Abstract: Information extraction from remotely sensed imagery consists of two parts：classification and specific class extraction.The accuracy of specific class extraction changes with the partition of the classes in the image.The effect of classes partition on specific class extraction is studied in this paper.A theoretical analysis of the effect is performed based on Bayesian decision rule.Two experiments of specific class extraction are carried out with different classes partitions.The experiments show that different classes partitions result in different accuracy of specific class extraction.A scheme of classes partitioning is presented based on class separability measure.The effectiveness of this scheme is verified by the experiment of specific class extraction.

Key words: feature space, information extraction, specific class

摘要： 遥感影像信息提取包括分类和特定类别提取，特定类别提取精度是随着特征空间的类别划分变化的，研究了类别划分对特定类别信息提取的影响。基于贝叶斯分类器，理论上分析了类别划分对特定类别提取的影响，对不同类别划分的特定类别提取进行实验研究，表明不同类别划分下的特定类别信息提取精度不同。为了确定合适的类别划分，提出基于散布矩阵的类间可分性的类别划分选择方法，并由实验结果进行了验证。

关键词: 特征空间, 信息提取, 特定类别