摘要
针对传统恒虚警率(constant false-alarm rate,CFAR)方法难以探测鸟类目标的问题,提出一种基于时频(time-frequency,TF)域鸟类目标微多普勒贝叶斯增强算法。首先,以鸟类目标扑翼模型为基础,建立雷达回波信号及微多普勒模型。其次,考虑短时傅里叶变换(short-time Fourier transform,STFT),对回波信号进行时频分析。针对STFT加窗操作影响分辨率及其对杂波敏感的问题,引入广义高斯分布对先验自适应建模,在贝叶斯推理方式下实现时频域微多普勒特征增强。考虑到目标非多普勒特征非平滑,后验分布计算困难,提出用近端未调整朗之万算法(proximal unadjusted Langevin algorithm,P-ULA)进行高效求解。仿真及实测实验数据表明,所提算法不仅能够有效抑制背景噪声,而且可以在一定程度上保留微多普勒特征的连续性。
Aiming at the problem that it is difficult to detect bird target using the traditional constant false-alarm rate(CFAR)method,a Bayesian enhancement algorithm of the micro-Doppler feature of bird target based on the time-frequency(TF)domain is proposed.Firstly,based on the flapping model of bird target,the corresponding radar echo signal and micro-Doppler model are established orderly.Secondly,the short-time Fourier transform(STFT)is considered to analyze the echo signal in TF domain.In view of the problem that resolution is influenced by windowing process of STFT and STFT is sensitive to clutter,a generalized Gaussian distribution(GGD)is introduced to model the prior in an adaptive way,and the micro-Doppler feature in TF domain is enhanced in a Bayesian inference manner.Considering the difficulty of calculating the non-smooth posterior distribution related to the micro-Doppler feature of the target,the proximal unadjusted Langevin algorithm(P-ULA)is proposed to solve it efficiently.Experimental results of simulated and raw data show that the proposed can not only effectively suppress the background noise,but also preserve the continuity of micro-Doppler feature to some extent.
作者
杨磊
孙卫天
毛欣瑶
夏亚波
桑婧隺
YANG Lei;SUN Weitian;MAO Xinyao;XIA Yabo;SANG Jinghe(College of Electronic Information and Automation,Civil Aviation University of China,Tianjin 300300,China;Tianjin Port Yuanhang International Ore Terminal Co.,Ltd.,Tianjin 300452,China)
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2024年第2期505-516,共12页
Systems Engineering and Electronics
基金
国家自然科学基金(62271487)资助课题。
关键词
雷达鸟类目标探测
微多普勒特征
短时傅里叶变换
广义高斯分布
贝叶斯学习
radar detection for bird target
micro-Doppler feature
short-time Fourier transform
generalized Gaussian distribution(GGD)
Bayesian learning
