U2: Oral Presentation C Friday, 11:00 – 11:30 am C Gooseberry Falls Room

Statistical, Spectral and Spatial Knowledge Based Classification System for Multi-Spectral Remote Sensing Imagery

Ranga Raju Vatsavai, Thomas E. Burk, Paul V. Bolstad, Marvin Bauer, Tim Mack, Jamie Smedso
University of Minnesota
Department of Forest Resources
Remote Sensing and Geospatial Analysis Laboratory
115 Green Hall
1530 Cleveland Avenue North
St. Paul, MN  55108
vrraju@gis.umn.edu

Thematic classification of multi-spectral remotely sensed imagery for large geographic regions requires complex algorithms and feature selection techniques. Traditional statistical classifiers rely exclusively on spectral characteristics, but thematic classes are often spectrally overlapping. The spectral response distributions of classes are dependent on many factors including terrain, slope, aspect, soil type, and atmospheric conditions. It is now possible to improve traditional classifiers and develop new classification systems that can incorporate the spatial knowledge derived from newly available geo-spatial databases. However, it is not easy to incorporate this additional knowledge into traditional statistical classification methods. On the other hand, knowledge-based and neural network classifiers can readily incorporate these spatial databases, but these systems are often complex to train and their accuracy is only slightly better than statistical classifiers. In this paper we present a new classification approach, combining knowledge-based systems and statistical classifiers to minimize the disadvantages of the traditional approaches. The remote sensing image is first stratified into spectrally homogeneous regions using the spatial knowledge derived from ancillary geo-spatial databases. A semi-automated supervised training is carried out for each of these regions. These regions are then classified using maximum likelihood classifier. Even in a complex geographic setting, this classification fusion approach has yielded an overall classification accuracy of about 90%. 

Keywords: Statistical and Knowledge Based Classifiers, Spectral and Spatial Knowledge, Classification Fusion, Thematic Classification