多時期高解像度衛星画像を対象とした空間的ニューラルネットワーク土地被覆分類

抄録

A neural network (NN) land cover classification model is proposed for multitemporal satellite image classification, which is drivel by co-occurrence matrix as spatial and spectral information source. The proposed method and the two kinds of conventional NN methods were evaluated by using the Landsat TM data set constituting four images observed in four seasons. As the result, the best performance was achived by the proposed model. The proposed model showed 93% of overall classification accuracy which was 4% to 8% higher than that of the conventional NN methods.