10:45 〜 12:15
[MGI31-P03] ハイパースペクトル衛星画像のバンド選択による鉱物含有率推定法の精度評価
キーワード：リモートセンシング、Hyperion衛星画像、反射率スペクトル、スペクトル分離、Cupriteフィールド（アメリカ合衆国）
Remote sensing using Earth observation satellites has been widely applied to detect signs of metallic deposits and hydrothermally altered minerals, because it can detect them in wide areas. With many observation wavelengths (bands) and a large amount of information, Hyperspectral satellite imagery can identify surface materials accurately. However, the areas covered by hyperspectral satellite imagery is still limited and much less than multispectral imagery with only several bands. Therefore, identification accuracy of minerals by multispectral imagery needs to be equivalent to that by hyperspectral imagery. For this, this study aims to develop a band selection method that enables to estimate mineral content with high accuracy. A feature selection technique, which has been developed in the field of machine learning, was applied to our method.
As a case study, one scene of EO-1 Hyperion image, a representative hyperspectral satellite imagery, was used for mapping mineral distributions in the Cuprite area in Nevada, USA where the surface minerals have been studied in detail through geological surveys and remote sensing analyses. Therefore, this area is suitable for verifying the accuracy of the proposed method. Four spectral unmixing methods were targeted to verify the mineral estimation accuracy by the band selection. Four main minerals: alunite, calcite, muscovite, and kaolin were selected as the endmembers. As a result, the optimal band selection method was specified, which brought the high mineral estimation accuracy almost equivalent to the result by the hyperspectral image analysis. Consequently, our band selection method was demonstrated to be effective.
As a case study, one scene of EO-1 Hyperion image, a representative hyperspectral satellite imagery, was used for mapping mineral distributions in the Cuprite area in Nevada, USA where the surface minerals have been studied in detail through geological surveys and remote sensing analyses. Therefore, this area is suitable for verifying the accuracy of the proposed method. Four spectral unmixing methods were targeted to verify the mineral estimation accuracy by the band selection. Four main minerals: alunite, calcite, muscovite, and kaolin were selected as the endmembers. As a result, the optimal band selection method was specified, which brought the high mineral estimation accuracy almost equivalent to the result by the hyperspectral image analysis. Consequently, our band selection method was demonstrated to be effective.