基于关联规则的面向对象高分辨率影像分类

以北京市昌平区Geoeye-1高分辨率遥感影像为试验数据,研究了关联规则挖掘和面向对象相结合的地物分类方法。首先探讨了关联分类法的原理,再通过图像分割、特征提取、关联规则挖掘、分类器构建一系列过程实现了基于关联规则的面向对象高分辨率影像分类,最终评估分类精度并与K-近邻法进行了对比。结果表明,该方法具有较好精度,能够在一定程度上摆脱地物分类对于专家知识的依赖。

Object-oriented High Resolution Image Classification based on Association-rule

This paper has explored the method of high resolution image classification by combining association rule mining and object-oriented method.Firstly,according to the theory of Classification Based on Association(CBA),and a modified classifier builder was discussed.Secondly,the object-oriented high resolution image classification was achieved by image segmentation,feature extraction,association-rule extracted and classifier building.After that,Class Association Rules(CARs) was mined by the process of CBA-RG.It was proved that these rules correspond with the features of the ground object.According to the order of "confidence → spectrum complexity → support → generation sequence",a modified classifier was built based on these rules.Finally,we evaluated the precision of the classification result and compared it with the result of K-Nearest Neighbors.The experiment shows that the precision is relatively high and can move away from the dependence on the expert knowledge in a certain degree.