| Résumé : |
(Auteur) The world's population is rapidly increasing, especially in urban regions to which many rural inhabitants are migrating. Such an effect results in the need for a more efficient method of monitoring cities, both in developing and developed countries. Present monitoring techniques are inefficient, and unable to effectively maintain up-to date information due to the population increase. Hence, the demand for settlement detection, urban classification and population estimation is apparent.
Radar remote sensing is showing great potential for assisting in such a matter. Its ability to discriminate between small buildings of sparse layout, and large, densely spaced, buildings is slowly being realised. This research addresses this issue through the development of a primarily theoretical model.
The urban environment is a complex mixture of built as well as natural elements. In order to simplify such a difficult situation, the backscatter expected from urban areas can be divided into simple scattering mechanisms. The most important ones chosen for this research are double bounce (from building walls and tree trunks), single bounce (from building roofs), and volume scattering (from tree canopies). The model has been designed to consider these scattering mechanisms, and their expected proportions, occurring in the urban environment.
One of the greatest difficulties created through using radar in the built environment, is due to the 'Cardinal Effect'. This occurs when there is strong double bounce scattering due to the intersection of the orthogonal building wall and ground surface being perpendicular to the look direction. An empirical investigation is conducted in the thesis to further understand this phenomenon.
In order to develop the model, many parameters, including those relating to the cardinal effect, need to be considered. Determination of such parameters is not a simple task, and requires some general assumptions to be made. The model has been designed to determine the backscatter and polarisation information for two different urban classes (ie., residential and commercial). These classes are chosen because they generally represent the low and high density urban areas respectively.
The backscatter and polarisation information from a single building is developed and then expanded into a block of buildings, with consideration of radar shadowing effects. Trees are also included in the simulated residential areas.
In order to test the accuracy of the model for residential and commercial land use, test sites representing a large range of orientation and look angles, are chosen in the city of Sydney, Australia. AirSAR data for these test sites are compared to model simulations representing the same characteristics. The results show that L-band model output is quite comparable with the real data. The P-and C-band are less reliable, with the model C-band results showing little resemblance to the AirSAR information.
For the model to be further tested, a simple classification is performed over a large area of the Eastern suburbs of Sydney. L-band data is used due to the model simulations closely represerding real data. Furthermore, the real P-, L- and C-band polarisation information were found to be quite similar, so a multiwavelength classification using radar data was not seen to contribute significant information compared to a single band.
The classification is based on a comparison between the model and AirSAR total power, polarisation index and polarisation phase difference. The classification shows that the model has the ability to distinguish between low and high density urbw areas. However, due to the difficulties in defining the characteristics of residential and commercial land uses, there is some overlap in the classification. Some tall, dense residential areas are classified as commercial. Similarly small scale commercial areas are classified to be residential. Such a problem is related to the definition of land use rather than land cover.
The most important classification parameter observed through this exercise is the total power. The polarisation index proved of little use due to its incapability of distinguishing between residential and commercial classes for the real data. Classification using the total power and polarisation phase difference together gave a less accurate result than total power alone.
However, overall results show that the potential which radar has as a remote sensing tool for distinguishing between low and high density urban areas, and for classification (under the appropriate circumstances), is quite high. This is especially so when radar is combined with other information gathering systems, such as optical remotely sensed data. Such a combination could be very beneficial for the growing need for urban monitoring and population estimation. |
Ajouter le résultat dans votre panier Affiner la recherche Interroger des sources externesDeep learning for multi-modal classification of cloud, shadow and land cover scenes in PlanetScope and Sentinel-2 imagery / Yuri Shendryk in ISPRS Journal of photogrammetry and remote sensing, vol 157 (November 2019)
Development of models for monitoring the urban environment using radar remote sensing / Catherine Ticehurst (1998)
