基于mean-shift聚类过程的遥感影像自动分类方法

提出了一种稳健的自动分类方法——自适应mean-shift(AMS)算法.该方法基于mean-shift聚类过程,不需要假定数据分布类型,也不需要指定类别的数目,自动化程度较高.自适应mean-shift算法根据数据分布特点自适应地确定带宽的大小,利用采样点估计模式来设计自适应估计器,自适应估计器将每个数据点与不同尺度的核函数联系起来,当核函数满足一定的条件时,AMS迭代过程收敛于极值点(mode),自适应AMS算法是一种归一化的密度梯度估计算法.采用TM影像进行分类试验,试验结果表明:该算法自适应程度高,精度也能满足要求,是一种稳健的自动分类方法.

A automatic classification method for remote sensing image using mean-shift procedure

A robust,auto classifying way named adaptive mean-shift(AMS) was put forward.This method is based on mean-shift clustering,which is not given the data distribution,not given the number of classes,which is high automatic.AMS can adaptively set the width of bandwidth, which designs the adaptive estimator by estimating the sample data,which makes every point related to different kernels varied different bandwidths.When the kernel satisfy certain conditions,the algorithm has a convergence of mean-shift procedure to the nearest stationary point of underlying density function.AMS is based on the estimation of the normalized density method.TM images were used for classifying,and the results indicate that the method is a automatic and robust method,which has better precision.