Resource information
Automatic information extraction from optical remote sensing images is still a challenge for large-scale remote sensing applications. For instance, artificial sample collection cannot achieve an automatic remote sensing imagery classification. Based on this, this paper resorts to the technologies of change detection and transfer learning, and further proposes a prior knowledge-based automatic hierarchical classification approach for detailed land cover updating. To establish this method, an automatic sample collection scheme for object-oriented classification is presented. Unchanged landmarks are first located. Prior knowledge of these categories from previously interpreted thematic maps is then transferred to the new target task. The knowledge is utilized to rebuild the relationship between landmark classes and their spatial-spectral features for land cover updating. A series of high-resolution remote sensing images are experimented for validating the effectiveness of the proposed approach in rapidly updating detailed land cover. The results show that, with the assistance of preliminary thematic maps, the approach can automatically obtain reliable object samples for object-oriented classification. Detailed land cover information can be excellently updated with a competitive accuracy, which demonstrates the practicability and effectiveness of our method. It creates a novel way for employing the technologies of knowledge discovery into the field of information extraction from optical remote sensing images.
