Classification of multitemporal remote sensing data using Conditional Random Fields

Land cover classification plays a key role for various geo-based applications. Many approaches for the classification of remote sensing data assume the features of neighboring image sites to be conditionally independent. However, using spatial and temporal context information may enhance classification accuracy. Conditional Random Fields (CRF) have the ability to model dependencies not only between the class labels of neighboring image sites, but also between the labels and the image features. In this work we present a novel approach for multitemporal classification in high resolution satellite imagery using CRF that is based on an extension of the CRF model by a time-dependant component. The potential of our approach is demonstrated using a set of two Ikonos and one RapidEye scenes of a rural area in Germany.