講演情報
[1301-12-10]マルチスペクトル衛星画像による詳細な鉱物分類のための観測波長帯と空間分解能の同時ダウンスケーリングに向けて
○グェン ティアン ホアン1、小池 克明1 (1. 京都大学)
司会: 柏谷公希（京都大学）
キーワード：
リモートセンシング、観測波長帯、空間分解能、高分解能化、鉱物分類
Hyperspectral remote sensing is more effective and provides greater accuracy than multispectral remote sensing for identification of surface materials in general. Airborne Visible Infrared Imaging Spectrometer (AVIRIS), a typical airborne hyperspectral sensor, has 224 spectral bands, while the Landsat series has the longest space-based record of Earth’s land and global coverage. We have developed a hyperspectral transformation method, Pseudo-Hyperspectral Image Transformation Algorithm (PHITA) to transform Landsat 7 ETM+ imagery into pseudo-EO-1 Hyperion imagery using correlations between ETM+ and Hyperion band reflectance data. This study extends PHITA to simulate pseudo-AVIRIS image from Landsat 8 OLI image. The pseudo-AVIRIS image has the same number of high-quality AVIRIS bands with a downscaled spatial resolution of 16.4 m, half of the original resolution. The Cuprite area, known as rich in hydrothermal alteration and metals such as gold and copper, was selected for a case study. The resultant pseudo-image was verified by a general appearance, six statistical quality indices, spectral reconstructions, and mineral mapping. Most pseudo-AVIRIS bands have large Pearson’s correlation coefficient (PCC), universal image quality index (UIQI), and structural similarity index (SSIM) than 0.95 and small Root Mean Square Error (RMSE) values mostly than 0.025. These results demonstrate strong correlations between the original and pseudo-images. Spectral reflectance profiles of main endmembers show equivalent behaviors in the original and pseudo-images, except for a small difference in absorption depth. In comparison with the original AVIRIS mineral map, the pseudo-mineral map has the overall accuracy of 68% and Kappa coefficient of 0.6. These results support that PHITA is highly capable of transforming OLI image into pseudo-AVIRIS image. However, before the transformation, an additional spatial downscaling technique of OLI image is necessary to improve the mineral classification accuracy. A more advanced spatial downscaling is our next step.