辅助数据在面向对象分类方法中的应用_以密云水库上游为例
- 格式:pdf
- 大小:725.52 KB
- 文档页数:8
基金项目: 中国科学院战略性先导科技专项项目( XDA05050109) ; 全国生态环境十年变化( 2000—2010 年) 遥感调查与评估项目(2013-10-10;
修订日期: 2014-10-23
* 通讯作者 Corresponding author.E-mail: wubf@ radi.ac.cn
Key Words: object-based method; land cover; Miyun Reservoir area; high spatial resolution; auxiliary data
近年来,遥感影像的空间分辨率逐步提高,能够 帮助人们在较小的空间尺度上观察陆地表层格局与 变化,进行大 比 例 尺 遥 感 制 图,为 土 地 资 源 调 查、土 地利用 \ 土地覆盖变化、生态环境监测等提供更详 实、时效性 更 强 的 数 据 源[1]。 高 分 辨 率 遥 感 影 像 与 中低分辨率遥感影像相比具有更加丰富的纹理和形 状信息,而且数据量更大,若利用传统的基于像素的 影像分类方法,则不能充分利用更为丰富的信息,显 得效率不高,而且“椒盐”效应也更为明显,数据冗余 增多[2]。面向对象的分类方法将遥感影像中特征近 似的相邻像素归并为同一个基本分类单元———影像 对象( Image object) ,除了光谱信息,高分辨率影像中 地物的纹理、形 状 和 空 间 关 系 信 息 也 能 在 影 像 对 象 得以体现,分类效率更高,正广泛应用于高分辨率遥 感影像分类中[3]。最近的研究中通过对最优分割尺 度的 确 定[4]、分 类 特 征 的 优 化[5] 以 及 分 类 器 的 选 择[6]等方面进行研究,不断提高了面向对象分类方 法的精度。面向对象的分类方法对于单一类型的提 取精度较高,如道路[7]、水体[8]等,但在同一影像中 同时提取多种土地覆盖类型时,该方法容易受“同物 异谱、同谱异物”的影响,分类中仍然存在不确定性, 需要借助多 源 异 构 数 据,以 提 供 的 更 为 丰 富 的 信 息 来辅助分类。 因 此,本 文 以 密 云 水 库 上 游 地 区 的 为 例,采用高分辨率遥感影像,着重研究辅助数据在基 于面向对象分类方法中的应用。
http: / / www.ecologica.cn
24 期
苑全治 等: 辅助数据在面向对象分类方法中的应用———以密云水库上游为例
7203
the shape parameter to 0.1; correspondingly,the spectra parameter was set to 0.9. The shape features of the image object included compactness and smoothness,which were both set to 0. 5 because of the complex shape of the land cover. The smoothness and compactness of the objects were almost equally important. We established the sample database after segmentation. Every sample was an image object in the data base and had object features of spectra,shape,and texture. Basing on the sample database,we trained these samples and used the supervised classifier supplied by eCognition to classify the land cover automatically. However,the software still had some uncertainty in recognizing similar objects with different spectra and different objects with similar spectra. We used numerous auxiliary datasets to modify the SVM classifier results. Results revealed 26 types of land cover in the study area; 85% of which are deciduous broad-leaved shrubs, deciduous broad-leaved forests,dry lands,and grasses. This study used two methods,namely,field validation and visual validation,in evaluating the product accuracy to ensure the objectivity and comprehensiveness of the accuracy evaluation. The result of the field validation accuracy was 85%,whereas that of the visual evaluation accuracy was 86%. This study distinguished the evergreen coniferous forest and cultivated land through numerous auxiliary data. Results proved that the auxiliary data were vital for improving classification accuracy of objects,especially similar objects with different spectra and different objects with similar spectra.
Application of auxiliary data in the object-based classification method: a case study on the Miyun Reservoir area
YUAN Quanzhi,WU Bingfang* ,ZHANG Lei,LI Xiaosong,ZENG Yuan
辅助数据在面向对象分类方法中的应用
———以密云水库上游为例
苑全治,吴炳方* ,张 磊,李晓松,曾 源
( 中国科学院遥感与数字地球研究所数字地球重点实验室,北京 100094)
摘要: 密云水库上游位于华北平原向蒙古高原的过渡带内,自然条件差异大,人类活动对该区域影响显著。因此,对该地区土地 覆盖类型遥感监测方法的研究具有典型的指导意义。基于高分辨率遥感影像,着重探讨辅助数据在面向对象分类方法中的应 用,对密云水库上游地区的土地覆盖进行分类提取。结果显示,研究区内共包含 26 类土地覆盖类型,其中落叶阔叶灌木林、落 叶阔叶林、草丛以及旱地,这 4 种类型的面积占总面积的 85%,是全区的主要土地覆盖类型。在分类时,采用多源异构辅助数 据,研究了北方山区常绿针叶林、旱地等土地覆盖类型的识别方法,有效降低了“同物异谱”和“同谱异物”现象对分类精度的 影响。 关键词: 面向对象; 土地覆盖; 密云水库上游; 高分辨率; 辅助数据
第 34 卷第 24 期 2014 年 12 月
生态学报 ACTA ECOLOGICA SINICA
Vol.34,No.24 Dec.,2014
DOI: 10.5846 / stxb201310102433
苑全治,吴炳方,张磊,李晓松,曾源.辅助数据在面向对象分类方法中的应用———以密云水库上游为例.生态学报,2014,34( 24) : 7202-7209. Yuan Q Z,Wu B F,Zhang L,Li X S,Zeng Y.Application of auxiliary data in the object-based classification method: a case study on the Miyun Reservoir area.Acta Ecologica Sinica,2014,34( 24) : 7202-7209.
Key Laboratory of Digital Earth Science,Institute of Remote Sensing and Digital Earth,Chinese Academy of Sciences,Beijing 100094,China
Abstract: The Miyun Reservoir area between the North China plain and the Mongolian plateau has various physical conditions and intensity of human disturbance. Thus,studying the method of monitoring the land cover of this area is important. On the basis of high-resolution remote sensing imagery,this article classified the land cover of the Miyun reservoir area and studied the application of auxiliary data in object-based classification method. The remote sensing data used were mainly high spatial-resolution images,including RapidEye and SPOT-5. Accurate geometric rectification was performed initially. To gain more object features for distinguishing the different classes more clearly,we referenced the digital elevation model and slope gradient data with spatial resolution of 25m and the thematic map of the land use with a scale of 1∶10000. Multi-temporal HJ-1 imagery was added to separate the evergreen needle forests and the dry land. A total of 687 field samples were collected from the Miyun reservoir area for classification and precision testing. The eCognition v8.7 software was also used in this study. First,the images were segmented into different image objects according to the object features in the spectra,where several parameters are needed. In RapidEye,five bands with the same weight were used for segmentation. Choosing the segmentation scales and parameters of the shape and compactness is important. Through constant experiments,we found that the suitable segmentation scale was 45. The configuration of the shape parameter will determine the weight of the spectra in the segmentation. Given that the classification was mainly based on the spectral feature,we set