基于混合高斯过程模型的高光谱图像分类算法

Hyperspectral image classification based on mixed Gaussian process model

提出了一种基于混合高斯过程模型的高光谱遥感图像分类算法,它不同于传统的基于多元统计的分类方法.为更好利用高光谱遥感图像的高谱分辨率特点,首先将函数数据分析的思想引进高光谱数据的分类问题,分类对象视为像元对应的谱线,故它们是函数型数据.为了有效模拟地物在空间上的分片聚集特性,则将混合高斯分布模型推广到混合高斯过程模型并用于高光谱数据分类算法中.数值实验表明,混合高斯过程模型是处理函数型数据的有效方法.

A novel hyperspectral image classification algorithm based on mixed Gaussian process is proposed,which is different from the traditional classification methods based on multivariate statistics. In order to make full use of the speciality of high spectral resolution of the hyperspectral image, functional data analysis is introduced in the model construction.The image elements of the objects to be classified are treated as spectral curves,hence they are functional data.In order to efficiently model the property of spatially piecewise assembling of objects,the mixed Gaussian distribution model is generalized to the mixed Gaussian process model,which is used in the proposed hyperspectral classification algorithm.Numerical experiments demonstrate that the mixed Gaussian process model is efficient in functional data analysis.