本研究使用自行開發之非監督式兩階段模糊及機率模式作為河川空間污染分類的研究，目前水質遙測分類大都使用監督模式，但監督式模式前提需建構一個多樣且代表性的學習樣本，相對的非監督模式毋需準備學習樣本，只需要少部分現場水質污染監測分類值作為最終判斷的參考，且少了監督模式學習建模之複雜性；本研究模式亦改善傳統非監督模式需要預設資料分類數的缺點，模式採自動最佳化分類結果，因河川污染是由多種不同水質綜合而成，分類數也隨不同河川有所差異，故不預設模式分類數，藉最佳化分類衛星遙測影像河川水體之污染情形。另外為提高對河川水質污染分類之精確度，透過篩選輸入最佳光譜變量組合，可進一步建立一套遙測河川表面水質空間污染的分類系統。

This story uses a two unsupervised fuzzy and probabilistic clustering method in order to research how predict the space of rivers is polluted. The river pollution includes many kinds of different water quality. But it is to need to possess various and representative study sample to utilize the supervised type to classify. Unsupervised method does not needing to study samples, but need local value as monitoring and classifying the basis finally. This method improves many traditional unsupervised methods that have been set up the number of clusters, and the automatic optimization classification. It is classing the pollution of river in the image of the satellites. In addition improve the accuracy predicted to the pollution of river, input the best variable association that is screened, in order to set up the prediction system that the space of river pollution in the image of the satellites.