Research On Detection And Recognition Of Life State Based On Biological Radar

Mechanical injuries to the human body caused by the collapse of heavy objects is a common type of injury after disasters such as earthquakes.Research has shown that injuries caused by heavy objects often lead to crush syndrome in injured individuals,which means that when heavy objects are removed,their injuries may suddenly worsen or even die.Therefore,it is crucial to understand the squeezed life status of the buried personnel in the early stages of search and rescue.Compared with traditional post disaster search and rescue equipment such as visible light or thermal imaging,bioradar can use electromagnetic waves to penetrate non-metallic obstacles to detect human vital signs.Therefore,if biological radar can be used to sense the radar life signals of injured individuals after obstacles from the echoes during the search for missing persons,and further identify the corresponding squeezing state,it can scientifically guide subsequent rescue and effectively reduce casualties.This article focuses on the above issues.Firstly,the working principle of pulse ultra wideband bio-radar and the original radar echo data are analyzed.In order to address the static background interference and electromagnetic attenuation phenomena in the detection environment,research on the preprocessing methods of bio-radar echoes is carried out;In order to achieve radial ranging of the detected target while accurately obtaining the corresponding human radar life signal at the corresponding position,research on target distance calculation methods has been carried out;And based on the traditional human radar life signal model and the physiological characteristics of post disaster casualties,an improved radar life signal mode is proposed.Due to the sensitivity of life signals detected by biological radar to environmental noise,this article attempts to use multiple methods to suppress noise.Variational mode decomposition,as a completely non recursive signal processing algorithm,combined with permutation entropy algorithm,can quickly filter out noise,but unreasonable parameter configuration can also directly affect the performance of VMD algorithm.Therefore,this paper proposes a particle swarm optimization method,which uses the maximum cross correlation as the fitness function to iteratively optimize the parameters of VMD.Performance verification of the proposed method based on an improved radar life signal model.The results show that compared with traditional IIR digital filtering,EMD,and CEEMD algorithms,this method improves the signal-to-noise ratio by 7.5dB,6.4dB,and 5.2dB,respectively.In addition,for 600 second long data segments,this method reduces processing time by 86.3%compared to the CEEMDAN algorithm.In the field of non-contact state recognition,in order to complete the training of the recognition network,the radar life signal sequence of the wounded is usually directly input as sample data.However,this may cause other features in the data to be ignored,resulting in the network model not being able to fully learn the data feature information and further affecting the accuracy of recognition.Therefore,this article proposes a crushed casualty state recognition network model based on radar life signal time-frequency images.The network takes the denoised radar life signal time-frequency images of the wounded as input,enhances the ability to extract data feature information through multi-scale convolution structures,and combines residual networks to suppress deep network degradation while avoiding information loss and redundancy.To verify the performance of the network model,this paper designed a human radar life signal acquisition experiment that simulates post disaster squeezing scenarios.The model was trained and tested based on more than 3000 sets of data collected from 10 subjects in different squeezing states(including trapped postures and compression types).The results showed that the recognition accuracy of the network for the squeezing state of the human body after obstacles reached 92.63%,Compared to the SVM model based on respiratory features and the 1D-CNN model based on radar life signals,it has increased by 4.19%and 6.01%respectively.In summary,this article designs and implements a software for detecting and recognizing the life status of squeezed patients based on the hardware equipment of bio-radar,radar echo processing algorithms,and a network for recognizing the squeezing status of patients.By combining multiple scene experiments,it has been proven that this software can effectively extract the vital sign information of patients contained in the radar echo signal,thereby improving the accuracy of recognizing the squeezing status of patients.This helps to early detect injured individuals who are suffering from mechanical injury and compression syndrome,and assists rescue personnel in adjusting rescue strategies in a timely manner to reduce casualties.