基于稀疏表示的SAR目标识别算法研究
发布时间：2018-08-22 07:42
【摘要】：合成孔径雷达(Synthetic Aperture Radar,SAR)是一种具有高分辨力的成像雷达。由于全天时、全天候的成像优势,它在民用和军用领域被广泛应用。作为应用之一的SAR目标识别由于对国防预警的重要意义成为广大学者研究的热点之一。稀疏表示从过完备字典中选取尽量少的原子线性重构信号,应用于识别问题时,不仅有天然的识别信息包含在稀疏表示系数中还表现出优良的抗噪性,而SAR目标识别的一个难题就是斑点噪声,因此以稀疏表示理论为基础的SAR目标识别具有广阔的研究前景。本文以稀疏表示为基础,结合SAR图像的特点,在SAR图像预处理、特征提取和目标识别方面展开研究,主要的研究内容如下:1.结合MSTAR数据库的SAR目标图像特点,研究了基于支持向量机(Support Vector Machine,SVM)分类思想的SAR图像预处理方法,经过对数变换、基于SVM的稳定SAR图像分割、后处理的流程后得到的SAR图像既保留了目标的细节信息又大大减弱了斑点噪声的影响,为后续的识别提供了更清晰的SAR目标图像。2.针对稀疏邻域保留嵌入(Sparse Neighborhood Preserving Embedding,SNPE)应用于SAR目标识别时在稀疏表示模型上的不足,提出改进的最大化稀疏重构间隙投影(Maximize Sparse Reconstruction Margin Projections,MSRMP),新算法不仅提升了识别率,还表现出对分类策略的不敏感性,只要特征维数足够大时,在不同分类器上识别率都能保持稳定,而且新算法传承了稀疏表示的抗噪性,在噪声严重的数据上仍保持较高的识别率。3.针对单个SAR目标拥有多角度图像的情况,对联合稀疏表示(Joint Sparse Representation,JSR)模型探讨,提出改进的联合稀疏表示(Improved Joint Sparse Representation,IJSR)模型,通过1范数最小化和低秩矩阵恢复措施寻求同一SAR目标多角度图像的共有模式,利用共有模式提取信息实现分类识别,将改进的联合稀疏表示和联合稀疏表示分别应用于MSTAR数据库上,结合稀疏表示分类策略的实验显示改进的联合稀疏表示提高了识别率。
[Abstract]:Synthetic Aperture Radar (Synthetic Aperture) is an imaging radar with high resolution. It is widely used in civil and military fields because of the advantage of all-weather and all-weather imaging. As one of the applications, SAR target recognition has become one of the hot research topics for many scholars because of its importance to national defense early warning. Sparse representation selects as few atomic linear reconstructed signals as possible from overcomplete dictionaries. When applied to recognition problems, not only natural recognition information is included in sparse representation coefficients, but also excellent noise resistance is shown. The speckle noise is a difficult problem in SAR target recognition, so the SAR target recognition based on sparse representation theory has a broad research prospect. Based on sparse representation and combined with the characteristics of SAR images, this paper researches on SAR image preprocessing, feature extraction and target recognition. The main research contents are as follows: 1. According to the characteristics of SAR target image in MSTAR database, a SAR image preprocessing method based on SVM (Support Vector Machine (SVM) classification idea is studied. After logarithmic transformation, stable SAR image segmentation based on SVM is achieved. The SAR image obtained by the post-processing process not only retains the detailed information of the target but also greatly reduces the influence of speckle noise, which provides a clearer SAR target image .2for the subsequent recognition. Aiming at the shortage of sparse representation model of sparse neighborhood reserved embedded (Sparse Neighborhood Preserving (SNPE) in SAR target recognition, an improved maximum sparse reconstruction gap projection (Maximize Sparse Reconstruction Margin projects (MSRMP) is proposed. The new algorithm not only improves the recognition rate, but also improves the performance of the algorithm. It also shows insensitivity to classification strategy. As long as the feature dimension is large enough, the recognition rate on different classifiers is stable, and the new algorithm inherits the anti-noise property of sparse representation. Still maintain a high recognition rate of. 3 on noisy data. For a single SAR target with multi-angle images, this paper discusses the joint sparse representation (Joint Sparse representation (JSR) model, and proposes an improved joint sparse representation (Improved Joint Sparse (JSR) model. The common pattern of multi-angle image of the same SAR target is obtained by minimizing 1-norm and restoring low rank matrix, and the common pattern information is extracted to realize classification and recognition. The improved joint sparse representation and the joint sparse representation are applied to the MSTAR database respectively. The experimental results with the sparse representation classification strategy show that the improved joint sparse representation improves the recognition rate.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52
，
本文编号：2196448
[Abstract]:Synthetic Aperture Radar (Synthetic Aperture) is an imaging radar with high resolution. It is widely used in civil and military fields because of the advantage of all-weather and all-weather imaging. As one of the applications, SAR target recognition has become one of the hot research topics for many scholars because of its importance to national defense early warning. Sparse representation selects as few atomic linear reconstructed signals as possible from overcomplete dictionaries. When applied to recognition problems, not only natural recognition information is included in sparse representation coefficients, but also excellent noise resistance is shown. The speckle noise is a difficult problem in SAR target recognition, so the SAR target recognition based on sparse representation theory has a broad research prospect. Based on sparse representation and combined with the characteristics of SAR images, this paper researches on SAR image preprocessing, feature extraction and target recognition. The main research contents are as follows: 1. According to the characteristics of SAR target image in MSTAR database, a SAR image preprocessing method based on SVM (Support Vector Machine (SVM) classification idea is studied. After logarithmic transformation, stable SAR image segmentation based on SVM is achieved. The SAR image obtained by the post-processing process not only retains the detailed information of the target but also greatly reduces the influence of speckle noise, which provides a clearer SAR target image .2for the subsequent recognition. Aiming at the shortage of sparse representation model of sparse neighborhood reserved embedded (Sparse Neighborhood Preserving (SNPE) in SAR target recognition, an improved maximum sparse reconstruction gap projection (Maximize Sparse Reconstruction Margin projects (MSRMP) is proposed. The new algorithm not only improves the recognition rate, but also improves the performance of the algorithm. It also shows insensitivity to classification strategy. As long as the feature dimension is large enough, the recognition rate on different classifiers is stable, and the new algorithm inherits the anti-noise property of sparse representation. Still maintain a high recognition rate of. 3 on noisy data. For a single SAR target with multi-angle images, this paper discusses the joint sparse representation (Joint Sparse representation (JSR) model, and proposes an improved joint sparse representation (Improved Joint Sparse (JSR) model. The common pattern of multi-angle image of the same SAR target is obtained by minimizing 1-norm and restoring low rank matrix, and the common pattern information is extracted to realize classification and recognition. The improved joint sparse representation and the joint sparse representation are applied to the MSTAR database respectively. The experimental results with the sparse representation classification strategy show that the improved joint sparse representation improves the recognition rate.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52
，
本文编号：2196448
本文链接：https://www.wllwen.com/kejilunwen/wltx/2196448.html
