衛星影像(Satellite images)係源自光譜反應值，為諸多地物（類）的綜合反映，故無論其解析度如何細緻皆不可避免地會產生混合像元(Mixed pixels)。衛星影像製圖的方法之一即是利用像元純化(Pixel unmixing)技術。所謂像元純化，簡言之，係運用光譜的多餘觀測量，以解析得像元內地物（類）混成比(Sub-pixel proportions of land cover)，因此又稱為次像元解析術。次像元解析術為遙感影像製圖領域中的重要方法之一，本研究旨在探討與該方法相關的數學與統計方法，分析其中利弊得失，並加以綜合、運用及發展出一套更具效能的運算模式，以提昇影像製圖之精度，期能為遙感探測學(Remote sensing)與地理資訊科學(Geographic information science)兩領域的整合應用提供實質助益。

A common problem in remote sensing is that the limited spatial resolution of the scanner inevitably leads to mixed pixels at class boundaries. This paper tries to propose some robust statistics for pixel unmixing to model the spectral mixture and classify at each pixel the proportions of the classes. Hopefully, these algorithms will help to improve the accuracy of satellite imagery mapping and contribute to integration of remote sensing and Geographic information science.