Knowledge-Based Object Oriented Land Cover Classification Using SPOT5 Imagery in Forest-Agriculture Ecotones

This paper describes a knowledge-based object oriented classification method using SPOT5 imagery in Forest-Agriculture Ecotones. It is based on optimized application of expert knowledge information extraction from remote sensing imagery, geographic data, investigated data, chessboard image segmentation and multi-resolution image segmentation technique. Due to these capabilities, the method represents a significant improvement in land cover classification. This approach can also be seen as a framework for integrating external knowledge with image classification procedures. Confusion matrix is used to do accuracy assessment and our assessment results show that he knowledge-based object oriented classification improves the total accuracy from 61.352% (pixel-based Minimum Distance classification), 91.30% (object oriented Nearest Neighbor classification) to 94.40%. The result indicates that this method leads to a higher classification accuracy.