Abstract:

Taking the oasis-desert transition zone in the eastern part of Kuqa City,Xinjiang as the research object and using GF2 remote sensing image as the main data source,on the basis of field investigation,supervised classification based on pixel and object oriented classification based on hierarchical multi-scale segmentation were used to accurately identify the vegetation information in the study area.The results showed that:1) The results of supervised classification and object-oriented classification were roughly the same.The overall classification accuracy rates of both methods could reach more than 94%,and the Kappa coefficient was greater than 0.93,both of which reflect higher classification accuracy.2)Compared with supervised classification,the object-oriented classification method improved the overall classification accuracy by 3.79%,and the Kappa coefficient increased by 0.032,which had a better classification effect and classification accuracy.By determining the optimal scale segmentation,the object-oriented classification method can more accurately extract vegetation information in the study area,and provide a scientific basis for the reasonable evaluation of the regional land desertification status.

Key words: high-resolution No.2 remote sensing image, supervised classification, object-oriented classification, deep learning, multi-scale segmentation