植生地域混合スペクトルにおける単位ベクトル化とNDVI関数を用いた植生影響低減手法の提案

抄録

In mineral exploration using remote sensing, vegetation disturbs to extract topsoil spectral features. Previous studies of spectral mixture analysis are using the linear mixture model with end member spectra. However, it is usually difficult to estimate the end member spectra, number of components and mixture ratio. This paper proposes a new method to reduce the vegetation influence from the mixed spectrum in vegetated areas.
In the mixture analysis of topsoil and vegetation, the mixture ratio is the vegetation cover ratio estimated by NDVI. However, NDVI may vary even on the same surface in the case of using the apparent reflectance. To address this issue, we introduce the unit vectorized reflectance (UVR). The UVR achieves to calculate the same NDVI values regardless of the differences in the apparent reflectance and stably express NDVI as vegetation cover ratio. For this reason, the regression curve, called “the reduction curve of vegetation influence”, is obtained from the NDVI and the UVR ratio. The ratio is calculated using the mixed and topsoil’s UVR. The regression model proposed presents therefore a new method to estimate the topsoil spectra.
To evaluate the applicability of this method to satellite images, ASTER and Hyperion data were used. For each band, NDVI and UVR ratio, calculated from the average of UVR at each NDVI value, had a significantly high correlation (R²>0.9). This result suggested the practical applicability of the method to satellite images. Therefore, the method proposed achieves to create vegetation-influence-reduced UVR images. Ground truth observation was conducted in central Chile to compare rock sample’s UVR and the vegetation-influence-reduced UVR. As a result, the vegetation-influence-reduced UVR seemed to be similar to the rock sample’s UVR. Therefore, this method was able to emphasize the absorption bands of minerals.