基于机器学习与高光谱数据的湿地植被物种识别研究

为了探索利用高光谱高空间分辨率遥感数据进行湿地植被物种识别，笔者在分析6 种湿地植被原反射光谱、二阶微分及连续统去除光谱的基础上，利用马氏距离法和相关系数法提取特征波段，并将其作为特征参数参与C5.0 决策树分类与信息提取。结果显示：（1）基于机器学习的C5.0 决策树法总体分类精度为79.87%，Kappa 系数为0.765，与监督分类最大似然法相比，植被信息提取总体精度提高9.95%，Kappa系数提高0.114；（2）机器学习C5.0 决策树法与最大似然法相比，其独特的优势在于对藻类的信息提取精度大大提升，狐尾藻和水蕴草的用户精度提升最大，分别提升了18.67%和15.86%。该方法能够实现湿地植被物种信息的高精度提取，为同类研究提供借鉴，为湿地生态健康评价提供科学与技术上的支持。

Species Identification of Wetland Vegetation: Based on Machine Learning and Hyperspcetral Data

The paper aims to explore the species identification of wetland vegetation by using hyperspectral and high spatial resolution remote sensing data. Based on the analysis of the original reflection spectra, secondorder differential spectra and continuum removal spectra of 6 wetland vegetation species, we extracted the characteristic bands by the methods of Mahalanobis distance and correlation coefficient, and took it as a characteristic parameter to participate in the classification and information extraction of C5.0 decision tree. The results showed that: (1) the overall classification accuracy of C5.0 decision tree based on machine learning was 79.87% , and the Kappa coefficient was 0.765; compared with the supervised classification maximum likelihood method, the overall accuracy of vegetation information extraction increased by 9.95% , and the Kappa coefficient increased by 0.114; (2) compared with the maximum likelihood method, the C5.0 decision tree method based on machine learning had the unique advantage that the information extraction precision of algae was greatly improved, and the user precision of watermifoil and Egeria increased the most, by 18.67% and 15.86%, respectively. The method can achieve high-precision extraction of wetland vegetation species information, provide reference for similar research, and supply scientific and technical support for wetland ecological health assessment.