遥感影像的分类自校正方法研究

A Study on Self-revised Classification of Remote Sensing Images

遥感图像分类是遥感图像研究的主要内容之一,分类精度高低直接关系到遥感数据的可靠性和实用性。多分类器系统可以提高单分类器分类的精度,但往往要求组成的子分类器分类误差相互独立,子分类器选择困难。支持向量机是新发展起来的一种非参数分类器,其分类原理和传统的基于统计的分类方法不同,表现出一定的独立性。为此本文尝试基于支持向量机和目前使用最广泛的最大似然法,构建一个性能高效且组合方式简单的复合分类器(称为遥感影像分类自校正方法)。同时,为了验证该分类器的性能,在北京市2006年4月27日的SPOT2图像上选择了一个研究区,分别利用最大似然法、支持向量机法和分类自校正方法进行分类对比试验。结果显示分类自校正方法的总体分类精度最高,比最大似然法和支持向量机法分别提高了4.35%和6.6%,而且各种地物类型的分类精度相对最大似然和支持向量机法都有提高。本文提出的分类自校正方法是一种性能高效且操作简单的分类方法。

Remote sensing classification is one of the main contents in remote sensing research. Classification quality is directly related to the reliability and practicability of the data. Multiple classifier systems can improve the classification accuracy than single classifier, although it is difficult to choose the diverse classifiers and make their errors independent. Support Vector Machine newly introduced into the remote sensing fields is a non-oarameter classifier and has different classification principle with commonly used Maximum Likelihood. Hence, in this paper, an effective and combining simple classifiers ensemble （defined as self-revised classification method） is developed based on the Support Vector Machine and Maximum Likelihood. At the same time, a study area was chosen from SPOT2 remote sensing image （received on April 27th, 2006） in Beijing to test the performance of the method presented in this paper. The results indicate that the self-revised classification method has the highest accuracy, the overall accuracy of which is higher respectively 4.35% and 6.6% than that of Maximum Likelihood and Support Vector Machine. Meanwhile the accuracies of all land cover types are increased. The self-revised classification method is accurate and simple.