RS classification information extraction of landuse in karst area by means of object oriented approach: A case in Bijie, Guizhou

Although traditional pixel-oriented classification can get good result in the extraction of information from the remote sensing image with marked spectral difference, the “salt and pepper phenomenon” cannot be avoided and the information of texture and shape cannot be fully applied, which resulting in large amount of information loss. In this paper, in order to improve the accuracy of remote sensing information extraction, the land-used information in Bijie, Guizhou, is extracted automatically by way of object-oriented approach. Firstly, the regional images of Landsat-5 TM is segmented multiscalely to create the image object layer. Then, remote sensing interpretation is done for karst area in light of knowledge decision tree classification and Suppot Vector Machine (SVM) classification techniques. The results show that the object-oriented classification techniques can accurately and efficiently extract land-use information in karst area, and can avoid the “salt and pepper phenomenon” meanwhile. Data verification by sampling in the field proves that the first classification accuracy of the first level is 91.7 % and the second level classification accuracy is 89.4 %, indicating the object oriented approach has a good application effect in Bijie, Guizhou Province.