Multi - Scale Remote Sensing Classification Of Land Cover

Land cover change is an important part of the global change research. Remote sensing technology is the main method of land cover information acquisition, and remote sensing image classification is a very important part in the land cover information acquisition.With the development and application of remote sensing technology, scaling,scale effect and optimal scaling choice as the main content to scale transformation,which were gradually become an important research direction. In this paper, SPOT5 HRG image, GF-1 image and landsat8 OLI image as the data source, taking Shangri-La as the study area, the scale of remote sensing classification of SPOT5 HRG images, GF-1 image and landsat8 OLI image was studied. First of all, the fusion of remote sensing image data for scale conversion, scale conversion effect evaluation, the best way to get the best scale conversion method. Secondly, the optimal scale of various types of objects in the study area was selected, and the remote sensing image data based on multi scale overlay were classified by remote sensing data. Finally, based on this method, the Landsat8 OLI remote sensing image data was used to classify the multi scale remote sensing data of Shangri-La city. The main content and conclusions as follows:(1) The method of scale conversion. By the local average method, median sampling method and nearest neighbor method, bilinear in method and the cubic convolution method, all of the five methods, which to scale conversion of SPOT5 HRG and GF-1 image scale conversion, and the four indicators of the mean, standard deviation, average gradient and information entropy of scale conversion effect were analyzed. The results showed that whether it is SPOT5 HRG image or GF-1 image,using cubic convolution method to scale the results obtained from the remote sensing image was better than other methods. Therefore, in the future study of scale conversion, the cubic convolution method should be widely used to carrying on the scale transformation.(2) Optimal scale selection. By calculating the local class of the variation function, the SPOT5 HRG image and GF-1 image of the class to select the best choice. The results showed that the optimal scale of the most types of objects was not on the original scale, but on the larger scale. Among them, the optimal scale of water was the largest, which followed by cultivated land, grassland, scrub and roads and residents, the floodplain of the shadow and remained in the original scale.(3)Multi-scale classification. By using the confusion matrix, the overall classification accuracy and Kappa coefficient, the results of the maximum likelihood classification of the original multi spectral images and the multi scale overlay images were evaluated. The results showed that the overall classification accuracy of classification results of SPOT5 HRG images and GF-1 images multi-scale overlay images are more than the overall classification accuracy of original multispectral images respectively higher than 12.84% and 14.76%, kappa coefficient respectively are 0.1306 and 0.1419. It could be seen that the classification accuracy of remote sensing image classification can be improved, which were based on multi-scale classification.(4) Shangri-La land cover classification. Based on the above method, the Landsat8 OLI remote sensing data in use of Shangri-La land cover classification.Results showed: Multi dimension classification method is also applicable to the Landsat8 OLI image, and is suitable for the whole of Shangri-La. In 2015Shangri-La land cover classification results indicated the most widely distributed forest, shrub, grassland and bare land.Scale problems in remote sensing research is extremely complex, there are still many places for in-depth study. Such as classification system only to second class and only analysis the optimal size for different classes of while ignoring the same class also may optimal scale is not the same. For example, experiments in other areas of applicability. All these need to be further studied in the future.