Comparison of pixel-based and object-based classifications of high resolution satellite data in urban fringe areas Noritoshi Kamagata 1 , Yukio Akamatsu 2 , Masaru Mori 3 , Yun Qing Li 4 , Yoshinobu Hoshino 5 and Keitarou Hara 6 1, 6 Graduate School of Informatics, Tokyo University of Information Sciences 1200-2 Yatoh-cho, Wakaba-ku, Chiba, 265-8501 JAPAN 1 h05002nk@edu.tuis.ac.jp, 6 hara@rsch.tuis.ac.jp 2, 3 Kokusai Kogyo CO., LTD. 3-6-1 Asahigaoka, Hino City, Tokyo, 191-0065 JAPAN 2 yakamatu@kkc.co.jp, 3 masaru_mori@kkc.co.jp 4 Japan Space Imaging Corporation 8-1 Yaesu 2-Chome, Chuo-ku, Tokyo, 104-0028 JAPAN yunli@spaceimaging.co.jp 5 Tokyo University of Agriculture and Technology 3-5-8 Saiwaicho, Fuchu, Tokyo, 183-8509 JAPAN hoshino@cc.tuat.ac.jp Abstract: This study compares pixel-based and object-based classification of land cover using high resolution satellite data available for urban fringe regions. In the pixel-based analysis the maximum likelihood method and the ISODATA method were applied. The results showed that in both methods misclassification tended to increase due to shadows. The pixel-based classification also experienced difficulty due to factors such as the varied shapes of the forest canopy and mixing of vegetation, etc. The object-based classification, in contrast, relies on abstraction of comparatively homogenous areas, and proved capable of extracting the boundaries among all the forest types. In addition, this study employed a high number of minute patches, and proved effective even in regions where tree species were mingled together. Some misclassification problems remained, which have to be addressed by future trial and error experiments in parameter setting. Still, object-based classification of high resolution satellite data was shown to be an effective tool for analyzing vegetation cover in semi-urbanized and countryside landscapes on the outskirts of large cities, where various vegetation types, as well as buildings and other infrastructures, are mixed together in small areas. Keywords: High resolution satellite data, Object-based classification, Pixel-based classification, Vegetation 1. Introduction In urban fringe areas where rapid segmentation and integration of land cover occur, a precise grasp of the current vegetation pattern, as well as regular monitoring of changes, are required for efficient management of the regional landscape. Remote sensing offers the possibility of quick and inexpensive methods for identifying and classifying forest types and other landcover [1] [2]. High spatial resolution data, such as IKONOS, is especially attractive as a tool for enhanced discrimination of cover types [3]. When using high spatial resolution satellite data, however, there is a limit to the effectiveness of conventional pixel-based classification, and object-based classification has been suggested as a more effective method. This research was implemented in a region where various vegetative and structural elements are intermingled in a complicated mosaic pattern. In addition, changes in the land cover and other factors are rapid. Object- based and pixel- based classification were compared for effectiveness in this sort of heterogeneous, rapidly changing landscape. The strengths and weaknesses of each system were identified and analyzed. This research was partially supported by the Ministry of Education, Culture, Sports, Science and Technology (MEXT) and the Japan Society for the Promotion of Science (JSPS), Grant-in-Aid for Scientific Research (B) 16380097, (C) 15510195 and supported in part by the Academic Frontier Joint Research Center, Tokyo University of Information Sciences, which is supported by MEXT, Japan.