基于中分辨率TM数据的湿地水生植被提取

利用湿地水生植被生长旺盛、光谱反射较强、光谱信息比较丰富的8月份中分辨率Landsat TM和ETM+多光谱遥感影像,采用面向对象的分类方法,进行野鸭湖湿地水生植被的提取。研究表明:在提取过程中,通过对原始影像进行主成分变换和穗帽变换,将主要信息与噪声分离,不仅减小了数据冗余和波段间的相关性,而且增大了影像上湿地水生植被与其他地物类型光谱和空间信息的差异性,并结合野外水生植被光谱特征分析,选择归一化植被指数NDVI与归一化水体指数NDWI辅助分类,构建特征波段或波段组合,然后,确定适当的隶属度函数和阈值范围,构建分类决策树,完成湿地水生植被的自动分类,提高了影像分割与面向对象分类的精度,取得了较为理想的湿地水生植被提取结果。2002年和2008年两景影像的总体分类精度分别达到86.5%和85.44%,表明中分辨率TM影像可以满足湿地水生植被提取的需要,又因为其具有较高的波谱分辨率、极为丰富的信息量、相对较低的价格、长时间序列,可以作为近20a湿地水生植被提取和动态变化监测的主要数据源。

The extraction of wetland hydrophytes types based on medium resolution TM data

Wetland hydrophytes are an important part of wetland resource, and the change of their number and extent have a direct impact to the wetland habitat quality. So the quantitative study of wetland hydrophytes change is of great significance for the protection of wetland change. We found that wetland hydrophytes classification using remote sensing images is less detail oriented, the application of object-oriented classification method applying medium-resolution remote sensing image in wetland hydrophytes classification is not sufficient. In this research, we applied object-oriented classification method to extract the wetland hydrophytes types of Wild Duck Lake Wetland Natural Reserve in Beijing, China. The medium-resolution Landsat TM and ETM+ multi-spectral remote sensing images which the resolution is 30 meter acquired in August, 2002 and 2008 were selected as the data source of the classification. The reason that the August imageries were selected because they show the vigorous growing season of wetland hydrophytes, and have strong spectral reflectance and plentiful spectral information. The results show: (a) during the extraction process, using Principal Component Transformation (K-L Transformation) and Tasseled Cap Transformation (K-T Transformation) to separate the key information from background noise can reduced data redundancy, the Principal Component Transformation only retains the first component of the largest eigenvalue, Tasseled Cap Transformation component outputs the first three features; (b) the method can correlate different bands; (c) the method can increase the diversity of spectral and spatial information between wetland hydrophytes and other surface features on land or water. Meanwhile, after analyzing of wetland hydrophytes, a typical wetland hydrophytes spectral characteristic curve was drawn comprehensively. Through the field spectral characteristics analysis, NDVI and NDWI were applied for classification, and characterizing the bands or band combinations were constructed. Then using decision tree method for data analysis, the automatic classification of wetland hydrophytes could be accomplished through the appropriate membership function and threshold range as constructing the decision tree. After the automatic classification, make use of manual editing module in eCognition, we combined spectral, size, texture and other structural information of features, and assigned the polygon objects to the correct category. This approach increased the classification accuracy. By making use of classification stability and classification accuracy test based on the polygon objects, the results of overall accuracy of 2002 and 2008 classifications were 86.5% and 85.44%, respectively. And classification stability standard deviation is less than 0.25, the average stability is greater than 0.84.The research indicated that medium- resolution TM image can meet the needs of wetland hydrophytes extraction. Moreover, because of the nature of high spectral resolution, plenty of vegetation cover information, relatively lower cost, and longer time accumulation of imageries, TM and ETM+ can be used as the main data source of wetland hydrophytes extraction and dynamic monitoring, in particular using the data from the last two decades. It can not only provide a more intuitive, scientific and accurate basis for the wetland hydrophytes protection in Wild Duck Lake area, but also provide a new way for other areas of wetland aquatic vegetation extraction and change detection.