论文部分内容阅读
利用遥感数据开展森林资源树种的分类对森林资源的监测、森林可持续经营及生物多样性研究都有重要意义。该文以江苏南部丘陵地区的北亚热带天然次生林为研究对象,利用Li CHy(Li DAR、CCD、Hyperspectral)集成传感器同期获取的高分辨率和高光谱数据,进行冠幅识别和多个层次的树种分类:首先,对高分辨率影像进行基于边缘检测的多尺度分割,提取出单木冠幅;其次,对高光谱影像进行特征变量提取,并对提取出的特征变量利用信息熵原理选取优化特征变量;然后,分别利用全部特征变量和经优化的重要特征变量对森林树种及森林类型进行预分类;最后,在预分类结果中加入单木冠幅信息对森林树种及森林类型进行重分类,并分析分类结果的精度。研究表明:1)利用全部特征变量进行4个典型树种分类时,总体精度为64.6%,Kappa系数为0.493;而针对森林类型的分类精度为81.1%,Kappa系数为0.584。2)利用选取的优化特征变量分类精度略低于利用全部特征变量的分类精度,其中对4个典型树种分类时,总体精度为62.9%,Kappa系数为0.459;而针对森林类型的分类精度为77.7%,Kappa系数为0.525。通过集成传感器同期获取的高分辨率和高光谱数据可以有效地进行北亚热带森林的树种分类及森林类型的划分。
The use of remote sensing data to classify forest tree species is of great significance to forest resources monitoring, sustainable forest management and biodiversity research. In this paper, the northern subtropical natural secondary forests in the hilly areas of southern Jiangsu Province were selected as research objects. The high resolution and hyperspectral data acquired by Li CHy (Li DAR, CCD, Hyperspectral) integrated sensor simultaneously were used to identify the crown and multi-level tree species Classification: First, multi-scale segmentation based on edge detection is applied to high-resolution images to extract single-wood crown; Secondly, feature variables are extracted from hyperspectral images and entropy principle is used to select the optimal features Then, the forest tree species and forest types are pre-classified by using all the feature variables and the optimized important feature variables. Finally, single-tree crown information is added to the pre-classification results to re-classify forest tree species and forest types Analyze the accuracy of the classification results. The results showed as follows: 1) The overall accuracy was 64.6% and the Kappa coefficient was 0.493 for all the 4 typical tree species, while the classification accuracy for forest type was 81.1% and Kappa coefficient was 0.584.2. The classification accuracy of feature variables was slightly lower than the classification accuracy using all feature variables. Among them, the overall accuracy was 62.9% and the Kappa coefficient was 0.459 for the four typical tree species, while the classification accuracy for forest types was 77.7% and Kappa coefficient was 0.525 . The classification of tree species and the classification of forest types in the northern subtropical forests can be effectively carried out by integrating the high-resolution and hyperspectral data acquired by the sensors over the same period.