基于改进自组织竞争神经网络的高光谱图像分类  

An improved hyperspectral remote sensing image classification method based on SOFM neural network

作　　者：赵春晖[1] 刘凡[1] 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《应用科技》2009年第8期8-12,共5页Applied Science and Technology

基　　金：高等学校博士学科点基金资助项目(20060217021);黑龙江省自然科学基金资助项目(ZJG0606-01)

摘　　要：针对传统的SOFM网络对高光谱图像分类精度低的缺点,提出了采用模糊积分与神经网络相结合的分类方法.即在改变网络的学习速率函数和邻域函数的前提下,同时对分类结果采用基于模糊积分的信息融合,使分类器之间相互补偿,并用高光谱图像的分类实验进行验证.与普通的SOFM网络和K均值聚类方法相比较,分类效果更好.Aiming at the low precision in remote hyperspectral sensing image classification of traditional SOFM neural network, this paper presents a classification method of combining neural network with fuzzy integral. On the premise of changing the network＇ s learning rate function and neighborhood function, this method of adopts fuzzy integral-based information fuse for the classification results, so that the classifiers compensate each other, and then verify it by hyperspectral remote sensing image classification. The effect of classification is better than the common methods of SOFM neural network and K mean clustering.

关 键 词：SOFM神经网络 模糊积分 高光谱图像分类 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]