Classification of farmland using multitemporal aerial images

Abstract

It is well known that there is a growing need for consistent and up-to-date GIS-data at various scales for administrative and regulatory applications. Especially farmland classes are of high interest in this context. A new automatic method for the classification of crops is described. The method is based on sequences of digital aerial orthophotos with a ground sampling distance of 0.17m. The applied image sequence consists of twelve images of the same region within one vegetation period. Expert knowledge about the crops together with extracted features leads to temporal models for each crop. The temporal change of the features along with a changing relevance of a feature is considered. The temporal models are applied during classification that is based on a weighting function. The approach is tested on a test site of about 700ha and achieves correct classification rates better than 90%.