基于高分辨率影像的信息提取技术的研究及应用

发布时间：2018-06-09 06:07

本文选题：高分辨率影像 + 面向对象　；参考：《中国地质大学(北京)》2017年硕士论文

【摘要】：随着各国在航空航天领域的不断角逐,人造卫星的成功发射,使得我们所获取的航空数据的分辨率迅速提高,高分辨率影像随之产生。高分辨率数据在给我们带来丰富的数据信息的同时,也带来一系列的技术上的难题。针对高分辨率数据,传统的信息获取方法显然无法满足我们的需求。由于高分辨率数据携带的信息过于丰富,如何能够从纷繁复杂的数据中快速地获取我们需要的信息,是当今遥感领域的一个技术的难题。本文以湖北公安县的高分二号影像数据为例,选取具有代表性的、地物类型丰富的地段作为实验区域进行研究。由于影像主要分为城镇和乡村,我们选取两个典型区域代表两者进行实验,详细阐述了高分辨影像信息获取的整个流程:影像的预处理,传统的影像解译方法,面向对象的影像解译:影像分割与分类。本文主要针对高分辨率影像解译时采用的面向对象的方法进行研究。为了能够彰显面向对象这种方法的显著优势,专门设置传统的解译方法作为对照,和面向对象的提取结果进行对比分析。研究成果主要分为以下几点:(1)实验一二区监督分类的精度分别为69.98%、73.53%;非监督分类精度分别为34.41%、46.02%。由于高分辨率影像存在着“同物异普”和“异物同谱”,以及“椒盐”现象,使得二者的精度不高,同时由于非监督分类没有训练样本,分类精度更低。(2)通过研究不同尺度下,亮度均值标准差的变化来获取不同对象的最佳分割尺度。当亮度均值标准差最大时的分割尺度,就是此种地物的最佳分割尺度。(3)实验一、二区面向对象的分类精度分别为76.44%、88.68%。面向对象的分类方法的分类精度虽然有一定程度上的提高,但由于实验一区由于地物类型复杂,房屋道路光谱信息相近,水体和各种地物阴影光谱信息相近,使得分类精度并不是很理想。
[Abstract]:With the continuous competition of various countries in the field of aerospace and the successful launch of artificial satellites, the resolution of the airborne data obtained by us improves rapidly and the high-resolution image is produced. High-resolution data not only brings us abundant data information, but also brings a series of technical difficulties. For high-resolution data, the traditional information acquisition method obviously can not meet our needs. Because the high resolution data carries too much information, how to get the information we need quickly from the complicated data is a technical problem in the field of remote sensing. Taking the image data of Gaofen 2 in Gongan County of Hubei Province as an example, this paper selects the representative area with abundant ground objects as the experimental area to study. As the image is mainly divided into towns and villages, we select two typical regions to represent the two experiments, and elaborate the whole process of high-resolution image information acquisition: image preprocessing, traditional image interpretation methods, Object-oriented image interpretation: image segmentation and classification. In this paper, the object-oriented method used in high-resolution image interpretation is studied. In order to show the obvious advantage of the object-oriented method, the traditional interpretation method is specially set as the contrast, and the result of the object-oriented extraction is compared and analyzed. The research results are divided into the following points: 1) the accuracy of the supervised classification in the first and second areas of the experiment is 69.98 and 73.53, respectively, and the accuracy of the unsupervised classification is 34.41 and 46.02, respectively. Because of the existence of "isomorphism", "foreign body isospectral", and "salt and pepper" in high-resolution images, the accuracy of the two images is not high, and because there are no training samples for the unsupervised classification, the classification accuracy is lower. The change of standard deviation of luminance means to obtain the best segmentation scale of different objects. When the standard deviation of luminance means is the largest, the best segmentation scale of this kind of ground object is. Experiment 1. The object oriented classification accuracy of the two areas is 76.440.88.68, respectively. Although the classification accuracy of the object-oriented classification method has been improved to some extent, because of the complex types of ground objects, the building road spectral information and the spectral information of water body and various ground objects, the spectral information of water body and all kinds of ground objects are similar. The classification accuracy is not ideal.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237

【相似文献】