【作者】 马婷婷；

【导师】 冯德俊；

【作者基本信息】 西南交通大学 ， 摄影测量与遥感， 2012， 硕士

【摘要】 以QuickBird, GeoEye-1为代表的高分辨率遥感影像数据在国土和规划等相关部门正得到越来越普及的应用,但无论是专业信息提取、动态变化检测,还是专题地图制作等应用都离不开对遥感影像分类。为了满足人们生产应用的要求,寻求一种高效的影像分类方式有着重要的意义。高分辨遥感影像具有更为丰富的光谱、形状、纹理、上下文、拓扑等信息,使得常规的基于像元的信息提取方法难以满足需求,而面向对象的方法弥补了基于像元信息提取技术在处理细节信息中的缺陷。但在面向对象的高分辨率遥感影像信息提取方法中,影像分割是基础,它直接决定着信息提取的精度。不同的分割参数和分割尺度会导致截然不同的分割结果(即影像对象),对于某种特定的地物目标,只有在合适的尺度上才能更有效、更准确的对其进行信息提取。本文首先分析了基于像元和面向对象信息提取的理论和研究现状,总结出影像分辨率的不断提高是面向对象思想被引入到影像信息提取中的关键因素,并且通过对影像分割参数的定量实验,总结了各个分割参数在影像对象生成中所起的作用。然后利用分类后目视解译法和最佳分割效果原则,论证了改进面积对比法能快速计算出个地物最佳分割尺度。其次,本文总结了一套相对完整的面向对象的信息提取技术思路：针对细小地物建立较小的分割尺度,提取出影像中较小的地物；在较小的分割尺度上采用同质性对生成的影像对象进行合并,提取出较大的地物；针对纷繁复杂的提取规则,建立了先光谱信息后纹理信息的规则模式。最后,通过面向像元的信息提取技术和面向对象的信息提取技术对实验区数据进行影像分类,并采用混淆矩阵、Kappa系数等指标对各自分类结果进行了精度评价,验证了面向对象信息提取技术在处理高分辨率遥感影像上的优势。利用遥感机理和阴影校正模型理论对第二个研究子区域中基于面向对象分类结果中的阴影其进行补偿,使阴影中的信息得到恢复,并利用面向对象方法对其进行了重新分类,提高了分类结果的应用。更多还原

【Abstract】 Now days, application of high resolution remote sensing image data, e.g. QuickBird, GeoEye-l,in land managing, planning and concerned application fields has been popularized increasingly. Generally, the classifacation of remote sensing images are significant to thematic information extraction, dynamic change detecting and thematic mapping. For actual demand in production, it’s reasonable to introduce high-efficiency approaches in generating information from the high resolution images. Compared with low resolution images, more spectrum, shape, texture, context and topology information are included in high resolution images, leading to inadaption of conventional information extraction methods that are on basis of image pixels. However, the object oriented method is capable to conquer such drawbacks in conventional methods for its adaptive capability in dealing with scale parameters of high resolution images because of the strategy of image segmentation.image segmentation is the premise and basis in object oriented information classification and extraction, and it is possible for image segmentation to significantly influence the accuracy of extracted information. Dependently, different segmentation parameters and scales lead to distinct segmented results (calling image objects), thus the appropriate parameters and scales must be determined in order to make the information extracting more efficient and accurate.In this paper, we analyze the pixel based and object oriented information extraction theory and their current research situation firstly, and indicate that the introducing of object oriented method in image analysis is to adapt the enhancement of image resolution. Based on the image segmentation parameter quantitative experiments, the role of each segmentation parameter in image object generating is summed up, and then the classification of visual interpretation method and optimal segmentation effect principle are used to demonstrate that the improved area contrast method can quickly calculate a feature optimal segmentation scale, furthermore, we produce an integrated strategy for information extraction with object oriented method: set a small scale for smaller features and the features are subsequently extracted; merge the smaller image objects in a small segmentation scale based on the homogeneity judgment standard, and the bigger features are then extracted; according to various extraction regulations, set up a rule mode based on spectral and texture information in sequence. Then, the classification of the study area is done used pixel oriented and object oriented technology,and the confusion matrix, Kappa coefficient and other indicators are examined to evaluate the accuracy of extraction results, respectively, verifying the advantages of object oriented information extraction technology in the processing of high resolution remote sensing images. Finally, we use the mechanism of remote sensing and the shadow correction method to compensate and restore the information in shadow areas, which is extracted by object-oriented information extraction technology from the second study area, Then use the object-oriented approach to re-image classification.from that it could improve the classification results of the application.更多还原