航空機ハイパースペクトルデータを用いた水稲の生育?収量?タンパク含有率の推定手法の開発

摘要

This paper describes an estimation method for crop conditions using Lasso (Least Absolute Shrinkage and Selection Operator) regression applying to airborne hyperspectral data. Lasso regression, which uses the multiple linear regression analysis with regularization, is suitable for analysis of a fewer measurement dataset with huge number of bands. The conventional estimation method of Lasso regression tends to select the regression model with an excessive number of bands. In addition, noises included in hyperspectral data have the potential to affect the Lasso regression model building. A method proposed in this paper achieves an improvement of the Lasso regression model building by using the moving average applied to hyperspectral data and Akaike's Information Criterion (AIC) applied in the estimation procedure for the regularization parameter. "rnIn the results of estimating the rice growth status using the airborne hyperspectral sensor AISA, Lasso regression with AIC accomplished to select more appropriate bands though the resultant correlation coefficient might be nearly equal to the one given by conventional method using the criterion of mean squared error with regularization. By adopting the normal distribution function with adjustable standard deviations for the window function in the moving average, it is able to exhibit various noise reduction levels. Our method currently proposed demonstrates that the optimal regression model for the hyperspectral data under noisy condition is necessary to have an appropriate moving average.%航空機ハイパースペクトルデータを用いて，水稲のrn生育?収量?タンパク含有率を推定する手法として，rnLasso回帰を検討した。その結果，Lasso回帰におけるrnMSEの評価基準では，明瞭に回帰モデルの精度を評rn価できず，選択されるバンド数が多いことがあった。rn評価基準として赤池情報量基準（AIC）を導入した結rn果，最適な回帰モデルを明瞭に得ることができ，少数rnのバンドで回帰モデルを構築できたことから，その有rn効性を示すことができた。