多光谱遥感图像BP网分类器学习样本选取法的研究  

LEARNING SAMPLE SELECTION IN MULTI-SPECTRAL REMOTE SENSING IMAGE CLASSIFICATION USING BP NEURAL NETWORK

作　　者：于秀兰[1] 钱国蕙[1] 周建林 贾晓光[1] 

机构地区：[1]哈尔滨工业大学航天电子与光电工程系,黑龙江哈尔滨150001

出　　处：《红外与毫米波学报》1999年第6期449-454,共6页Journal of Infrared and Millimeter Waves

基　　金：国防基金!(编号Y96- 01)

摘　　要：通过分析多光谱遥感图像学习样本在光谱空间不同位置对BP神经网络分类器分类精度的影响,提出基于x2 分布的学习样本选取方法,并应用于TM 图像分类.对6 种地物应用不同样本选取法训练的BP网分类器和Bayes分类器的分类结果比较表明:本方法分类精度明显高于Bayes分类器和其它样本选取法得到的BP网分类器。Through analyzing the influence of the learning samples’ location in the spectral space on the accuracy of multi spectral remote sensing image classification using BP neural network, a method for learning samples selection based on x 2 distribution was presented and used in TM image classification. The classified results of the 6 ground objects with BP classifier using different learning samples selecting methods and the Bayes classifier show that the BP classifier with the presented learning samples selection method can not only reduce the number of learning samples greatly which leads to shorter learning time, but also improve the classification accuracy compared with the existing methods.

关 键 词：学习样本选取 BP神经网络 多光谱遥感图像 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]