Detection Of Respiratory Signal And Heartbeat Signal And Human Behavior Recognition Based On Radar

Pulse doppler radar can be used in the study of vital signs detection and target behavior recognition.The detection of vital signs(mainly respiratory and heart signals)based on pulse radar has become one of the research focuses in the field of life detection in recent years.Accordingly,the traditional human behavior method requires manual extraction of features,and the classification recognition performance is affected by the selected features and the accuracy of extraction features,so the universality is not strong.Aiming at the above two points,the research work of this paper is carried out from the following two aspects.On the one hand,this paper studied the extraction and separation methods of vital signs(mainly respiratory and cardiac signals),and proposed an improved data length change technology based on wavelet transform to realize rapid heart rate detection.The most suitable parent wavelet is selected for the study of data length change technology through the wavelet analysis,and verified with the measured data.The interference of respiratory signal harmonic on heart rate detection is well suppressed,and the heart rate rapid detection is realized.On the other hand,this paper studies the human behavior recognition method based on convolutional neural network(CNN).As long as the original information such as spectrum diagram is input,high accuracy classification and recognition can be achieved,which can be easily applied to the classification and recognition of other targets.On the basis of studying the existing CNN classification recognition based only on time spectrum diagram,the target distance image information is added,and the human behavior recognition method based on the fusion of CNN slow time-distance image and time spectrum diagram is proposed.The experiment analyzed the classification and recognition performance of the four conditions of only input slow time-distance image,only input time spectrum diagram,input time spectrum diagram and slow time-distance image vertical stack and input time spectrum diagram and slow time-distance image horizontal stack.The experimental results show that the best classification accuracy can be obtained by the time spectrum diagram and the slow time-distance image.