An Automated Preprocessing Chain for Optical Satellite Imagery

Wolfgang Lück
Technical Manager
Earth Observation Data Services,
South Africa
Email: wluck@csir.co.za

One of the largest hurdles encountered by the remote sensing community is the inability to derive thematic information at the required speed, consistency, quality and cost, demanded by the geoinformation market, from available imagery.

This obstacle can only be overcome by automating image classification techniques. As a prerequisite a migration from the currently dominant qualitative, to quantitative classification approaches and the necessary image rectification and radiometric calibration has to take place. An automated preprocessing chain producing imagery and classification products suitable for quantitative image analysis is being implemented at the Satellite Application Centre of the CSIR / South Africa. This system based on libraries from the commercial software PCI Geomatica 10, Geomatica X and Definiens Enterprise, is characterized by its flexibility and use of cutting edge algorithms for all preprocessing steps.

The system supports a wide range of optical imagery from QuickBird, Ikonos, Spot 2 / 4 / 5, Landsat TM & ETM+, ASTER and MODIS. Level 1 raw imagery are deposited in a predefined directory structure from where they are processed in batch mode. The preprocessing steps include image import, metadata extraction, image stitching, GCP collection, Tie Point collection, image triangulation via bundle block adjustment, DEM extraction where stereo imagery is available, ortho-rectification, mosaicking and tiling. In addition various radiometric corrections are applied concurrently to the geometric corrections. These consist of a conversion from digital numbers to at satellite reflectance, atmospheric correction via radiative transfer modeling, topographic normalization and true color reconstruction for sensors such as ASTER and Spot.

The preprocessing chain provides imagery to a thematic information extraction engine based on Definiens Enterprise Server, using predefined classification protocols as input for an object oriented classification. Classification products are served in the form of a geodatabase including the raw imagery, classification rules, fuzzy classification results and ISO conforming metadata.

The system is currently operational in its first stages with the preprocessing chain supporting SPOT, Landsat and ASTER. It should be fully operational by the end of 2007 with its integration with Oracle 10G, a web based catalogue and portal conforming to all ISO standards.