基于光谱特征的湿地湿生植物信息提取研究

Information Extraction of Wetland Aquatic Vegetation Based on Spectral Characteristics

基于湿生植物光谱特征分析,采用面向对象分类方法提取湿地典型植被信息。在大量的野外实地调查基础上,确定提取的典型植物类型为芦苇(Phragmites australis)、长芒稗(Echinochloa crusgallii)和睡莲(Nymphaeatetragona);采集湿地优势植物光谱数据,将优势植物与提取对象芦苇、长芒稗和睡莲的光谱特征进行相关性分析,获取物种间区分性好的波段及波段组合,参与影像分割权重的设置;根据典型植被的分布特征,确定面向对象分割尺度(其中芦苇的分割尺度为50,长芒稗的分割尺度为20,睡莲的分割尺度为100)。通过研究发现:基于光谱特征分析的面向对象的分类精度为96%,而未利用光谱特征的面向对象的分类精度为87.3%,传统监督分类精度仅为82.3%。证明在面向对象提取前对植物光谱特征分析得到区分性好的波段及波段组合参与分割,对提高分类精度起关键作用。

Based on a spectral characteristic analysis of typical wetland plants, the authors used the object - oriented classification method to extract such specific plants in Hanshiqiao area of Beijing as Phragmites australis, Echinochloa crusgallii and Nymphaea tetragona on the basis of a lot of survey work. First, the authors collected spectral data of these three typical wetland plants in the field, which constituted the basis of spectral correlation analysis and could help make classification between different species. The correlation analysis results of highly distinct spectral band combination and vegetation indexes with remote sensing images were involved in segmentation weight assignment. Second, based on the distribution of typical plants, the authors decided the split - scale of object- oriented （Phragmites australis split - scale being 50, Echinochloa crusgallii split - scale 20, and Nymphaea tetragona split - scale 100）. A comparison of different classification methods shows that the classification accuracy of the object - oriented method based on spectral features is about 96% , and the classification accuracy is 87.3% in case it is not based on spectral characteristics. Traditional supervised classification accuracy is only 82. 3%. The results show that the object -oriented classification method with spectral feature analysis is effective in the information extraction of Hanshiqiao wetlands typical plants, and the differentiation between spectral bands as well as band combinations will play a key role in that it can highly improve the accuracy of classification.