衛星画像における非線形モデルと近傍画素フィルタを適用した混合画素推定法

概要

論文の詳細を見る
The several papers reported that the errors in land cover classification have been generated around the edge of categories especially. The mixed pixel that includes the multiple categories is one of the error factors. All remotely sensed images include the mixed pixel, particularly the influence of the mixed pixels in the middle- or low-resolution images is more serious.In this paper the method using the nonlinear mixed pixel model and the filtering of the neighborhood pixels have been proposed. In order to verify the effectiveness of the proposed method, the six simulations have been made; two patterns without the proposed filter are for the linear and for the nonlinear models, and four patterns with the proposed filter are for the proposed nonlinear method.The results of the simulations have been quantitatively evaluated by calculating the correlation coefficients between the simulated results obtained from Landsat-7/ETM+ image and the model answers made by QuickBird-2 images. It was confirmed that the correlation coefficients of the proposed nonlinear model and the proposed filter were improved than that of the other results.
社団法人電気学会の論文