Abstract: BNs classifications for remote sensing image is divided into three courses include preprocessing, the BNs model construction and image classification. Preprocessing is the foundation of the other two steps. The preprocessing step has large effect on the whole classification results. The goal of preprocessing is to extract the information high-efficiency and exactly and to eliminate disturbance from image features. This paper introduces the process about data preprocessing technique and the algorithm implementation. Because of the characteristic of remote sensing data and BNs method, the preprocessing is divided into two parts, one is spectrum space segmentation, and the other is mutual information computation. The principles and the algorithms of two parts are introduced. It is very important for the development of preprocessing technique.

Key words: Remote sensing image, Classification, Bayesian networks, Preprocessing, Implementation

摘要： 遥感图像BNs分类分为预处理、BNs模型构建和分类3个前后联系的过程。其中对预处理技术是后面两个步骤的基础，其算法实现过程对分类结果的影响很大，预处理的目标是高效、准确地提取图像分类所需要的重要特征，剔除干扰因素。该文介绍了一种简单数据预处理技术和其算法实现过程。针对遥感数据的特点和BNs方法的需要，将该预处理过程分成波谱空间分割和关系信息计算两个部分，分别介绍了两部分的原理并给出了实现的算法。对于遥感数据分类预处理技术的研究和实现具有很强的借鉴作用。

关键词: 遥感图像, 分类, 贝叶斯网络, 预处理, 实现

CLC Number: 

• TP391