Proceedings Article | 25 October 2012
KEYWORDS: Image segmentation, Image classification, Agriculture, Image processing, Image analysis, Multispectral imaging, Landsat, Accuracy assessment, Silver, Earth observing sensors
Considering the dramatic change occurred in the Blue Nile region of Sudan, this study is of great value for developing a method for identification of forestland cover extents, integrating rate of changes and causes. The study utilizes three consecutive optical multispectral images, two LANDSAT TM images of 1990 and 1999 as well as TERRA ASTER image of 2009 to evaluate forest cover dynamics during the period 1990 to 2009. The method adopted in this research consists in cross operation of classified images of different points in time, which utilizes the overlaying images to be compared for change detection. New layer of segments was created representing the change areas as well as the overlapped areas of each pair of classified images. Consequently, a series of optimized algorithms have been developed to estimate the change in Land Use Land Cover (LULC). At the fundamental stage, smooth and accurate classified images are very essential for any post-classification change detection technique, which were typically achieved by object-based approach (OB) with overall accuracy 91 %, 93 % and 95 % for the years 1990, 1999 and 2009 respectively. Nine LULC classes were generated from each, i.e. agriculture (Ag.), bare-land (Br.), crop-land (Cr.), dense-forest (DF), grassland (Gr.), orchard (Or.), scattered-forest (SF), settlements (St.) and water (W). Therefore, and considering the dramatic change observed in the area, the fusion operation of multi-temporal data results initially in quite numerous change "from-to" information classes, which allows for aggregation of these classes at any hierarchical level of details. Moreover, the developed approach allows the operator to effectively know the spatial pattern of change, trend and magnitude of the dynamics occurred in each of the classified LULC classes. While many change-detection techniques have been developed, a little has been done to assess the quality of these techniques. Hence, the change maps resulting from cross operation were assessed, which reveals that, the accuracies of the change maps for the two time intervals were consistently high.
