【作者】 何少林；

【导师】 徐京华；

【作者基本信息】 西南交通大学 ， 地图制图学与地理信息工程， 2013， 硕士

【摘要】 随着遥感技术的发展,尤其是影像数据源的丰富和影像处理水平的提高,使得遥感技术的应用越来越广泛。以大飞机、卫星等为平台的航空航天遥感测量已经得到广泛的应用,但是在西南多云雾地区和飞行困难地区,利用大飞机和卫星等为平台的航空航天遥感测量难获取满足要求的高分辨遥感影像。而运用无人机遥感可以在多云雾地区低空飞行,获取高分辨率遥感影像。利用高分辨率影像较之低分辨率影像能更容易的获取地物的类别属性信息,更好地实现对土地利用信息可视化。但是传统的影像分类方法导致分类精度不高,影响着无人机高分辨率遥感影像在土地利用的应用效率。如何利用一种新的影像分类方法提高无人机高分辨率遥感影像提取土地利用信息的精度有着十分重要的意义。本文利用无人机低空遥感技术获取的高分辨率影像数据,选取四川省德阳市某村庄为实验区域,结合无人机高分辨率影像相对于卫星影像能提供更多的纹理、形状以及地表相互空间信息,采用面向对象多尺度分割分类方法对实验区域进行土地利用分类研究。面向对象多尺度分割分类方法避免了传统基于像元分类方法产生的“同物异谱”及“同谱异物”现象,同时克服了面向对象单一尺度产生“过分割”和“欠分割”问题。该实验为利用无人机影像提取土地信息及制图等方面的工作提供一定的指导意义。论文所做的主要工作如下：1.分析国内外无人机遥感技术在获取信息方面的应用,分析了国内外遥感影像土地信息提取技术应用现状。2.根据RTK采集的无人机影像像控点,进行离散点内插规格格网DEM实验,探讨样本点选取和内插算法选择等因素对构建规格格网DEM的精度评价。3.总结常规航测数据处理方法和流程,以无人机影像为数据,经过必要数据预处理,选取LPS (Leica Photogrammetry Suite)为平台,制作实验区的数字正射影像DOM和数字高程模型DEM。4.阐述影像分割技术和面向对象多尺度影像分割技术的原理和方法,通过实验对影像分割中的光谱因子、形状因子(包括光滑度和紧密度)参数进行确定,分析一些最优分割尺度的选择方法,以均值标准差法对影像进行实验并选择出不同地物的最优分割尺度。5.对无人机影像研究区域进行多尺度多层次的土地利用分类实验,通过实验分析比较得到不同土地类型的最优分割尺度,并在此基础上建立分类规则进行多尺度土地信息提取。并进行了精度分析评价,制作土地利用专题图。更多还原

【Abstract】 With the development of remote sensing technology, the rich of image data source and the raise of image processing level make the application of remote sensing technology more and more widely. The aerospace remote sensing on large aircraft and satellites platform have been applied widely, but it is difficult to get the high-resolution remote sensing images in the southwest region with more clouds and the flight difficult areas. The UAV(Unmanned aerial vehicles) of flying at low altitude can obtain high-resolution remote sensing image in more clouds areas. The high-resolution image can obtain the type of the feature attribute information more easily and land use information visualization better than low-resolution image. However, the traditional image classification method leads to classification accuracy is not high, and affects UAV high-resolution remote sensing images application efficiency in land-use. How to use a new image classification method to improve UAV high-resolution remote sensing images accuracy of Land Use information has a very important significance.This thesis combining with the UAV high-resolution images can provide more texture, shape and mutual spatial information on the surface than satellite images, and take object-oriented multiscale segmentation classification to study UAV high-resolution images land use classification in the experimental area of a village, Deyang City, Sichuan Province. The object-oriented multiscale segmentation classification method avoid the pixel-based traditional classification phenomenon of different objects which have the same spectrum and the same objects have different spectrum, and overcome the object-oriented single scale problem of over segmentation and less segmentation. This experiment provides some guidance for land information extraction and mapping on UAV image.The main contents are as follows:1. The domestic and foreign UAV technology application of obtaining information and the land information extraction technology application status on remote sensing image were analyzed.2. According to the UAV images control points collected by RTK, it had explored the accuracy of regular grid DEM construction about sample points and interpolation algorithm selection factors such as in experiments of grid DEM construction interpolated by discrete points.3. The conventional methods of aerial survey data processing were summeried. LPS was selected as a platform, the experimental zone DOM and DEM were completed on UAV images data were preprocessed.4. The principles and methods of object-oriented multi-scale image segmentation techniques were elaborated and the optimal segmentation scale selection methods were analyzed. Spectral factor and shape factor (including smoothness and compactness) were determined in image segmentation experiment and optimal segmentation scales of different objects were selected in mean standard deviation experiment.5. In multi-scale multi-level land use classification experiment of UAV image study area, the optimal segmentation scales of the different land types were analyzed and obtained, and on this basis the classification rules for multi-scale were established and land information were extracted. The accuracy of result was analyzed and evaluated, and the thematic map of land use was made in experiment.更多还原