一种多特征结合的遥感图像分类方法

摘要

As the basis of remote sensing image analysis and comprehension, remote sensing image classification is one of the main topics in remote sensing image research.To improve classification effect, the integrated use of multiple features is usually required in the classifica-tion.In this paper, we propose a new classification method which is based on feature level data fusion.After extracting the multiple image space features and spectral features, we feed each type of feature into classifiers.The probability outputs of these classifiers are jointed to form the middle level features, and the classification is applied once again.The method effectively avoids the scale problem in direct conjunction of multiple features.Experiments on two classical data sets, that is, Indian93 and Flightline C1, show that the method has obvious advantage.%遥感图像分类是遥感图像分析和理解的基础，是遥感图像研究中的重要内容之一。为提高分类效果，遥感图像分类中通常需要综合运用多种特征。提出一个新的基于特征级融合的遥感图像分类方法。将多种图像空间特征和光谱特征分别作为分类器的输入，将各分类器的概率输出拼接起来作为中间层特征再进行分类。该方法有效避免了多特征直接拼接存在的尺度问题。在In-dian93和Flightline C1两个数据集上的实验结果表明该方法具有一定优势。