Human Vital Signs And Multi-Targets Detection Based On Convolutional Neural Network

In recent years,radar has been gradually applied to the detection of human targets and vital signs in daily life.At present,there are many kinds of radar which can be used for multi-target detection and human vital signs detection,such as MIMO radar,FMCW radar,CW radar and so on.These radars usually need large-scale antenna arrays.UWB radars have the advantages of low power consumption,high range resolution and non-contact detection of multi-target and human vital signs with only one pair of antennas.Therefore,this paper uses IR-UWB radar to extract multi-target and human life characteristics.In order to apply the implementation method to different detection platforms such as servers and mobile devices,convolution neural network and lightweight convolution neural network are used to recognize multi-target and vital signs respectively.1.PCA is used to extract the main motion features,eliminate the interference of motion harmonics,and obtain relatively pure texture features of moving objects.Combining with GoogLeNet,the number of moving objects in corresponding scenes is estimated.The estimation accuracy of 99.84%is obtained by using GoogLeNet method,and 71.6%is obtained by combining MobileNet method.2.Combining with GoogLeNet method,learning multi-target motion characteristics in different feature scenarios,eliminating interference from different environments,judging the number of targets in unknown environments,86.8%of the test accuracy is obtained,and 65.4%of the detection accuracy is obtained in MobileNet network,which solves the problem that the existing detection data depend on the environment and can not predict the number of unknown scenarios.3.In the detection of human vital signs,a differential cross-multiplication and state space method(DACM-SSM)based on IR-UWB radar is proposed.The DACM method is used to deal with the multi-value problem of the phase when arctan function is numerically calculated.The high signal-to-noise ratio(SNR)respiration and heartbeat signals are extracted by SSM algorithm based on threshold optimization.The algorithm improves the operation speed,restrains the influence of random displacement on human respiration and heartbeat extraction,and eliminates the interference of respiration harmonics on heartbeat extraction.It solves the problems of high computational complexity,large consumption of computational resources and low accuracy of real-time judgment results.Using VGG convolution neural network to judge whether the target's respiration and heartbeat are abnormal,more than 90%of the detection accuracy is obtained,and 71.1%of the detection accuracy is obtained by using MobileNet.By combining radar signal processing with in deep learning,the examiner can extract the physiological characteristics of the examinee for analysis,prediction and more comprehensive understanding according to the advantages of in-deep learning.