商空间理论下面向对象的遥感影像分类

Object-oriented Classification Method Based on Quotient Space Theory

传统的分类方法仅仅基于像素光谱特征,不适合于高分辨率遥感影像。本文提出了一种新的基于商空间理论,面向对象的高分辨率遥感影像分类方法,即综合云模型、模糊支持向量机和决策树的分层合成分类技术。针对决定分类效果的两个因素,影像分割和分类算法,分别做出了一些改进。第一,本文提出了一个自适应的基于云模型的区域增长分割策略。第二,本文提出了一个新的基于商空间理论,结合模糊支持向量机和决策树的分层合成分类技术。从实验结果来看,本文提出的方法,在分割效果上基本能满足人眼的视觉要求,在分类精度上比传统的分类方法有更高的准确性,而且能实现分类过程的自动化。

Traditional classification methods only based on spectrum features of pixels are not suitable for high-resolution remote sensing image. A new object-oriented classification method is proposed based on quotient space granular theory, cloud model, fuzzy support vector machines and decision tree algorithm. Considering that the two factors which determinate the accuracy of classification results are image segmentation and classification algorithm, we make some improvements and our work includes the following aspects. Firstly, an adaptive region growing method is proposed based on the Cloud Model. Secondly, a hierarchy-synthesis classification technique is proposed based on quotient space granular theory, fuzzy support vector machine and decision tree. A new algorithm is designed to calculate the fuzzy membership of sample. By combining FSVM with ISODATA, we not only improve the quality of training sample, but also make objects classified automatically. More importantly, the strategy of hierarchical classification makes different categories discriminated more effectively. By using synthesis theory of quotient space, we can get the final classification result from classification results of multi-granular quotient spaces. According to the experiment result, our method can not only improve the accuracy of classification result and satisfy human eye, but also make objects classified automatically.