基于随机森林特征变量优化的湿地植物分类与密度反演

Classification and Density Inversion of Wetland Vegetation Based on the Feature Variables Optimization of Random Forest Model

摘要:

以长江口滨海湿地为研究区域，采用随机森林算法对滨海湿地植被进行分类。在提取Landsat?8 OLI影像植被指数和水体指数的基础上，提出利用植被指数季节差值对模型进行特征变量优化，分析了长江口滨海湿地植物群落分布的空间特征。以所占面积最大的互花米草（入侵物种）为例，采用多元线性回归模型结合实地测量数据，估算了秋季的互花米草植物密度的空间特征。提出的多时相遥感数据结合随机森林特征变量优化方法，可以较为便捷地提取长江口湿地3种优势物种的空间分布特征，与最大似然法相比，分类精度有较大提高，总体分类精度由78.35%提高至87.55%，Kappa系数由0.72提高至0.84。该方法适用于存在“异物同谱”问题的湿地植物群落研究。

Abstract:

Taking the coastal wetland of the Yangtze River Estuary as the research area， the random forest model was used to classify the vegetation of wetland. Besides the vegetation index and water index extracted from Landsat-8 OLI image， the seasonal difference value of vegetation index based on plant phenology characteristics was proposed in this paper as the optimization of feature variables to analyze the spatial characteristics of vegetation distribution in the coastal wetland of the Yangtze River Estuary. Bassd on the vegetation classifications， the multiple linear regression model combined with the field data was used to estimate the vegetation density of the Spartina alterniflora， which was an invasive species occupying the largest area. It is indicated that the proposed multi-temporal data combined with the optimization of feature variable of random forest model can be used to conveniently analyze the spatially vegetation distributions in wetland . Compared with the maximum likelihood classification method， the proposed method in this paper greatly enhances the classification accuracy with an overall accuracy and Kappa coefficient of the classification results increasing from 78.35% and 0.72 to 86.02% and 0.82， respectively. The proposed method is proved to be applicable for solving the problem of “same spectrum for different surface features” in the study of wetland plant community distributions.