Physical Aspects of Multispectral Remote SensingProject of CENSIS, supported by the Volkswagen-Foundation Researchers: Prof. Dr. Johann Bienlein, Dr. Ramon Franck, Dr. Martin Kollewe, Dipl.-Phys. Marc Hermann, Dr. Gerhard Meister, Dr. André Rothkirch, Dipl.-Phys. Niklas Rega, Prof. Dr. Hartwig Spitzer, Dr. Rafael Wiemker, Dipl.-Phys. Stefan Zenk Working Areas - Group ProfileThe importance of remote sensing of the earth by airborne and satellite based sensors for research and application purposes grows steadily. A working group led by Prof. Dr. Hartwig Spitzer at the 'Center for Science and International Security at the University of Hamburg (CENSIS)' investigates the physical principles of remote sensing and develops new methods for computer based image processing. The objective is to contribute to the methodology of digital processing of multispectral remote sensing images (e.g. classification algorithms), taking physical aspects of the measurement process into account. Possible applications include:
Up to the 1990's, black-white-photographs have been a major data source for monitoring in cadastre and urban planning applications. Their main advantages are relatively easy and well-known handling, very high spatial resolution and easy data storage possibilities. However the technological trend points towards increased use of digital cameras and digital electro-optical multispectral or hyperspectral sensors with up to 300 spectral channels. This considerably eases the production of thematic maps, where every pixel is assigned to a certain utilization category. We have therefore specialized on the analysis of multispectral aerial images from a sensor which combines reasonably good spectral and spatial resolution For the development of algorithms, we have used image data from the airborne based scanner DAEDALUS AADS 1268. This sensor has 11 spectral channels from visible light to thermal infrared and a geometrical resolution of 2.5 mrad, corresponding to a spatial resolution of 75 cm at a flight altitude of 300 m. Five image sets of industrial, residential and vegetation areas at Nuremberg, Germany were acquired in cooperation with the German Aerospace Center (DLR, Oberpfaffenhofen). In addition to the DAEDALUS data aerial photographs were taken. The image sets cover the same area in five years and three seasons (August 1991, April 1992, Oktober 1994, Juli 1995, August 1997). For three years (1994, 1995, 1997) calibration measurements with a spectroradiometer on the ground were also performed. To compare different takes of the same scene, the measured raw data must be converted first to radiances, then to reflectances. To obtain the correct reflectances, the illumination by sunlight and skylight as well as the transmissivity of the atmosphere must be known. The necessary parameters can be estimated by deriving visibility and aerosol distribution from simulation, or by comparison with reference measurements on the ground. In order to distinguish between a real change in reflectance and a change produced by a different illumination or viewing geometry, the bidirectional reflectance characteristics of urban surfaces are studied. On one hand, the illumination of a reflecting surface depends on its slope relative to sun- and skylight. On the other hand, angle dependent reflection (BRDF effects) influence the spectrum measured at the sensor. In situ measurements with a spectrometer on several sample surfaces allow quantitative predictions of the influence of surface orientation and sensor position for different reflecting surfaces. One goal is the development of classification algorithms that are not mislead by these effects. The georeferencing of images acquired by airborne scanners is more difficult than for aerial photography, because they depend heavily on the movements of the plane. For a coordinate transformation based on landmarks, locally adaptive transformation functions are necessary. Our working group has implemented, compared and improved several algorithms for geocoding, especially for images acquired by airborne scanners. Supervised and unsupervised algorithms have been used and developed for classification and change detection. The simultaneous use of spectral and context based characteristics yields interesting improvements.
The working group cooperates with Prof. Dr. Leonie Dreschler-Fischer from the division 'Cognitive Systems' of the computer
science department of the University of Hamburg. |