针对简单贝叶斯分类器(NBC)中“天真”而又苛刻的条件独立假设,本文结合Fisher准则提出了一种在贝叶斯网络中引入隐藏节点的方法,用来松弛或放宽这个过于严格而且不现实的假设条件。隐藏节点的引入可以更好地描述问题,进而更好地解决贝叶斯网络在分类中的应用。实验表明,本文提出的方法可以比NBC获得更高的分类精度和更好的稳定性。在训练样本不多的情况下,平均分类精度比PCA—NBC高3%之多。
In this paper, on the basis of the study of Naive Bayes Classifiers ( NBC), a method, Bayesian Network with hidden nodes, was proposed to relax the independent assumption and applied to texture classification of aerial image. The experiment results demonstrated that the method performed better in overall classification precision than NBC and PCA-NBC.