Systeme prototype pour le suivi des changements de l'occupation du sol en milieu urbain fonde sur les images du satellite RADARSAT-1

Land use and land cover mapping is of great importance for urban environment management. However, these maps become very rapidly outdated and require constant updating. Even in this day and age, most of the updating process is performed manually with the aid of aerial photography. The essential stages of this process consist in detecting the changes, establishing their nature and performing the database updating accordingly. Automating this operating procedure, even partially, is of utmost importance. Radar imagery may be acquired over large territories even in the presence of clouds. It has been the subject of very few research for urban mapping and there exists no operational application that makes use of it for map updating in this type of environment. However, it is well known that the imaging radar in general is sensitive to raised structures such as buildings. The apparition of new buildings in vacant areas, where most land cover and land use changes occur in urban areas, is clearly an indication for change detection. In the framework of this thesis, RADARSAT-1 imagery acquired in fine mode (10 meter resolution) has been investigated for its use in an innovative image processing method: the map-guided analysis of remote sensing data. The outdated map serves as a guide to segment a more recent image in order to compute characteristic features allowing to detect and identify changes. Tests show that texture measurements obtained with a fast method derived from the grey level co-occurrence matrix constitute such characteristic features allowing new buildings detection. Using 11 pixels square windows, contrast and entropy measurements are computed on a fine mode RADARSAT-1 image. Choosing appropriate thresholds is sufficient to pinpoint new buildings in vacant areas. Validation of this detection method is carried out over 5 sites presenting diverse physiographic characteristics using a dataset of 14 fine mode RADARSAT-1 images acquired with various incident angles throughout 1999. In the best cases, an accuracy of over 90% can be reached, while in the worst case this accuracy is around 75%. The results show that image acquisition using large incidence angles (F4 or F5) during a dry period (summer or fall) is preferable. With regard to the identification of the new taxons, it cannot be achieved with the radar imagery. Thus, a prototype system is developed to perform change detection using a RADARSAT-1 image, which then constitutes the key for the search by an operator into territorial databases (aerial photographs, land register, valuation rolls, etc.) to identify the new taxons. This prototype system, applied in a real case test, shows its potential for an operational use to accelerate and facilitate land use and land cover map updating over large territories.; Key words. cartography, land use/land cover, urban areas, automatic updating, map-guided analysis, RADARSAT-l, texture measurements, grey level co-occurrence matrix.