時系列データの最尤法分類における分類結果の安定化手法

Abstract

For the purpose of improvement of the stability of classification results of multi-temporal images using supervised maximum likelihood classification method, the author proposed a classification method combining the method by unifying training data sets of two temporal images with the method using multi-layer classification images. By the experiment using LANDSAT/MSS image and MOS-1/MESSR image taken almost at the same time, the occupation rate of classification classes was proved to become stable by normalization of two temporal images and unification of both training data sets, although the pixel-wise class coincidence rate was very poor. The classification method using dual multi-layer classification images was proved to achieve very high pixel-wise coincidence rate and also make class occupation rate more stable. The experiment using two or three multi-temporal images proved that the proposed methods can be used effectively in order to obtain stable classification patterns and class occupation rates among the temporal images, which brings a reasonable result for landcover change extraction.