分析了目前遥感影像的统计分类、神经网络分类及基于符号知识的逻辑推理分类方法的优缺点.以GIS为平台,构建了多源空间数据库,将数据挖掘的思想和方法引入遥感影像分类中,提出了面向分类规则挖掘的遥感影像分类框架.针对遥感光谱数据及其他空间数据的特点,定义了连续属性样本分类概念和分割点评价指标,提出了一种新的连续属性样本分类规则挖掘算法.选择一个试验区,采用该算法分别对遥感光谱数据、遥感光谱和DEM数据相结合的数据进行分类规则挖掘、遥感影像分类和分类精度比较.结果表明：（1）该算法具有较高的分类精度;（2）加入DEM等与分类相关的其他空间数据可以提高遥感影像的分类精度.通过挖掘分类规则进行遥感影像分类,扩展了基于知识的逻辑推理分类方法中知识获取渠道,提高了分类规则获取的智能化程度.新的连续属性样本分类规则挖掘算法,扩展了归纳学习算法对连续属性样本分类的适应性.

The advantages and disadvantages in remote sensing image classification methods at present are analyzed,in cluding maximum likelihood classification((MLC);) Artificial neural networks((ANN)) classification;Logic reasoning classification based on symbol and rules.In order to advance remote sensing image classification,multi-resouce spatial database is set up by means of GIS,the thought and concept of data mining are imported,and the remote sensing image classification frame is brought forward.The concept of classification,and the index of judging splitting point on continuous valued attributes samples are defined,a new algorithm is proposed,which mines classification rules on the continuous valued attributes spatial database.A trail area is selected,the two projects are put in practice,which are based on Spectra Data,combining of Spectra data and DEM data respectively,mining classification rules,classifying remote sensing image by the way of the algorithm,and comparing their classification accuracy.Trail outcome shows that:(1) the classification accuracy of the algorithm is good;(2) integrating DEM and other correlative data into the spatial database can advance classification accuracy.The way of mining classification rules being applied to remote sensing image classification broaden the channel of knowledge discovery in logic reasoning classification based on symbol and rules,advance the capability of discovering classification rules automatically.The new algorithm mining classification rules on the continuous valued attributes enlarge the adaptability of inducing learning algorithm to the classification of the continuous valued attributes samples.