基于多特征的高光谱遥感土地利用信息提取

Land use information extraction using multiple features derived from hyperspectral images

【目的】为了降低高光谱遥感数据噪声,提高土地利用分类信息提取精度,探索结合纹理和空间信息的分类方法。【方法】以河南镇平县Hyperion高光谱成像光谱仪获取的高光谱影像为数据源,借鉴决策树分类思想,采用了一种结合光谱、纹理和空间信息的高光谱遥感多特征地类提取方法,先通过提取光谱特征初步提取地类,再分别采用提取纹理特征和基于空间信息的植被提取进行详细地类信息的分层提取,最后,用地面实测样点验证各类土地利用类型的分类精度,比较了用不同方法对不同地类的提取效果。【结果】基于多特征的地类分层提取体系中,采用各波段光谱反射率区分大的地类,再用纹理特征进行光谱差异较小的地类划分,而基于空间信息进行植被的分类。通过结合纹理和空间信息提取方法的总分类精度达86.7%,较最大似然法分类精度提高13.3%。【结论】高光谱与纹理和空间信息相结合的遥感分类方法能有效减小噪声,提高分类精度,可为土地利用分类提取研究提供一定的参考。

【Objective】To reduce noise and improve land use classification accuracy,we need to explore a classification method that combines texture and spatial information.【Method】Using remote sensing images obtained from hyperspectral imaging spectrometers in Henan Zhenping and learning from the classification of a decision tree,a hyperspectral remote sensing classification method combining spectrum,texture,and spatial information was used. The land use type was initially determined by extracting the spectral features,and then hierarchically determined in detail by extracting the texture features and vegetation,which was based on spatial information. Finally,the accuracy of the classification of various land use types was verified by ground test points,and the extraction efficiency of different methods was compared. 【Result】The land-type layered-extraction system combined many characteristics,adopted various band spectral reflectances to classify large ground classes,used textural features to classify ground classes with smaller spectral differences,and then classified vegetation based on spatial information. The total classification accuracy of the method combining texture and spatial information was 86.7%,which was 13. 3% higher than the total accuracy of maximum likelihood classification.【Conclusion】The remote sensing classification method that combined hyperspectral and texture features,and spatial information can effectively reduce noise and improve the classification accuracy. This provides a reference for the study of land use classification.