基于稀疏表示的SAR目标识别方法研究

发布时间：2018-02-27 00:30

本文关键词： 合成孔径雷达自动目标识别特征提取稀疏表示稀疏保持投影　出处：《中国民航大学》2014年硕士论文　论文类型：学位论文

【摘要】：合成孔径雷达(Synthetic Aperture Radar,SAR)是一种主动的、全天候的远程传感器,可以在白天或者夜晚运行,能够穿透云层,从空中或者空间传播的平台构造地面高分辨率图像,在现代战场中有很重要的作用。基于SAR的自动目标识别(Automatic Target Recognition,ATR)技术在军事方面有很重要的应用,一直是国内外学者研究的热门课题。近几年压缩感知的提出为稀疏表示的发展提供了工程应用的土壤,得到很多学者的广泛关注,并被应用到多个领域,如图像压缩、去噪等。之后有学者将稀疏表示应用于图像识别中,并在人脸数据库上进行实验,实验结果证明该方法能够取得比传统方法更好的效果,基于此本文将稀疏表示方法应用于SAR图像目标识别中。首先介绍了稀疏表示的基本理论,包括稀疏字典的构造以及稀疏求解算法,在此基础上研究了稀疏理论在识别中的应用,给出了结合KPCA(Kernel Principal Component Analysis)和稀疏表示的SAR目标识别方法。该方法首先利用KPCA方法提取样本特征,然后在特征空间内构造稀疏表示模型,通过梯度投影法(Gradient Projection for Sparse Reconstruction,GPSR)求得测试样本的稀疏系数,最后根据稀疏系数的能量特征进行分类识别。利用美国运动与静止目标获取与识别(Moving and Stationary Target Acquisition and Recognition,MSTAR)实测SAR数据进行实验,实验结果表明该方法在方位角未知的情况下能够明显提高目标的识别结果,是一种有效的SAR目标识别方法。另外本文还研究了一种新的特征提取方法即稀疏保持投影(Sparsity Preserving Projections,SPP),该方法将由稀疏表示得到的稀疏系数引入到特征提取中,通过数据的稀疏重建关系构造目标函数得到特征向量。在该方法的基础上,本文给出了基于改进的稀疏保持投影特征提取方法,该方法在SPP特征提取的基础上保持样本之间的稀疏重构关系,同时借鉴了局部保持投影(Locality preserving Projection,LPP)特征提取方法的思想,使得提取的特征不仅能保持稀疏重构特性,还能使同类样本间的距离变小。将改进的特征提取方法应用到SAR目标识别中,利用MSTAR数据进行实验,实验结果证明了该方法的有效性。
[Abstract]:Synthetic Aperture radar (SAR) is an active, all-weather remote sensor that can operate during the day or at night, can penetrate clouds and construct high-resolution ground images from a platform in the air or in space. The automatic Target recognition (ATR) technology based on SAR has a very important application in the military field. In recent years, compression perception has provided the soil of engineering application for the development of sparse representation, which has been widely concerned by many scholars, and has been applied to many fields, such as image compression. After that, some scholars applied sparse representation to image recognition, and carried out experiments on face database. The experimental results show that this method can achieve better results than traditional methods. In this paper, sparse representation method is applied to SAR image target recognition. Firstly, the basic theory of sparse representation is introduced, including the construction of sparse dictionary and sparse solution algorithm. Based on this, the application of sparse theory in recognition is studied. In this paper, a method of SAR target recognition based on KPCA(Kernel Principal Component Analysis and sparse representation is presented. Firstly, the KPCA method is used to extract the sample features, and then a sparse representation model is constructed in the feature space. The sparse coefficients of test samples are obtained by gradient Projection for Sparse Reconfiguration (GPSRs). Finally, the sparse coefficients are classified and identified according to the energy characteristics of the sparse coefficients. The SAR data of moving and Stationary Target Acquisition and recognition are obtained and recognized by moving and still targets in the United States. The experimental results show that the method can improve the target recognition results obviously when the azimuth is unknown. This paper also studies a new feature extraction method, namely sparse preserving projection Preserving projects, which introduces sparse coefficients obtained from sparse representation into feature extraction. The feature vector is obtained by constructing the objective function through sparse reconstruction of data. Based on this method, an improved sparse preserving projection feature extraction method is presented in this paper. Based on SPP feature extraction, the sparse reconstruction relationship between samples is maintained, and the idea of local preserving projection preserving projection LPP feature extraction is used for reference, so that the extracted features can not only maintain sparse reconstruction characteristics. The improved feature extraction method is applied to SAR target recognition and the experimental results of MSTAR data show that the method is effective.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN958

【参考文献】