Lithologic Mapping Using ASTER Imagery and Multivariate Texture

Peijun Li,Haikuo Yu,Tao Cheng
DOI: https://doi.org/10.5589/m09-019
IF: 2.6
2009-01-01
Canadian Journal of Remote Sensing
Abstract:Lithologic mapping using Advanced Spaceborne Thermal Emission and Reflectance Radiometer (ASTER) data and multivariate texture features from different wavelength regions was investigated in an area mainly composed of sedimentary rocks. Nine reflected spectral bands were used, including three visible and near-infrared (VNIR) bands and six shortwave infrared (SWIR) bands. The multivariate texture features are measured by multivariate variogram and derived from multiple bands. The obtained texture features, as additional bands, were combined with spectral data for lithologic classification, and their performances were quantitatively evaluated. The results showed that lithologic mapping using nine ASTER reflected spectral bands alone produced a relatively high overall accuracy (80.3%). However, there are still obvious confusions between different lithologic units in the classification result. Lithologic mapping by combining spectral features and textural features from VNIR or SWIR bands each produced higher accuracy than that using spectral images alone, with an approximately 7% improvement in overall accuracy and 8.5% improvement in kappa coefficient. Moreover, the combined use of these multivariate textural features from both VNIR and SWIR bands and the spectral data in lithologic classification improved the mapping results even further, with a 9.5% increase in overall accuracy and an 11.6% increase in kappa coefficient compared with those using the spectral data alone. The results demonstrated the additional discrimination capability of image texture features for lithologic identification, which significantly reduced the confusion between lithologic units with similar spectral properties but different textural features.
What problem does this paper attempt to address?