特征权重优化高分辨率遥感影像模糊分类研究

The Research on Weight Optimization in Multi-feature Fuzzy Classification of High Resolution Remote Sensing Image

在针对SPOT5等高分辨率遥感影像的面向对象模糊分类过程中,一般对影像对象的特征赋予相同的权重。为了体现不同特征对分类作用的差异,本文在分类时根据特征的重要与否,对参与分类的特征赋予不同的权重,提高重要的、区分度好的特征的权重,降低次要特征的权重。以北京市昌平区的SPOT5影像为例,利用多特征模糊分类和经过权重优化的多特征模糊分类进行分类对比实验。实验结果表明,经过特征权重优化的分类与权重相同的分类结果相比,分类总精度由原来的86.3%提高到了92.6%,Kappa系数由原来的0.8096提高到了0.8947。结果表明,经过权重优化的多特征模糊分类有助于提高模糊分类法的分类精度和适用性。

In the process of fuzzy classification of high resolution remote sensing image,people are used to give the same weight to all image objects＇ features.This paper attempts to raise the weight of important features and to bring down the weight of subordinate features in the image classification.We classify SPOT5 image data of Changping area in Beijing using two methods in order to see the contrast.One method is traditional multi-feature fuzzy classification and the other is the method that has been weight optimized.Compared with the first method,the second method obtained better results.The total classification precision increased from 86.3% to 92.6% and the Kappa coefficient raised from 0.8096 to 0.8947.The results show that weight optimization in multi-feature fuzzy classification contributes to improve the accuracy and applicability of multi-feature fuzzy classification.