基于多源遥感数据的第四系覆盖物分类方法研究:以内蒙古旗杆甸子幅1∶5万填图试点为例

Classification of Quaternary Coverings in Desert Grassland Shallow Cover Area Based on Multi-Source Remote Sensing Data:A Case of 1∶50 000 Pilot Geological Mapping in Qigandianzi,Inner Mongolia

遥感图像分类技术对于荒漠草原浅覆盖区第四系覆盖物分类具有重要意义。以内蒙古旗杆甸子幅1∶5万填图试点为例,基于ASTER、GF-2等多源遥感数据,利用植被抑制法、波段比值法、主成分分析以及纹理信息提取等多种方法,充分考虑了多光谱数据的光谱信息和高分辨率数据的形状、空间结构、纹理信息等特征,结合面向对象分类法,对研究区第四系覆盖物进行了分类,并比较分析了不同分类方法的分类效果与精度。结果表明:将波段比值、主成分分析以及纹理分析多种特征作为辅助数据参与分类,其分类效果优于基于单一ASTER数据进行的分类;通过几种不同分类方法的比较分析,发现多特征面向对象分类的总体精度最高,达到85.40%,比多特征传统监督分类的总体精度提高了约11%,分类影像上地物边界清晰。该法分类技术可以为荒漠草原浅覆盖区的地质填图提供相关技术支持。

Remote sensing image classification is of great significance for the classification of Quaternary coverings in desert grassland shallow cover area. The 1∶50 000 pilot geological mapping in Qigandianzi, Inner Mongolia was studied as a case. In order to consider the spectral information of multi-spectral data and the shape, spatial structure, texture information of high-resolution data, the ASTER GF-2 and vegetation suppression, band ratio, principal component analysis, texture information extraction were used and combined with object-oriented classification to classify the quaternary coverings. The classification effect and precision of different classification methods were compared. The result shows that the features of band ratio, principal component analysis and texture analysis utilized in the classification as auxiliary data lead to better classification than the classification based on single ASTER data and the classification accuracy is also significantly improved. The object-oriented classification which considers information of spectrum, space and textureis superior to the traditional supervised classification effect which mainly depends on spectral information. The comparison and analysis of several different classification methods reveal that the overall accuracy of multi-feature object-oriented classification is the highest, reaching 85.40%, which is about 11% higher than that of multi-feature traditional supervised classification. The boundary of the classification image is clear. The classification technique discussed in this paper intends to provide help geological mapping in grassland desert shallow cover area.