GF-1影像和OLI影像协同土地利用模糊分类方法研究

针对高分辨率遥感数据进行土地利用类型分类时出现的“同谱异物”现象,以及中分辨率遥感数据划分土地利用类型时受空间分辨率限制产生的“混合象元”问题,本文以高分一号数据(GF-1)和Landsat-8数据(OLI)为例,提出了一种协同利用高分辨率遥感数据和中分辨率遥感数据进行土地利用类型模糊分类的方法。首先,利用主成分变换的方法分别对GF-1纹理信息和OLI光谱信息进行压缩和增强,并将增强后的纹理信息和光谱信息进行特征协同;然后,根据各地物类型的光谱、纹理特征,对特征协同数据进行60、80、100共3个尺度的分割;最后,根据地物类型间的光谱特征和纹理特征的差异,构建各地物类型的模糊逻辑隶属度函数,实现对影像土地利用类型的模糊分类。实验结果表明,主成分变换的方法有效地将研究区GF-1和OLI数据的光谱、纹理信息压缩、增强,为面向对象分类中分类特征的选取提供了一种思路;同时,本文方法成功划分了研究区土地利用类型,并获得了较高分类精度,总体分类精度达到93.52%,对其它高空间分辨率与高光谱分辨率遥感数据协同分类研究具有一定借鉴意义。

The Research on the Method of Combining Images of GF-1 and OLI for FuzzyClassification of Land use

In order to improve the phenomenon that different objects perform the same spectral characteristics in land use mapping of high spatial resolution data and the “mixed pixel” problem caused by lower spatial resolution in land use mapping of medium spatial resolution data, this study took GF-1 and OLI as a case and proposed a method of combining high spatial resolution data and medium spatial resolution data for fuzzy classification of land use. Firstly, texture information of GF-1 and spectral information of OLI were compressed and strengthened by principal component analysis (PCA), respectively. Compressed texture information of GF-1 and compressed spectral information of OLI were layer stacked. The combined data of three bands feature was received. Then, the feature combined data was segmented into three different levels of 60, 80, 100 based on texture and spectral characteristics of the different land use types in feature combined data. Finally, the fuzzy logic membership functions of the land use types were built based on texture and spectral difference of the different land use types. In this way, the fuzzy land use classification of the study area was carried out. Results shows that the PCA method compressed and strengthened GF-1 and OLI of study area effectively and the proposed method classified the land use of study area successfully receiving a high total accuracy of 93.52%. The method proposed in this paper offered a new idea for classification feature selecting in object-oriented classification and had some significance for other classification research of combining high spatial resolution data and high spectral resolution data.