论文部分内容阅读
遥感影像单类信息提取是一种特殊的分类,旨在训练和提取单一兴趣类别。研究了基于最近邻分类器的单类信息提取方法,包括类别划分和样本选择问题。首先分析论证了最近邻方法提取单类信息只与所选择的样本相关,而与类别划分无关,因此可以将单类信息提取作为二类分类问题进行处理。然后在二类分类问题中,根据空间和特征邻近性选择非兴趣类别的部分训练样本,简化了分类过程。实验结果表明,所提出的方法可以有效实现遥感影像单类信息的提取。
Remote sensing image information extraction is a special kind of classification, designed to train and extract a single interest categories. The single-class information extraction method based on nearest neighbor classifier is studied, including classification of class and sample selection. First of all, it analyzes and proves that the nearest neighbor method extracts the single information only related to the selected sample, but has nothing to do with the classification, so the single information extraction can be treated as the second class classification problem. Then in the second category of classification problems, some training samples of non-interest categories are selected according to the spatial and feature proximity, which simplifies the classification process. Experimental results show that the proposed method can effectively extract single-class remote sensing image information.