地学空间数据聚类方法研究

本文介绍了地学空间数据迭代聚类的算法原理。根据类别分离度在线性映射变换下的不变性，用Ｋ－Ｌ变换进行地学空间数据的特征提取。在特征空间上两类分布的分离度只依赖于特征值最大和最小的两个特征向量，且其大小依赖于上述两个特征值与某个定值的偏差。据此可设计一个迭代过程，使分类后样本集的特征值都尽可能趋近某个定值，从而达到最佳分类结果。用该算法对乌鲁木齐河流域土壤类型进行了分类研究，取得了较好的效果。

RESEARCH ON CLUSTERING APPROACHES FOR GEO-SPATIAL DATA

This paper presents algorithmic principles for approaching clustering of geo-spatial data.According to the unchangeness of linear mapping transformation of classification divergence,the feature information of geo-spatial data is extracted by K-L transformation.In the feature space,the divergences on two classificational distributions are only dependent on the two characteristic vectors,i.e.,maximun and minimum in feature values,and their varying values depend upon the declinations of the aforementioned two feature values and the certain constant values.On these grounds,an approaching process can be designed so that the feature values in classified sample sets tend to a certain value as much as possible,therefore the best classification results are reached.The test researches on soil categoris in Urumqi River basin are made with the algorithms mentioned above,and better effect is obtained.