عنوان مقاله [English]
This research addresses the automatic extraction of alluvial fans using four methods of segmentation from satellite data. This segmentation method divides images into partitions. It is typically used to recognize objects or other relevant purposes in digital images (Fu, 2013:3260). Alluvial fans have always been a landform that attracts human because they are suitable areas for living due to freshwater and appropriate soil for drinking, cultivation, making pottery, making mud-brick and other activities (Maghsoudi and Azizi, 2012: 23). Therefore, alluvial fan extraction is significant in the planning of engineering geomorphology and other related disciplines.
During the recent years, many segmentation techniques have been developed (Ranasinghe, 2008). In this research, the most popular segmentations are presented and then those that are appropriate to identification of alluvial fans of geomorphology were introduced. In general, land-surface segmentation has demonstrated to have a great potential to improve geomorphological mapping with better representations of geomorphological objects. Segmentation divides land-surface into relatively homogeneous areas, by polygons based on input criteria. Segmentation results are used to identify objects and their classification (Drăguţ et al., 2013). The main objective of this research is to introduce and implement algorithms for geomorphological landforms segmentations that the target landforms are alluvial fans and bahada in this research. The selected study area is in Yazd Basin and to test the ability for generalization of the selected methods, the similar alluvial fans of the central city of Yazd province have been selected. Briefly, importance of segmentations in geomorphology is in the extraction of landform objects, landform classification, landform isolation and identification details of landforms.
The methodology of this study is based on processing segmentation on high-resolution images of Geoeye-1 as well as the ASTR-1 multispectral satellite images within the E-Cognition Developer© software from Trimble company. Arc Catalog and ArcGIS are used for production of the required layers in the proposed flowchart. In this study, two main approaches have been used in the construction segmentation. In the Top-down Segmentation, the objects of image are divided into smaller parts. Top-down approaches are approximately implemented by three algorithms: 1. Chessboard segmentation, 2. Quadtree-based segmentation, and 3. Contrast Split Segmentation. The forth method for segmentation is called multi-resolution segmentation that is the most popular method in the bottom-up segmentation approach (Baatz and Schäp, 2000).
We have described the four methods, and then each of those methods has been executed on the satellite images within the mentioned platforms. The outputs of each segmentation processing have been evaluated based on visual interpretation of the images. According to the flowchart proposed, outputs of segmentation have been separately overlapped on the high-resolution Geo-eye images that are used in ArcGIS environment. The existing map of geomorphology was used to improve visual interpretation. In this study, we used not only the top-down segmentation but also Bottom-up Segmentation approaches.
Results and Discussion
The segmentation results of the four methods in the E-Cognition Developer© software from Trimble company was as follows:
In short, the first method converts the image into a square shape that its output is a chessboard image. In the second method, the entire image is divided into four squares of the same size using the standard deviation or other criteria as a separate factor, and then each square is also divided into four smaller parts until to a defined threshold. These divisions continue until the objects are separated from each other based on shape and color homogeneity. This method will produce narrow strip initial segments for features with a large length-width ratio (e.g. roads, waterways, strip erosion types.), that is suitable for extraction of narrow objects in the context images. In the third method, the objects are separated by polygons from each other based on threshold values. This indicates the degree of difference between darkness and brightness. We were able to extract details of the alluvial fans (e.g. Shadow of gully erosion, oued,) using contrast segmentation method. The forth method, the image pixels or small objects are combined based on the criterion of homogeneity in successive with neighboring pixels or objects to lead to the production of larger objects. Therefore, the objects with homogeneous color and shape are combined to form a larger one. This technique is based on region growing concepts, in other words one or some known pixels are developed by the rest of unknown pixels based on a criterion.
On the basis of the results, it was concluded that two algorithms are popular and applicable in geomorphology: A) The multi-resolution algorithm is precise and high performance to identify the geometry of the alluvial fans of Yazd Basin; B) Contrast split segmentation has been successful to identify details on the body of the alluvial fans like to gully erosion, shadows, roads, Oued. Finally, in order to examine the testability of the selected methods, the multi-resolution algorithm has been executed in the similar fans in other parts of the central city of Yazd province. Its results have proved the generalizability of these methods, because the algorithm is repeated four times to identify and extract the boundaries of the alluvial fans, as the outputs appeared quite similar in the morphology of alluvial fan.