Radiance transformation of multitemporal LANDSAT image for land cover classification

Recently classification of remote sensing images using neural network approach is studied. However multitemporal LANDSAT image data is not used for classification. A problem in classification of remote sensing images is that we cannot get images in fixed wide area such as states of prefectures at one time because the range of sensors of artificial satellite is limited. For example, full area of Fukuoka prefecture is observed two separate images. For multitemporal images, several factors affect the spectrum at different observation dates. In this paper, we concentrate on the sunbeam factor from which we can estimate the intensity. Using the sun elevation angle from the data, we can estimate sunbeam intensity. We transform multitemporal images using radiance transformation which is based on path radiance model. We confirmed that radiance transformation is effective to the classification of multitemporal images from several experiments.