基于宽带雷达RCS数据的空间物体识别方法
Space object recognition method based on wideband radar RCS data
马腾 1周兰凤 1李建鑫1
作者信息
摘要
本研究提出了一种基于时频分析和混合神经网络的雷达RCS数据目标识别技术.通过小波变换、傅里叶逆变换和对数处理方法来处理2 种极化的宽带雷达散射截面(radar cross section,RCS)数据,实现数据的高效去噪和频域分析,从而得到更为精准和清晰的一维距离像.设计了一种混合神经网络结构来处理得到的一维距离像.该网络结构综合利用卷积神经网络(CNN)来高效提取特征,并采用长短时记忆网络(LSTM)来捕捉时序依赖关系,从而实现了对雷达RCS两种极化数据的高效识别.为验证该技术的有效性,使用某研究所提供的数据集进行了验证性实验,并与CNN、SVM、MLP等主流方法进行比较.通过参数的优化和调整,模型达到了 97.50%的识别准确率.结果表明,该方法能够充分利用时频信息,并成功整合局部和全局特征,为雷达RCS数据目标识别提供了一个高效和精准的解决方案.
Abstract
This study proposes a radar RCS data target recognition technology based on time-frequency analysis and hybrid neural network.Wavelet transform,inverse Fourier transform and logarithmic processing methods are used to process the two polarization broadband radar cross-section(RCS)data to achieve efficient denoising and frequency domain analysis of the data,thereby obtaining a more accurate and clearer Dimensional range profile.Furthermore,a hybrid neural network structure is designed to process the obtained one-dimensional range profile.The network structure comprehensively utilizes convolutional neural networks(CNN)to efficiently extract features,and uses long short-term memory networks(LSTM)to capture temporal dependencies,thereby achieving efficient identification of two polarization data of radar RCS.In order to verify the effectiveness of this technology,a confirmatory experiment was conducted using the data set provided by a certain research institute,and compared with mainstream methods such as CNN,SVM,and MLP.Through parameter optimization and adjustment,the model achieved a recognition accuracy of 97.50%.This result shows that this method can make full use of time-frequency information and successfully integrate local and global features,providing an efficient and accurate solution for radar RCS data target identification.
关键词
雷达目标识别/卷积神经网络/长短时记忆网络/傅里叶逆变换/小波变换Key words
radar target recognition/convolutional neural network/long short-term memory network/inverse fourier transform/wavelet transform引用本文复制引用
出版年
2024
NETL
NSTL
万方数据