基于Landsat 8 OLI数据的树种类型分布提取

树种分布是林业调查工作中的关键环节,遥感影像中对树种信息的提取具有重要意义。论文以福建省长汀县为研究区,收集2016年3月美国陆地卫星影像(Landsat 8 OLI)数据,提出基于知识库的图像膨胀分类方法,结合不同树种类型分布点的影像像元灰度与影像不同波段的重新排列,对杉类、松类、竹类与阔叶树类4个类型提取分布信息,结合地面验证坐标点进行精度评价,同时与专家知识分类法比较。结果显示图像膨胀法结合了地物分布特征与光谱信息,在Landsat 8 OLI影像中是可行的,并提高了精度,整体精度为83.01%,Kappa系数为0.77,与专家知识分类法相比分别提高了8.25%、0.11。该研究为快速精确地使用Landsat 8 OLI影像提取树种分布信息具有一定参考。

Extracting Tree Species Distribution with Landsat 8 OLI Data

Investigation the distribution of the tree species is significant in the forestry work, and extracting the tree species information from remote sensing images plays an important role. Changting County is located in southwest part of Fujian Province, China. The topography in Changting is characterized by mountains and hills. The climate is humid. The mean temperature is 18 ℃, and the annual precipitation is 1 742.8 mm. Forests cover large proportion of the area in Changting County, more than 80%. In this study, multispectral Landsat-8 OLI imagery data obtained on 22 March 2016 were used. Normalized difference vegetation index (NDVI) was used to distinguish the distribution of the vegetations. From December 2015 to February 2016, 550 sample points were detected which contained four different tree species. One hundred points were used to build knowledge based system (KBS), and the rest 450 points were taken to verify the accuracy of the classification. For the KBS, different tree species have different means of spectral threshold values. Half standard deviation of the threshold value of the 100 points was taken to build the initial KBS, and then it was modified by the good result of the classification. Image expansion method was used to classify the forests. The results of the classification were validated with ground verification data, and compared with results derived from expert knowledge classification. The results show that image expansion method combining spectral and spatial characteristics of different tree species improve the classification accuracy. The overall accuracy and Kappa coefficient were 83.01% and 0.77, respectively, increased by 8.25% and 0.11 when compared with expert knowledge classification. The presented study could provide a reference for distinguishing tree species for forestry investigation with Landsat 8 OLI data.