基于混合核函数的自组织神经网络遥感图像分类  

Hybrid-kernel SOM neural network and its application in remote-sensing image data classification

作　　者：尹汪宏[1] 李朝峰[1] 张俊本[1] 王正友[2] 

机构地区：[1]江南大学信息工程学院,江苏无锡214122 [2]江西财经大学信息管理学院,江西南昌330013

出　　处：《计算机工程与设计》2009年第2期388-391,共4页Computer Engineering and Design

基　　金：国家自然科学基金项目(60665001)

摘　　要：自组织神经网络SOM作为一种无监督学习的竞争式网络,已经得到了广泛的应用,它通过对输入信号的竞争学习,将样本划为不同的类别,但其分类效果常很难令人满意。提出了一种基于混合核函数的SOM神经网络改进方法,并和传统的SOM网络进行了对比,Iris数据和Wine数据的分类实验表明,该方法可以明显改进SOM网络的分类效果。然后对某地Landsat卫星遥感图像数据进行分类实验,实验结果表明,与传统的SOM网络、基于多项式核的SOM网络以及基于RBF核的SOM网络相比较,基于混合核函数的SOM神经网络方法的分类效果有较明显的提高。Self organizing maps （SOM） is widely used as a unsupervised learning network, passing the competition study toward the importation signal, it classifies the sample as different category, but its classification result often can hardly give satisfaction. An Hybrid- kernel improved method for self-organizing map （SOM） neural network is presented and a camparison with traditional SOM neural network is made. Iris data and wine data classification experiments show that new method may efficiently enhance the classification accuracy for SOM neural network. Then the classification experiment is finished to landsat remote-sensing image data in certain place. The ex- perimental result indicates that the classification result of SOM network based on Hybrid-kernel function has a more obvious exaltation, comparing with traditional SOM network, the SOM network based on polynomial-kernel function and SOM network based on RBF- kernel function.

关 键 词：混合核函数 自组织映射 神经网络 模式分类 自组织神经网络 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]