高分辨率遥感影像天然林与人工林植被覆盖信息提取  

Extracting Natural and Artificial Forest Information Based on High Resolution Remote Sensing Data

作　　者：王荣[1,2] 江东[2] 韩惠[1] 张峰 赵少华 

机构地区：[1]兰州交通大学测绘与地理信息学院,兰州730070 [2]中国科学院地理科学与资源研究所资源环境科学数据中心,北京100101 [3]国家环境保护部卫星环境应用中心,北京100094

出　　处：《资源科学》2013年第4期868-874,共7页Resources Science

基　　金：高分辨率对地观测系统重大专项项目;中国科学院重点部署项目(编号:KZZD-EW-08)

摘　　要：森林类型识别是森林资源遥感监测的基础工作,低分辨率遥感影像缺少纹理细节信息,高分辨率遥感影像不仅具有光谱信息,而且提供丰富的空间、纹理特征信息,因此基于高分辨率遥感影像的纹理特征进行森林内部信息的提取成为近年研究的热点及难点,传统基于单个象元纯光谱及面向对象建立规则集的方法难以有效区分天然林与人工林植被覆盖信息。本文利用面向对象多尺度分割算法、Sobel算子边缘检测及骨架线提取等方法,提取天然林与人工林的纹理线特征,构建了一种新的特征指数——纹理线条密度指数(TLDI)。研究表明:与常用的NDVI、SAVI、EVI等植被特征及VAR、HOMO、CON等GLCM纹理特征指数相比,TLDI指数的离散度更好、分类效果更佳;当TLDI>0.1时,为天然林植被覆盖区;当0<TLDI<0.1时,为人工林植被覆盖区。典型区域的实验表明,除极少数稀疏天然林对象斑块错分人工林,总体分类精度较高,人工林错分率小于0.83%,面向对象的TLDI指数可以有效地提取森林内部天然林与人工林植被覆盖信息。The recognition of forest type provides an important basis for forest monitoring and ecological protection. Forest land can be easily distinguished from other land cover types in the multiple spectral bands of satellite images, but traditional methods based on single pixel and pure spectra, common vegetation indexes （NDVI, SAVI, EVI） or image texture feature values （HOMO, ENT, Con, DISS, SEC and VAR） cannot effectively distinguish between different types of forest vegetation. Although there are some differences between natural forest and plantation in GLCM texture feature value of RG, GB, the differences of GB is only 0.2 between natural forest and plantation and it is very difficult to classify natural forest and plantation. The objective of this study was to raise a new method for artificial-natural forest classification using multiresolution object-oriented effective segmentation based on optimal split scales, selecting samples, sobel edge detection and an extracting skeleton. We then construct a new feature index, the Texture Line Density Index （TLDI）. The performance of the new method was tested with several geo-statistical texture measures from IKONOS multiple spectral images in South China. Compared with commonly used vegetation indexes and the GLCM texture index, the discrete degree of TLDI and classification result are superior. When TLDI〉0.1, the area is natural forest vegetation coverage;when 0〈 TLDI 〈 0. 1, the research area is plantation vegetation coverage; and when 0 〈 TLDI 〈0.1, plantation vegetation coverage is present. With rare natural forest segmentation patches as exceptions, overall classification accuracy is higher, and the misclassification rate of plantation less than 0.83%.

关 键 词：高分辨率遥感 多尺度分割 森林资源监测 植被信息提取 

分 类 号：S757.2[农业科学—森林经理学]