Abstract
Various environmental problems observed in areas ranging from a small region to the whole area of the earth are closely related to land use changes resulting from the human activity. To prevent further environmental problems, it is necessary to determine quickly changes of land use, predict the effect of the change and plan countermeasures. Therefore, estimating the present state of the land use is very important for the management of the terrestrial environment.
Land cover distribution measurement by an artificial satellite has proven to be an effective means of estimating land use changes. However, most images observed by an artificial satellite are composed of mixed pixels which include various categories in a pixel. In particular, considering mixed pixels is necessary for land cover analysis using low spatial resolution image.
In this paper, two methods of the gradient method and the matched filter method using hyper-spectral data to estimate the land cover ratio of interest in a pixel are proposed. The gradient method estimates coverage of interest using the characteristic wavelength range such as the absorption in the observation wavelength range, and the matched filter method estimates coverage of interest with finding the maximum value of cross correlation between signal of observation and interest. The common features of proposed methods are to use hyper-spectral data and estimate the coverage from only the spectral radiance of a specific category. The effectiveness of proposed methods were also examined. As a result, it was shown that the two methods proposed are effective for estimating land cover ratio of interest in a pixel.
