognition and classification algorithms are recommended.(3)The method of human motion recognition of through-wall radar based on feature enhancement and CNN is proposed.Firstly,the noise in the time-frequency map is eliminated by adaptive threshold filtering.Then,the energy of the processed time-frequency map is gathered by S-method to enhance the time-frequency feature.Finally,the CNN is used for motion recognition.An experimental system of stepped-frequency through-wall radar is built,and six kinds of human discontinuous motions data and three kinds of human similar motions data behind the wall are collected.The experimental results show that the time-frequency features are effectively enhanced after adaptive threshold filtering and S-method processing.The average recognition accuracy of CNN for discontinuous and similar motions is 98.61% and97.22%,respectively.Compared with traditional Support Vector Machine(SVM)and KNearest Neighbor(KNN)methods,CNN has a higher recognition accuracy.(4)A recognition method of human continuous motion based on time window segmentation and Bi-LSTM network is proposed.Firstly,a small time window is used to segment the time-frequency map of human continuous motion into several frames.Then a BiLSTM network is built and trained.Finally,the five groups of human continuous motions data collected by the experiment are identified and classified.The results show that when the time window length is 10 s,the average recognition accuracy is the highest,reaching 66.81%.