基于PCA的面向对象的耕地信息提取方法

Information Extracting for Cultivated Land Using Object-oriented Method of PCA

耕地是人类社会最重要的自然资源之一,高分辨率影像提取耕地有着重要意义和影响。本文以高分一号为研究对象,对高分一号影像的波段进行主成分分析转换(PCA),分析耕地在转换后影像的波段特征,纹理特征以及几何特征,再使用面向对象的方法对影像进行提取地物。相对于最大似然法的精度,经过主成分分析能够增强各种不同地物的可分离性。最大似然法的整体精度是77.05%,kappa系数是0.7246。PCA转换波段的面向对象方法提取整体精度达到了85.76%,kappa系数是0.8290,PCA的方法精度提高了8.71%,kappa系数提高了0.1044。提取耕地的制图者精度平均达到了81.38%,用户精度平均达到了85.09%,提取耕地的精度分别平均提高了13.32%,7.15%。这说明了基于主成分分析的面向对象方法可以有效的进行提取各种地物,可以为提取地物的研究提供支持。

Cultivated land is one of the most important natural resources, and extracting cultivated land from high resolution image is significant. This study was based on 'GF-1' satellite, whose bands were processed by peripheral component analysis(PCA). In this paper,analyze the spectral, texture and space features in the bands, and extract cultivated land by the object-oriented method. The total accuracy for cultivated land information acquired from'GF-1' image data by maximum likelihood method is 77.05% and kappa coefficient o is 0.7246.And, The total accuracy of the object-oriented method of PCA is 85.76%, and kappa coefficient is 0.8290 which was improved by 8.71%, and kappa coefficient was improved 0.1044. The results show the object-oriented method of PCA can be more effective to extract cultivated land information, which is an available and feasible method.