傳統的像元式影像判釋只能得到在光譜上呈均質的覆蓋區域,並無法正確區格出由不同土地覆蓋所組成的土地利用,因此地物導向區格化的概念就應運而生。地物導向的概念其實就是一種加入大量空間資訊的高階分類,運用人對地物的認知所產生的知識讓電腦知道影像中到底存在什麼地物。地物導向區格化理論已被證明可以成功區格人為定義的地物類別,但是在實際的運作過程中,仍有許多環節需要更細緻化、更系統化、更正規化,所以本研究的定位為建立地物導向區格知識正規化,即只要是對地物特徵的描述經過正規化的知識抽取及整理後,都可以套入一個階層性區格的主要流程來完成區格。 雖然本研究大部分針對複合式有特徵地物之地物類型來做區格化的實作,但由此建立的邏輯架構也可繼續應用至其他不同類型地物的區格化作業,唯後續尚有許多可再深入研究的部分,以補全地物導向區格化理論的完整性,使遙測影像判釋的能達到自動化與精確度兼具的境界。
Due to that traditional image processing methods can only produce land cover of homogeneity in spectra, but cannot generate the segments of the landuse made up of different land covers, the category-guided segmentation comes with the tide of fashion of searching better image analyses. The concept of the category-guided segmentation is a kind of high-level classification making use of abundant spatial information and knowledge derived from the human understanding of surface features, and this new method has been successfully proved to separate artificial surface features from images. Even so, there are still many links in the segmentation procedure needing to be more refined, organized and normalized. Therefore, this study aims to set up a normalization of category-guided segmentation, that is to say, after extracting knowledge of the description of landuse characteristics, most landuse can be segmented with a main procedure of the hierarchical segmentation. Although this study practiced segmentation most concerned with the compound landuse with characteristic features, the derived logical framework can be applied to other kinds of landuse types as well. Nevertheless, many details need discussing to complete the theory of category-guided segmentation, making segmentation process more automatic and results more accurate.