基于贝叶斯网络分类的遥感影像变化检测

遥感成像过程中,地面、大气等诸多要素的不确定性和波段之间的相关性等原因影响了分类精度,导致变化检测的不准确性。为了提高分类精度往往需要引入先验知识。贝叶斯网络是一种新的数据表达和推理模型,对数据没有严格的正态分布前提要求,通过动态地调整先验概率密度,能有效提高分类精度。以北京通州地区1996-05-29和2001-05-19两个时相的陆地卫星Landsat TM遥感影像为例,介绍了基于贝叶斯网络的分类算法,并在此基础上实现了两个时相遥感影像的变化检测。实验结果表明:基于贝叶斯网络分类算法的后分类比较变化检测方法是遥感影像变化检测的一种新的有效方法。

Remote Sensing Change Detection Based on Bayesian Networks Classifications

The traditional post classification comparison change detection of remote sensing is greatly restricted by the classification accuracy which is influenced by the uncertainty of many factors such as the atmospheric condition,the correlation between the bands of remote sensing data etc.The prior knowledge is often introduced into the classification in order to improve the accuracy.The Bayesian Networks model is a new model for data expression and learning.It has no strict precondition of normal distribution of the input data and can increase the classification accuracy efficiently though adjusting the prior probability density dynamically.The Bayesian Networks classification algorithm was developed in this paper,taking the Landsat TM data in Beijing acquired on May 29~(th),1996 and May 19~(th),2001 as an example in detail and then the change detection using the temporal remote sensing data was realized.The experimental result indicates that the post classification comparison based on Bayesian Network classification algorithm is a newly effective approach for remote sensing imageries change detection.