An application of Remote Sensing: Object oriented analysis of satellite data

More and more often remotely sensed satellite images are used for monitoring and managing the land surface. Orbiting around the earth, satellites acquire data in short intervals anytime it is necessary. Elaborating satellite’s data is possible to recognize the landslides, to know the land use, to help agriculture. Many are the fields in which Remote Sensing is useful, as in image processing. Traditional image processing and image interpretation methods are usually based only on the information extracted from features intrinsic of single pixel: the object’s physical properties, which are determined by the real world and the imaging situation – basically sensor and illumination. The pixel-oriented analysis of satellite data has a main limit: the acknowledgment of semantic low level information, as the amount of energy emitted from the pixel, where the context does not assume any role. The application of Object Oriented Image Analysis on very high resolution data allows to obtain, by an automatic or semi-automatic analysis – with a minimal manual participation - a good classification also in presence of high and very high resolution data of small cities, where higher is an error possibility. In an object oriented analysis the semantic level is raised: relation rules join space are added, topological information and statistics and so the context is defined. Recognition is so based on concepts of Mathematical Morphology applied to the image analysis and elements of Fuzzy Logic for a human-likely classification. In this application, by using a specific tool, we operate a segmentation of the entire scene on more levels. The segmentation multiresolution obtains the automatic creation of vectorial polygons, directly extracted from the raster, with the remarkable advantage of having therefore a perfect coincidence in the superimposition on raster. The final classification, predisposing an adapted hierarchy of classes that hold account of the relations between the produced segmentation levels, may be highly accurate. In this contribute this methodology is applied (through the proposition of an integrated package we are realizing for a segmentation and the successive NeuroFuzzy classification) to the aim characterizing the detection of burned areas in Landsat and Ikonos images. It is opportune to emphasize that in this note we are not using algorithms and methods (more rigorous for the resolution of the problem in examination) known and tested in literature, i.e. NBR, BAI, NDVI etc., for locating burned areas and optimizing results, methods on which moreover we are working to the aim of integrating them in the package now proposed: the goal of our work is exclusively to test the integrated fast methodology proposed and results obtainable on this application, and it isn’t to experiment known resolution methods or algorithms and/or innovative for the application.