高分辨SAR图像机动目标检测与识别技术研究

发布时间：2018-06-30 18:19

本文选题：SAR图像 + 目标检测　；参考：《西安电子科技大学》2014年硕士论文

【摘要】：随着合成孔径雷达成像技术与应用领域的快速发展，合成孔径雷达的信息采集能力不断增强，人工解译已较难适应其数据量的快速增长，借助于计算机、机器学习和模式识别相关技术对合成孔径雷达图像进行自动或半自动地解译可以在较大程度上提高对数据的处理效率，这对于军事和民用领域都具有良好的应用价值。随着SAR图像分辨率的提高，它在军事领域中的应用也由最初的检测目标逐步扩展为检测并识别目标，基于高分辨SAR图像的机动目标检测、鉴别、识别是SAR图像解译领域中重要的研究方向，，已成为国内外的研究热点。本文从高分辨SAR图像中机动目标的检测、鉴别与识别三个方面进行研究，从分类的角度考虑上述三个问题，所做主要工作如下： (1)研究了经典的恒虚警率检测方法，分析了SAR图像中机动目标与常见背景区域的统计特性。针对SAR图像分辨率的提高，提出了对目标和背景同时建模的基于贝叶斯分类器的机动目标检测算法。此外，由于在SAR图像的目标检测阶段所要处理的数据量是巨大的，因此，在上述算法的基础上进一步提出用视觉显著注意模型先求取图像中的显著区域，再进一步用检测方法进行检测，算法不仅具有较好的检测性能，更有良好的实时性。 (2)研究了表征学习的基本理论方法，并将表征学习应用于高分辨SAR图像的机动目标特征提取上。在此基础上分别提出了基于稀疏表示的高分辨SAR图像机动目标鉴别方法和基于One-Class SVM的高分辨SAR图像机动目标鉴别方法。基于稀疏表示的目标鉴别在正样本较少的情况下也获得了较好的鉴别性能，而基于One-Class SVM的目标鉴别在正样本未知的条件下，也具有良好的鉴别性能，克服了实际中目标训练样本未知的鉴别难题。 (3)研究了在语义分析中常用的词袋（BoW）模型的基本思想及其典型应用，分析了高分辨SAR图像中机动目标的常用特征，提出了一种基于BoW模型的高分辨SAR图像机动目标识别方法。并通过实验比较了该方法与稀疏表示分类器的在高分辨SAR图像机动目标识别中的性能。本文的工作得到了国家重点基础研究发展计划（973计划）： No.2013CB329402，国家自然科学基金（61072108，60601029，60971112，61173090），新世纪优秀人才项目：NCET-10-0668，高等学校学科创新引智计划（111计划）：No. B0704，教育部博士点基金(20120203110005)，武器装备预研基金项目(9*****)，以及华为创新研究计划项目(IRP-2013-01-09)的资助。
[Abstract]:With the rapid development of synthetic Aperture Radar (SAR) imaging technology and applications, the ability of SAR information acquisition has been enhanced, and it is difficult to adapt to the rapid growth of data volume by manual interpretation. The automatic or semi-automatic interpretation of SAR images by machine learning and pattern recognition techniques can improve the efficiency of data processing to a large extent, which has good application value in both military and civilian fields. With the improvement of SAR image resolution, the application of SAR image in the military field is gradually expanded from the original detection target to detect and recognize the target, and the maneuvering target detection and identification based on high-resolution SAR image. Recognition is an important research direction in the field of SAR image interpretation, and has become a hot spot at home and abroad. In this paper, the detection, identification and recognition of maneuvering targets in high-resolution SAR images are studied. The above three problems are considered from the perspective of classification. The main work is as follows: (1) the classical CFAR detection method is studied. The statistical characteristics of maneuvering targets and common background regions in SAR images are analyzed. Aiming at the improvement of SAR image resolution, a maneuvering target detection algorithm based on Bayesian classifier is proposed. In addition, since the amount of data to be processed in the target detection phase of SAR images is huge, a visual salient attention model is proposed to obtain the salient regions of the SAR images first based on the above algorithms. The algorithm not only has good detection performance, but also has good real-time performance. (2) the basic theoretical method of representation learning is studied. Representation learning is applied to feature extraction of maneuvering targets in high resolution SAR images. On the basis of the above, the maneuvering target identification methods based on sparse representation and One-Class SVM are proposed, respectively. The target discriminant based on sparse representation also obtains better discriminant performance under the condition of fewer positive samples, while the target discriminant based on One-Class SVM also has good discriminant performance under the condition of unknown positive samples. It overcomes the problem of unknown target training samples in practice. (3) the basic ideas and typical applications of the word bag (BoW) model commonly used in semantic analysis are studied, and the common features of maneuvering targets in high-resolution SAR images are analyzed. A maneuvering target recognition method for high resolution SAR images based on Bow model is proposed. The performance of this method is compared with that of sparse representation classifier in high resolution SAR image maneuvering target recognition. The work of this paper has been obtained from the National key basic Research and Development Program (973 Program): No. 2013CB329402, the National Natural Science Foundation of China (61072108), the New Century Outstanding Talent Project: NCET-10-0668, the discipline Innovation and introduction Program of higher Education (111 Program): no. B0704, Ministry of Education Ph.D. Foundation (20120203110005), weapons and equipment Pre-Research Foundation (9 *), and Huawei Innovation Research Program (IRP-2013-01-09).
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【参考文献】