摘要
以黑龙江省漠河县为研究区域,采用陆地卫星-8遥感影像为数据源,结合影像的光谱信息和数字高程模型辅助数据,分别采用最大似然分类法(MLC)和随机森林模型法(RFM)对研究区森林植被进行分类,并分析和评价光谱特征变量对模型的重要性、2种分类方法对森林植被类型分类的适用性。结果表明:随机森林分类方法的总体分类精度为81.65%、卡帕(Kappa)系数为0.812。与传统的MLC方法相比,RFM法均提高了3种森林类型的生产者精度和使用者精度,其中针阔混交林精度提高最多。通过分析特征变量的重要性,发现高程、归一化植被指数、红光波段、近红外波段、短波红外波段对模型分类精度有较重要的影响。说明随机森林模型方法结合多源信息是森林植被类型遥感分类的一种有效手段。
Taking Mohe County of Heilongjiang Province as study area,with the Landsat8 remote sensing images as data source,and the spectral signatures of the image and DEM as additional data,we classified the types of forest vegetation based on classification method of the Maximum Likelihood Classifier（ MLC） and random forest model（ RFM）. We analyzed the importance of the characteristics variables of the spectral for the model,and evaluated the suitability of two methods in forest vegetation classification. The overall classification accuracy and the Kappa coefficient of RFM were 81.65% and 0.812,respectively. Compared with the MLC method,the RFM method improved the user accuracy and production accuracy for three forest types. By analyzing the importance of the variables,elevator,NDVI,red band,NIR band,and short wave IR band played an important role in classification accuracy. Therefore,the RFM based on multiply types of data is a fast and effective method in the classifications of forest vegetation types.
出处
《东北林业大学学报》
CAS
CSCD
北大核心
2016年第6期53-57,74,共6页
Journal of Northeast Forestry University
基金
科技部科技基础性工作专项项目(2013FY111600-7)
关键词
随机森林模型法
陆地卫星-8遥感影像
森林植被分类
Random forest model（RFM）
Landsat8 remote sensing image
Forest vegetation classification
作者简介
张晓羽，女，1989年2月生，东北林业大学林学院，硕士研究生。E—mail：949205203@qq．com。
通信作者：赵颖慧，东北林业大学林学院，副教授，E-mail：zyinghui0925@126．com。
