【作者】 于鹏；

【导师】 刘富；

【作者基本信息】 吉林大学 ， 控制理论与控制工程， 2011， 硕士

【摘要】 高分辨率可见光遥感图像舰船目标自动检测和识别在军事上具有十分重要意义。本文以工程应用为背景,以图像理解理论为基础,在充分分析可见光遥感图像成像特点的基础上,重点围绕可见光遥感图像舰船目标检测、目标特征提取、分类器设计等方面展开研究。建立了一个舰船目标自动识别系统,能够快速准确地从宽幅遥感图像数据中检测和识别舰船目标。主要研究内容主要包括以下几个部分:1、针对海洋背景下舰船目标检测,提出了基于Top-Hat变换的图像恒虚警率舰船目标检测算法,能够实现舰船目标的快速检测和定位。通过大量实验数据分析,验证了该算法的有效性。2、港口内舰船目标检测方面,采用基于卫星辅助数据和港口先验信息的目标检测方法,先在宽幅成像的遥感图像中准确定位港口位置,然后根据港口先验模板信息,检测并识别港口内舰船目标。3、本文采用了一种新的采用独立分量分析(ICA) Zernike矩的目标形状特征识别方法。该方法基于目标分割,对分割后的目标区域进行独立分量分析,利用三阶中心矩将目标形状转化为标准形式并提取Zernike矩作为特征向量进行识别。通过理论分析与实验证明此方法性能鲁棒。4、采取基于支持向量机分类器的方法进行分类器设计。先通过稳定性好、抗干扰能力强的特征进行粗分类,降低快速检测中的高虚警率。然后对初次分类结果再提取ICA Zernike不变矩作为特征向量进行分类识别。这种方法可以大大提高识别系统的运行效率。详细论述了可见光遥感图像舰船目标自动识别系统的工作流程和关键技术,本文最后对研究内容进行了总结并给出了下一步研究方向。更多还原

【Abstract】 Automatic ship target detection and recognition in high-resolution optical remote sensing images have a great value in the military. This paper, which background is engineering application, is based on the theory of the understanding of optical remote sensing image. It studies on the aspects of ship target detection, Ship feature extraction, classification design and establishment of the Sea Area target Surveillance System. It can distinguish ship target from massive remote sensing image data quickly and accurately. The paper mainly includes the following several facets:1、For offshore ship target automatic detection, Proposes a image constant false alarm rate method base on Top-Hat transformations. And a lot of ship detection experiments were carried out and the results demonstrated the efficiency of the algorithm.2、For in-shore ship target automatic detection. A method of target detecting and recognition based on Satellite auxiliary data and prior information of the harbors has been proposed. Firstly, the method makes sure the accurate position of the harbor target in the image of large size. Then, it detects and recognizes the ship target which lie in the harbor area, according to the structure information of the harbor target.3、Proposed a novel feature recognition method for ship target based on the Independent Component Algorithm (ICA) zernike invariant moments. Firstly, we analyze the divided Region of Interest (ROI) with ICA method. And, we transform the shape of target to the canonical form. Then, the invariant moments of normalized shapes can be extracted. It will be used as feature vector to do further recognition. The experiment demonstrates that the performance of this method is robust.4、A pattern recognition classifier based on support vector machine is designed in this Paper .rough classification based on stability and anti-interference characteristics used to excluded false target. Then ICA zernike invariants moment were extracted to classification. Furthermore, this method can recognize the ship target from the remote sensing images effectively.Finally, the functional design and key technologies of the Sea Area Target Surveillance System for Remote Sensing Images are discussed. concludes the research of this thesis. Some problems and interesting area for future research are pointed out.更多还原