Abstract: In this study,Minqin oasis,Gansu Province,China was taken as the research area. Based on the GF-1 remote sensing image and wild typical sampling data,the object-oriented classification method was used to classify the images in order to obtain the vegetation spatial distribution information. According to the specific characteristics of the study area,the segmentation scale 5 was taken as the initial value,and the segmentation experiment was carried out with the increase of step length from 5 to 100. In the first layer,the Normalized Difference Vegetation Index(NDVI)was used to distinguish vegetation and non-vegetation,and the threshold value was 0.16. In this layer,In order to distinguish the smaller vegetation,the segmentation scale was 10. In the second layer,the Normalized Difference Water Index(NDWI)was used to extract the water in non-vegetation,and the threshold value was -0.02. According to the size,shape and other characteristics of water,the segmentation scale was 35. In the third layer,with the training samples obtained from field sampling survey,the nearest neighbor classification method was used to extract the vegetation from the cultivated land,woodland and grassland. In this layer,the segmentation scale was 25. The accuracy of the classification results was evaluated by the sampling data and the Google map data. The overall accuracy was 83.02%,and kappa coefficient was 0.745 1,compared with supervised classification of the pixel-based method,whose overall accuracy was 69.37% and kappa coefficient was 0.497 0. It indicates that the object-oriented classification method has more advantages than the supervised classification of the pixel-based method in extracting the vegetation information of arid area. In this paper,according to the rule of different features using different segmentation scales,the hierarchical classification could not only decrease the occurrence of over-segmentation and under-segmentation,but also improve the classification accuracy and shorten the time. It was a new attempt in the study of vegetation classification in arid regions.

Key words: GF-1, Minqin oasis, object-oriented, supervised classification