期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2016
卷号：XL-3/W4
出版社：Copernicus Publications
摘要：It has been proven that roof runoff water plays an important role in the high metallic concentration levels in urban rainwater since metallic elements are generated by corrosion of roof materials before being swept away by rainwater. The aim of TOITEAU project is therefore to model this phenomenon, evaluating the metallic .ows from roofs in rainwater. To achieve this goal, an important work has already been done to model those .ows at roof scale. But, it has now to be extrapolated to a whole drainage area, requiring knowledge about the areas concerned by the different kinds of roof coverage, that is to say that a map of roof materials is needed. Such information can be extracted from aerial (ortho-)images owing to (supervised) classification techniques. In the present situation, only six classes corresponding to the following kinds of roofs were defined : zinc plates, slates, red tiles, brown tiles and .at roofs. Nevertheless, classification results are limited because of several factors that have therefore to be dealt with. First of all, some distinct classes have very similar radiometric distribution (such as for instance zinc and at light slates), making it hard to distinguish between them. That's why derived channels computed from initial red-green-blue channels of the ortho-image have been used to improve the classification results. Texture channels have also been tested especially to discriminate zinc from other light coloured roof materials. For the same reason and in order not to obtain a too "noisy" result, per region classification algorithms have been used : homogeneous regions will be classified instead of pixels. Secondly, roofs are the only interesting parts of the ortho-image in this study. As a consequence, a building mask is first computed from digital topographic database BDTopo in order to classify only roofs. However, several elements concerning data precision have to be taken into account at this step. For instance, the ortho-image and the topographic database can obviously not have been captured at the same date and, as a consequence, buildings can have been destroyed, modified or built between these two distinct capture times. In addition, as the used ortho-image is not a "true ortho-image", building objects from digital topographic database and ortho-image roofs are not perfectly superposed. However, these topographic database building objects can be registered to the ortho-image. Nevertheless, it must be said that these database objects often remain caricatures of true buildings. Besides, most of the time, homogeneous regions to be classified do not directly correspond to database buildings since those database objects can be groups of buildings or buildings of which the roof is composed of different materials. Therefore, it is necessary to segment building areas (according to the topographic database) of the ortho-image into homogeneous regions that are then classified. Lastly, shadows can be quite important in roof areas because of the presence of roof superstructures or higher buildings in the neigh- bourhood. That's why an additional class "shadow" is also defined in order to take into account shadow areas where radiometric information is not sufficient to discriminate between the different kinds of materials. Tests have been carried out on two distinct study areas with 50cm resolution orthophotos for the first one and 12cm resolution orthoim- ages for the second one. The first study area was a dense urban centre, whereas the second could be divided into several parts : a residential suburb consisting of houses, a dense urban centre with buildings having up to 4-5 levels and a mixed residential / service area consisting of higher buildings
