EM算法的模糊化策略及其在遥感影像分类中的应用

摘要

Among the statistical classification methods of Remote-Sensing images,the distribution of pixel feature sets was always viewed as a mixture of the distributions with different density functions,and the Expectation-Maximization algorithm was one of the most frequently used methods to estimate the parameters of the mixture models.But with appropriate initial parameters,it is very sensitive to classification noises,which would probably slow down the running speed and reduce the accuracy of its results.On this point,a fuzzy strategy of this algorithm was proposed in order to depress the negative influences.The classification experiment on the Remote-Sensing image shows the better characters of our algorithm.%遥感影像的统计分类中,通常都将像点特征的集合视为概率密度函数的混合分布,EM算法是求解这种混合模型参数的一个常用方法。但EM算法在给定合适初值的情况下,对训练数据中的噪声非常敏感,这将严重影响算法的运行效率和求取参数的精度。为了解决这个问题,本文提出了EM算法的模糊化策略,以此来减少噪声在参数学习过程中的影响。对遥感影像的分类实验表明,经过模糊化的EM算法能够更好地完成影像数据的分类。