基于高分辨雷达目标特征提取与识别方法研究

发布时间：2018-03-31 01:19

本文选题：一维距离像　切入点：平移不变特征　出处：《沈阳理工大学》2015年硕士论文

【摘要】：利用高分辨雷达对目标进行识别是当代雷达系统的一个主要发展趋势。目前基于高分辨雷达的目标识别在军事及民用方面都已经有了一定程度的应用。本论文利用实测高分辨雷达回波数据进行实验,重点研究了雷达识别系统中目标的特征提取与融合,并利用支持向量机分类器对选定的特征进行识别。主要工作包含以下内容:首先,在针对雷达目标一维距离像的形成和性质进行研究的基础上,发现雷达目标的一维距离像中包含了大量的目标结构以及形状等信息,因此,针对一维距离像进行特征的提取。在研究了一维距离像平移敏感性的基础上,采用提取平移不变特征的方法来克服这一问题。分别提取了目标的功率谱特征、中心距特征以及幅度谱差分特征,并且研究发现了目标的这三种提取特征具有差异性。其次,为了获取有效特征,本文提出了一种基于粗糙集改进的主成分分析融合方法应用于雷达目标识别上。在雷达目标特征识别系统中最重要的环节就是目标的特征选取。选取的特征既要能够描述目标,又要与其他相似目标有一定的差异性。在不影响特征信息含量的同时还应该尽量的减少特征维数,争取利用最少的特征来包含目标最有效的信息,进而做到快速、高效、准确的识别目标。对目标的功率谱特征、中心距特征以及幅度谱差分特征进行主成分分析,然后基于粗糙集理论对目标特征进行属性约简,使得融合后的特征具有大量的目标信息,同时大幅度地降低了特征维数,从而保证该融合特征的优越性。最后,详细的介绍了支持向量机分类器的原理及应用,并应用三种不同的算法配合多种核函数将一维距离像的目标融合特征和单一特征分别进行识别,研究结果表明,基于粗糙集改进的主成分分析融合的特征不仅在识别方面强于其他特征,而且其特征维数也大幅度的降低了。这样既提高了识别系统的识别率,同时也节省了识别系统的存储空间,减轻了系统的运算复杂程度。
[Abstract]:Target recognition based on high resolution radar is a main developing trend of modern radar system. At present, target recognition based on high resolution radar has been applied in military and civilian fields to a certain extent. The experiment is carried out with the measured high resolution radar echo data. The feature extraction and fusion of targets in radar recognition system are studied emphatically, and the selected features are recognized by support vector machine classifier. The main work includes the following: first, Based on the research on the formation and properties of radar target one-dimensional range profile, it is found that the one-dimensional range profile of radar target contains a lot of information such as the structure and shape of the target. Based on the study of the translation sensitivity of the one-dimensional range profile, the method of extracting the translation invariant feature is used to overcome this problem. The power spectrum features of the target are extracted, respectively. The center distance feature and amplitude spectrum difference feature, and the study found that the three extraction features of the target have differences. Secondly, in order to obtain effective features, In this paper, an improved principal component analysis fusion method based on rough set is proposed for radar target recognition. The most important part of radar target feature recognition system is the feature selection of the target. The selected feature must be able to describe the target. We should not affect the content of feature information, but also try to reduce the dimension of the feature, try to use the least features to include the most effective information of the target, so as to be fast and efficient. The power spectrum feature, center distance feature and amplitude spectrum difference feature of the target are analyzed by principal component analysis, and then the target feature is reduced based on rough set theory. The fusion features have a large amount of target information and greatly reduce the feature dimension, so as to ensure the superiority of the fusion feature. Finally, the principle and application of SVM classifier are introduced in detail. Three different algorithms combined with a variety of kernel functions are used to identify the target fusion feature and the single feature of the one-dimensional range profile, and the results show that, The feature of principal component analysis fusion based on rough set is not only stronger than other features in recognition, but also its feature dimension is greatly reduced, which not only improves the recognition rate of the recognition system, but also improves the recognition rate of the recognition system. At the same time, the storage space of the recognition system is saved, and the complexity of the system is reduced.
【学位授予单位】：沈阳理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN957.52

【参考文献】