基于支持向量机无限集成学习方法的遥感图像分类

Remotely sensed imagery classification by SVM-based Infinite Ensemble Learning method

基于支持向量机的无限集成学习方法(SVM-based IEL)是机器学习领域新兴起的一种集成学习方法。本文将SVM-based IEL引入遥感图像的分类领域,并同时将SVM、Bagging、AdaBoost和SVM-based IEL等方法应用于遥感图像分类。实验表明:Bagging方法可以提高遥感图像的分类精度,而AdaBoost却降低了遥感图像的分类精度;同时,与SVM、有限集成的学习方法相比,SVM-based IEL方法具有可以显著地提高遥感图像的分类精度、分类效率的优势。

Support-vector-machines-based Infinite Ensemble Learning method （SVM-based IEL） is one of the ensemble learning methods in the field of machine learning. In this paper, the SVM-based IEL was applied to the classification of remotely sensed imagery besides classic ensemble learning methods such as Bagging, AdaBoost and SVM etc. SVM was taken as the base classifier in Bagging, AdaBoost. The experiments showed that the classic ensemble learning methods have different performances compared to SVM. In detail, the Bagging was capable of enhancing the classification accuracy but the AdaBoost was decreasing the classification accuracy. Furthermore, the experiments suggested that compared to SVM and classic ensemble learning methods, SVM-based IEL has many merits such as increasing both of the classification accuracy and classification efficiency.