| Summary: | 碩士 === 國立中興大學 === 資訊科學與工程學系 === 96 === Along with the fast development of remote sensing technology, and the availability of international market of high definition commercial satellites, satellite technology field has been expanded, and spatial resolution of imagery has also been substantially raised up. Regarding the application of using satellite images to investigate the earth’s surface, it is hard to meet all actual application requirement and raise accuracy by traditional information extraction methods because the surface features of Taiwan island have become more and more complicated.
Traditional pixel-based classification methods only focus on single pixel spectral information for classification, while ignore the important spatial information such as shape and adjacent relationship. In this thesis, image classification based on object-oriented method is adopted. The entire research includes not only using spectral and spatial factors to divide the image into several meaningful object sections, but also classifying the knowledgebase through a category structure according to both surface features and space characters. By this object-oriented classification method, ambiguity in classification can be avoided and high accuracy is reached. In this research, SPOT-5 and QuickBird satellite images are used for testing materials, and “object-oriented method” and “maximum likelihood classifier” are adopted to compare the results . According to the analysis for the results, the accuracy of “object-oriented method” is approximately 86% and 83% for a SPOT-5 satellite image and a QuickBird satellite image,respectively. while that of “maximum likelihood classifier” is about 82% and 77% for a SPOT-5 satellite image and a QuickBird satellite image,respectively. In terms of classification effect, we conclude that “object-oriented method” is obviously superior to “maximum likelihood classifier”.
|
|---|
