Abstract: The karst topography in southwestern China is filled with fantastic caves and curved areas. And the classification of groundobject is relatively fragmented, which leads to the reductionin in the accuracy of the traditional one-time classification of spectral features.Based on the high-definition of UAV orthophoto and terrain factor, this paper makes full use of the spatial, spectral, texture and terrain features of UAV remote sensing images and adopts the object-oriented CART decision tree algorithm and hierarchical strategy to pickup the coverage type of the land inspecific study area. The method of combining spatial terrain factors and hierarchical strategies reduces the interference among the fragment edground objects in the karst area actually has a much higher classification accuracy. The whole classification accuracy reach 91.2%and the Kappa coefficient is 0.87, which is respectively 9.8% and 0.13 higher than the traditional one-time classification accuracy and Kappa coefficient. It shows that this method is suitable for decoding land coverage in southwest karst area, which could provide favorable reference for land-use monitoring.

Key words: terrain factor, stratification strategy, UAV image, multi-scale segmentation, decision tree classification