局部信息保持极限学习机的遥感图像分类

REMOTE SENSING IMAGE CLASSIFICATION WITH LOCALITY INFORMATION PRESERVING EXTREME LEARNING MACHINE

极限学习机ELM(Extreme learning machine)以其简单快速和良好的泛化能力在模式识别和机器学习领域得到了广泛的应用。近年来,研究人员将其应用到高光谱遥感图像分类问题中。然而,由于数据样本有限,极限学习机及其相关技术在遥感图像中存在数据学习不充分的问题。流形学习算法揭示了数据内在的几何结构信息。根据遥感图像的特点,基于流形学习的思想,将遥感图像数据样本的流行结构引入到ELM模型中,提出一种基于局部信息保持极限学习机LPKELM(locality information preserving extreme learning machine)。为了验证所提算法的有效性,使用两个高光谱遥感图像数据集进行实验。实验结果表明,LPKELM的分类性能优于SVM、KELM、KCRT-CK和MLR算法。

Extreme learning machine(ELM)has been widely used in pattern recognition and machine learning with its simple,fast,and good generalization abilities.In recent years,researchers have applied ELM to the area of hyperspectral remote sensing image classification.However,due to the limited data samples,ELM and its related technology have insufficient data learning in remote sensing images.Manifold learning method reveals the inherent geometric structure of data.According to the characteristics of remote sensing images and the idea of manifold learning,we introduced the manifold structure of remote sensing image data samples into ELM model,and proposed a local information preservation extreme learning machine(LPKELM).In order to verify the effectiveness of the proposed algorithm,two hyperspectral remote sensing image datasets were used to conduct experiment.Experimental results show that LPKELM is better than SVM,KELM,KCRT-CK and MLR algorithms in classification.