基于贝叶斯网络分类的土壤盐渍化遥感监测

贝叶斯网络是一种将贝叶斯概率方法和有向无环图的网络拓扑结构有机结合的概率模型。采用贝叶斯网络分类对具有典型干旱特征的库车县土壤盐渍化情况进行监测，首先应用条件独立性测试原理建立贝叶斯网络结构，把研究区遥感数据进行离散化，然后应用贝叶斯定理作为分类原则，将每个像元分为像元最大概率的类别。研究结果表明该方法分类6种地类的整体分类精度达到96％，为该区盐渍地面积、空间分布等特征监测提供了较好的依据。

Remote Sensing Monitoring of Soil Salinization Based on Bayesian Network Classification

The Bayesian network is one kind of probability model, which is based on probability theory and graph theory. The network shows that random variables are nodes and conditional dependencies are edges in a directed acyclic graph. In this paper, soil salinization of Kucha county in arid region was monitored with the Bayesian network classifier. Firstly, using the conditional independence test built the Bayesian network structure, and desensitizes remote sensing data of study area. Secondly, applying Bayesian network theorem as classification principle classifies every single pixel to a pixel of maximum probability sort. The results indicate that the classification accuracy is above 95%, which provides reasonable base for monitoring the areas and spatial distribution of salinization soil.