Unsupervised Machine Learning For Unmanned Aerial Vehicles And Birds Of Micro-doppler Feature Enhancement Technology

In the process of Radar detection of UAV and birds targets,the Radar cross-section of the target is small,and it is difficult to detect by using the traditional Constant false-alarm Rate method.Due to the obvious fretting characteristics of UAV and birds targets,it is possible to detect and identify targets by using the target micro Doppler characteristics.However,the resolution of micro-Doppler characteristic frequency extracted by typical time-frequency analysis techniques is relatively low.In addition,due to the influence of noise,the available echo data is limited,the feature distortion is serious,and the micro-Doppler continuity features obtained are not obvious.To solve the problems of low resolution of time-frequency map of micro-Doppler features and not obvious relevant features,this thesis starts from the original radar echo data and combines unsupervised machine learning algorithm to enhance the micro-Doppler features of UAV and birds targets.Firstly,the micro-motion models of UAV and bird targets are introduced,the radar echo signal model is established,and the principle of short-time Fourier transform and the reason of low resolution are analyzed.At the same time,an appropriate dictionary is constructed in the time-frequency domain to express the micro-Doppler features after time-frequency analysis in the form of matrix for the convenience of later operation and processing.Secondly,the sparse feature enhancement of micro-Doppler signals in the time-frequency domain is studied.This thesis adopts ₁ norm regularization idea to model the sparse problem,and solves it under the framework of Alternating Direction Method of Multipliers.The sparsity enhancement of time-frequency maps of UAV and birds micro-Doppler characteristics in different scenarios is realized through simulation,which verifies the effectiveness of the algorithm.Thirdly,the continuity features enhancement of micro-Doppler signals in time-frequency domain is studied.The Generalized Gaussian Distribution is used to model the prior Distribution adaptively.Proximal-Unadjusted Langevin Algorithm was proposed to solve the posterior distribution of target features.The continuous characteristics of the UAV and birds under different scenarios are enhanced through simulation,which verifies the high efficiency of the method.Finally,study the dynamic evaluation method of performance index of unsupervised Bayesian machine learning algorithm in the detection and identification process of UAV and birds.The positioning accuracy and detection probability are taken as examples to evaluate the dynamic performance indicators,and compared with the traditional Monte Carlo method.At the same time,the interference subject to different distribution is introduced to verify the generality of the proposed algorithm,and the processing gain of related indexes is introduced to quantitatively verify the efficiency of the proposed method.