一种建立在CPN基础上的分类方法在遥感影像分类中的应用研究

Application of a classification method based on counter propagation neural network in remote sensing image classification

对偶神经网络利用了自组织映射近似函数的一种新的映射神经网络,其结构组合了Kohonen的自组织映射和Grossberg的外星(Outstar)结构,网络结构相对简单。本文以对偶神经网络分类方法原理为基础,研究了一种理想化的分类方法,并以MATLAB平台为基础对遥感影像进行分类处理,实验结果表明,其分类总精度为94.17%,分类精度较传统监督分类结果有所提高。

Counter Propagation Neural Network is one kind of new mapping neural network based on self-organization mapping approximate function. Its structure combines Kohonen self-organization mapping and Grossberg outside star(Outstar)structure. The structure of the network is relatively simple. Based on the classification theory of Counter Propagation Network, an idealistic classification method is researched in this paper. Image processing is finished in MATLAB. An experiment indicates that its total precision of classification is 94.17%, which is higher than that of supervised classification method.