基于遥感图像中港口目标的分割算法研究与实现

发布时间：2018-06-15 13:43

本文选题：遥感图像 + 图像分割　；参考：《西安电子科技大学》2014年硕士论文

【摘要】：遥感图像目标分割,是遥感图像识别和图像解译的重要步骤。遥感图像分割具有非常强的目标特性和图片类型针对性,因而目标分割算法必须通过精心设计,图像分割才能达到良好的效果。港口目标作为一个国家重要的军事基地和航运交通枢纽,研究关于它的信息,具有重大战略意义和民用意义。遥感图像中港口目标的分割和研究,当今时代,无论在军事领域还是民用领域都是至关重要的研究任务,也是图像解译领域的一个热点研究问题。针对遥感图像中港口目标的分割和提取研究任务,因其目标与背景的复杂性和特征的多样性,要达到精确的分割效果,需要针对特定目标选择合适的分割算法。本文的主要内容是遥感图像中港口目标的分割算法研究与实现,遥感图像来源于google earth。选取两幅遥感图像,分析港口目标的特征信息:灰度特征及结构特征。在此基础上选择阈值分割法和脉冲耦合神经网络分割法,实现图像分割。本文的主要研究内容如下:1.介绍了图像阈值分割法,在一维最大类间方差法的基础上研究了二维最大类间方差法,确定图像分割的最佳阈值,实现港口目标的分割。2.详细研究了一种改进的图像阈值分割方法,加权参数直觉模糊熵阈值分割算法。在直觉模糊熵分割法的基础上增加了权参数,改善了直觉模糊集缺少对模糊不确定性度量的局限性。目的为改善传统阈值分割法的分割效果,以及在一定程度上消除或抑制云雾因素对分割结果的影响。3.考虑到阈值分割方法自身的局限性,一个或一组阈值无法精确分割遥感图像中的港口目标,在此基础上研究了基于神经网络的分割算法:脉冲耦合神经网络(PCNN-Pulse Coupled Neural Networks)分割算法,介绍了此算法的核心内容和分割原理,考虑到算法自身的局限性,加入了图像边缘算法及去噪步骤,对分割结果进行优化。实现上述几种分割算法,通过对分割结果的分析与讨论,验证它们的可行性。结果表明了这些算法在一定程度上提高了分割效率并且优化了分割结果。
[Abstract]:Object segmentation of remote sensing image is an important step in remote sensing image recognition and image interpretation. Remote sensing image segmentation has very strong target characteristics and image type pertinence, so the target segmentation algorithm must be carefully designed in order to achieve good results. As an important military base and shipping transportation hub of a country, it is of great strategic and civil significance to study the information about port target. The segmentation and research of port targets in remote sensing images is an important research task in both military and civil fields, and it is also a hot research issue in the field of image interpretation. Due to the complexity of the target and background and the diversity of the features, it is necessary to select the appropriate segmentation algorithm for the specific target in order to achieve the accurate segmentation effect for the research task of port target segmentation and extraction in remote sensing image. The main content of this paper is the research and implementation of the segmentation algorithm of port target in remote sensing image, which comes from google earthland. Two remote sensing images are selected to analyze the feature information of port target: grayscale feature and structure feature. On this basis, the threshold segmentation method and the pulse coupled neural network segmentation method are selected to realize image segmentation. The main contents of this paper are as follows: 1. In this paper, the method of image threshold segmentation is introduced. Based on the one-dimensional maximum inter-class variance method, the two-dimensional maximum inter-class variance method is studied to determine the optimal threshold value of image segmentation, and to realize the segmentation of port target. An improved image threshold segmentation method, weighted parameter intuitionistic fuzzy entropy threshold segmentation algorithm, is studied in detail. Based on the method of intuitionistic fuzzy entropy segmentation, the weight parameter is added, and the limitation of the lack of fuzzy uncertainty measurement in intuitionistic fuzzy set is improved. Objective to improve the segmentation effect of the traditional threshold segmentation method and to eliminate or suppress the influence of cloud and fog factors on the segmentation results to a certain extent. Considering the limitation of threshold segmentation method, one or a group of thresholds can not accurately segment port targets in remote sensing images. On this basis, a neural network segmentation algorithm based on neural network is studied: PCNN-Pulse coupled Neural Network (PNN) segmentation algorithm. The core content and segmentation principle of the algorithm are introduced. Considering the limitations of the algorithm, the image edge algorithm and denoising steps are added to optimize the segmentation results. By analyzing and discussing the segmentation results, the feasibility of these algorithms is verified. The results show that these algorithms improve the segmentation efficiency to some extent and optimize the segmentation results.
【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】：TP751

【参考文献】