基于判别空间条件熵加权的土地覆盖分类方法研究

针对遥感专题类别信息的机理问题,从土地覆盖参考数据的偏差程度对分类精度的影响角度,提出了一种基于判别空间条件熵加权的土地覆盖分类方法。引入判别空间模型概念,基于此模型生成土地覆盖数据类别,并分析了土地覆盖信息类别与数据类别的语义偏差出现的深层次原因;计算信息类别与数据类别的对应关系矩阵,据此得到二者的条件熵,实现对土地覆盖信息类别与数据类别的语义偏差的量化;根据信息类别与数据类别的条件熵计算修正判别变量的权重因子,实现基于判别空间条件熵加权的土地覆盖分类。采用一景SPOT-5影像进行分类实验,并利用同一地区的Landsat 5TM影像进行方法验证。实验表明,条件熵加权修正方法使土地覆盖分类精度有了显著提高,并对不同分辨率的遥感影像具有适用性。

A Weighted Method Using Conditional Entropy for Land Cover Mapping Based on Discriminant Space Model

In allusion to problems on the mechanism of thematic category information of remote sensing,and on the consideration of the semantic bias existed in different land cover reference data which has large influence on the classification accuracy,a weighted method using conditional entropy for land cover mapping based on discriminant space model is suggested in this paper.Discriminant space models,according to which land cover with respect to data classes can be generated,are introduced.Based on these models in discriminant space,the deep seated reasons of the semantic bias existed between land cover with respect to information classes and data classes are analyzed.Correspondence matrix between land cover with respect to information classes and data classes is generated,and the conditional entropies of land cover types with respect to information classes and data classes are also obtained,which realize the quantification of the semantic bias existed between land cover with respect to information classes and data classes.Then,weight factors to realize discriminant variables correction are calculated,and land cover mapping through the weighted method using conditional entropy based on discriminant space model is proposed.A SPOT-5 image is selected and used to make classification experiments,and a Landsat 5 TM image in the same region is used to validate the applicability of proposal method.Experimental results show that the proposal method is an effective approach to improve land cover mapping quality and is applicable for remote sensing images with different resolution.