Study On The Classification Of Land Use And Scale Effect Of The Oasis Regions Of South Xinjiang Basin Based On The Multi-source Remote Sensing Images

With the rapid development of computer and remote sensing technology, multi-spectral remote sensing sensors are emerging in an endless stream. Multi-source remote sensing information based on object-oriented multi-scale has become an important part in the research of remote sensing classification. Especially, with the appearance of object-oriented classification method, the scale effect of multi-source remote sensing image segmentation has become the key to efficient extraction of remote sensing data in a large extent. Economic forest as the main body of an oasis in southern Xinjiang region as the study area, this study used the object-oriented method to carry out (0.61 m,2 m,2.4 m,8 m,16 m and 30 m) 6 kinds of spatial resolution remote sensing image multi-scale segmentation. The optimal segmentation scale and the best temporal of multi-source remote sensing data are defined, and the land use classification of study area was carried out by using multiple classifiers, in order to provide technical reference for the study of land use classification based on remote sensing. The main conclusions are as follows:(1) Multi-source remote sensing image segmentation scale effect analysis shows that the scale image segmentation of oasis region with economic forest as the main body in southern Xinjiang, increases with the increase of image spatial resolution under different image spatial resolution. When the image spatial resolution is consistent, the segmentation scale increases with the increase of the area of the object, and decreases with the decrease of the area of the object. Furthermore, the remote sensing image of 2.4 m spatial resolution has best effect on the land-use type partition of small, medium, large area among the remote sensing images of 0.61 m,2 m,2.4 m,8 m,16 m,30 m spatial resolution, which segmentation scale is set as 60,80,120, and segmentation parameter is set as Shape 0.9, Compactness 0.1.(2) Changes of temporal and the response analysis of land use types shows that within the oasis region with economic forest as the main body in southern Xinjiang, there is no obvious difference when segmentation scale, image segmentation degree and the area of image object changes along the temporal,at different temporal (April to September), in the remote sensing image of the same spatial resolution. However, in different temporal time, there is obvious difference of the NDVI value of ground feature in August. By the evaluation of classification precision, this paper believes that August is the best time for the exaction of the ground feature of Southern Xinjiang oasis area, with the overall classification accuracy of up to 83.32%, under this temporal, there is obvious difference between various land-use types, which is most suitable for land use classification.(3) The analysis of edge feature extraction in oasis area based on object-oriented shows that with the combination of the nearest neighbor classification and membership functions classification, the ground feature exaction and identification effect within the research scope are the best, with the overall classification accuracy of 93.08%, Kappa coefficient of 0.71. compared with Support Vector Machine, the overall classification precision is 10.83%higher, compared with Regression Tree method, the overall classification precision is 20.28%higher.