Extraction Of Land Use/Cover Information Based On Object-Oriented Technology

With the continuous development of the remote sensing technology, especially the rapid development of new sensors, there are more and more high spatial resolution images. At the same time, researches on information extraction in land cover from it are also increasing. Compared with low or moderate spatial resolution remote sensing image, high spatial resolution sensing image has richer structure information and texture information. But now its utilization rate is not high. It is mainly because that traditional classification methods are based on spectral of pixels and ignores other information in the images, which produce a lot of broken and invalid pattern spots in classification result, thus lead to a low accuracy. So we often improve classification accuracy by artificial visual interpretation. However, it requires much manpower and time. In general, the traditional classification methods no longer has the universal applicability of high spatial resolution sensing image.In 2013,China successfully launched "GF-1" satellite which is the first satellite of the high resolution earth observation system in our country. It can provide important data to support the studies for land use situation. But the researches are still insufficient at present. So in order to solve the problems above,explore the applicability of "GF-1" image of monitoring urban land applicability,and more efficiently and accurately extract the land cover information. Based on GF-1 image in Leshan, land cover information was extracted by object-oriented classification technology in the paper. And some features are added into the image in order to obtain the influence of classification accuracy. At the same time, traditional pixel-based classification results were compared with the object-oriented classification results by subjective visual evaluation and object accuracy evaluation. The main research results for this paper are as follows:(1) A lot of fusion methods Sharpening were used for the study area image. The results were evaluated index of Information entropy, Average gradient, Relative deviation, Spectral Angle and Spectral information divergence. By synthesizing the visual evaluation and calculation results, the NNDiffuse is confirmed to be the best fusion method in the study area.(2)Through the combination of the texture feature, NDVI and multispectral image, a new remote sensing image was obtained. Then the unsupervised classification and supervised classification experiments were conducted with the help of the new sensing image and multispectral image. And it showed that the pixel-based classification method with multi-feature can obtain more accurate land cover information.(3) The multi-scale segmentation experiment was conducted in the eCognition software with different segmentation schemes. The optimal segmentation scale and parameter for the study area was obtained. The multi-scale segmentation hierarchy(80,30) with two levels was set up, so did the classification system. Finally the object-oriented classification experiments based on multi-spectral image and the image added multi-feature were completed. It showed that the object-oriented classification method is better.(4) Compared with the traditional pixel-based classification method, whether based on multispectral images or the image added multi-feature, the object-oriented classification technology is more efficient and accurate, and can get more ideal and practical results.