A Comparative Study Of Endmember Extraction Algorithms On Hyperspectral Image

• Main Authors: Guo, Chun-Yin, 郭純吟
• Other Authors: Wu, Jee-Cheng
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/08002653538408153643

碩士 === 國立宜蘭大學 === 土木工程學系碩士班 === 100 === Because of the factor of spatial resolution, there are mixed image pixels in hyperspctral images which have a large number of band image information. This research is to study how to extract the pure endmember from hyperspctral image and analyze the pros and the cons of extraction methods in the presumption of the pure endmembers existing in hyperspectral image. This research uses four methods to extract endmembers from hyperspectral images, and to compare the efficiency of each method of the extraction. The four endmember extraction methods are Pixel Purity Index Algorithm (PPI), N-FINDR Algorithm, (N-FINDR), Automatic Target Generation Process (ATGP), and Sequential Maximum Angle Convex Cone (SMACC). First, we use two hyperspectral images and five figures of extracted endmembers (5, 7, 9, 11, and 13) to find out the pure endmembers. Second, we compare similarities of ground truth pure endmember and endmember which are extracted by the four methods. Last, we analyze the pure endmembers obtained by the four extraction methods, and the good and bad points about each method. The results are shown as: (1) the chosen five endmembers with the four methods are unable to extract pure endmembers. (2) The most similar endmembers are found when the number of endmember extraction is figure 9. (3) The method of PPI and SMACC is able to obtain the most similar pure endmembers.