基于Adaboost的高光谱与LiDAR数据特征选择与分类

Feature Selection and Classification of Hyperspectral Data and LiDAR Data Based on Adaboost

提出一种基于Adaboost的高光谱与LiDAR数据特征选择与分类方法。方法首先提取地物的光谱、高程、纹理以及植被指数等特征,接着利用Adaboost评估不同特征的重要度,实现特征选择,最后基于特征子集进行Adaboost分类。甘肃省张掖城区的地物分类实验表明所提方法能选择出有利于类别区分的特征。

Hyperspectral remote sensing can accurately describe the objects’spectral characteristics,and has become an effective way of identifying land covers.However,the hyperspectral imagery does not readily provide the targets’3-D position information which is necessary for recognition of targets with similar spectral signatures and distinct topologies. Supplementation of the hyperspectral images with LiDAR data can compensate the shortage,and then the classification performance can be improved.We propose an adaboost-based feature selection method to integrate hyperspectral images and LiDAR data for classification.The spectral,altimetric,textural features as well as vegetation index are first extracted.The importance of each feature is then evaluated using Adaboost,and the feature subset is produced by discarding features less importance values.The final classification results can be obtained based on the subset.The land cover classification experiments on Zhangye city,Gansu province show that the proposed method can select more useful features for classification.