Research On Moving Target Classification And Recognition Technology Based On Radar Micro-doppler Signal

The detection and classification of targets have various applications including the monitoring of species in territorial boundaries,the monitoring of patients' activities in medical treatment,the confirmation of unknown person in security protection work,etc.,which has become the focus of governments,companies and scholars around the world.In recent years,radar has gradually become the focus of the field of human detection and classification.Compared with other sensors,radar's advantages in human detection and classification mainly reflected in: it is not affected by light and weather,so it can monitor people all day.However,the technology is still in the initial stage,the theory is not perfect.This paper studies the identification of specie,the discrimination of human activities and identification of human targets,which has important theoretical significance and application value.The main work and results are as follows:(1)The basic model of a moving human body has been studied,aiming at simulating the radar echo of human body,and verifying the possibility of using radar to realize identification theoretically.The non-rigid human body is simplified into a simple rigid body model composed of upper limbs,lower limbs and torso.By adjusting the length of this parts the echo of different human model can be obtained.Then,traditional machine learning algorithm and deep convolutional neural network are used to identify these radar echoes to realize identity recognition theoretically.(2)The recognition of different species based on radar micro-doppler signal has been studied.Due to the distinction between the structures of different objects,the micro-doppler signal generated by the movement will also be different.Therefore,different species can be recognized by using the radar micro-doppler signal,and then the human signal can be extracted for in-depth study.We use K-band radar to collect data which is processed by Short-Time Fourier Transform.Followed by the selection and extraction of the corresponding features,and then classied by traditional machine learning algorithm and deep learning algorithm.In this paper,a species identification method based on radar micro-Doppler signal using machine learning algorithm is proposed.(3)Activities classification and human identification based on radar micro-doppler signal have been studied.For the same person,the swing frequency and amplitude of the limbs are different in different motion states,resulting the difference between the micro-Doppler characteristics of different activities.Therefore,micro-Doppler characteristics can be used to classify different activities.Meanwhile,when different people in the same state of movement,due to the difference of weight and heignt,micro-Doppler characteristic generated by movement will be distinct,which can be used to realize the identification of human identity.In this paper,the K-band radar were used to collect data,and the Short-Time Fourier Transform are used to process the signal.Followed by the selection and the extraction of the micro-Doppler feature which is used as the input of the traditional machine learning algorithm.When using the deep convolutional netural network,the time-frequency diagrams are used as the input of the convolution of the neural network for the classification and recognition.In this paper,methods about classifying human activities and human identification based on radar micro-Doppler signal using machine learning algorithm are proposed.To sum up,this paper has completed the human body modeling,the recognition of different species,the classification of activities and the identification of human based on radar micro-doppler signal.Also,the performance of traditional machine learning,the fusion algrithms of several traditional machine learning methods and deep convolution neural network has been compared.The results show that in these studies,the recognition accuracy of deep convolutional neural networks is generally higher than that of traditional learning algorithms and their fusion algorithms,together with better anti-noise performance.For identity recognition,when the number of people is less than a certain number,the accuracy rate of deep learning can be comparable with traditional human recognition methods such as methods based on fingerprint and video,which have important military value and social significance.