| Land use and land cover change is closely related with global climate change, biodiversity loss, environmental change, ecological safety and human and environmental sustainable development. Especially, farmland has a more special social and academic research significance.For the land use and land cover, remote sensing has become a powerful tool in the real-time identification of land-use and land cover classes and satellite images are commonly used in land and crop classification. The problme how to class the landuse and land cover more qucikily and accurately is an impotant part of application reasch of remote sensing.There are many types of remote sensing image sensors, not only 1 km resolution large scale images, but also sub-meter high-resolution images. However, the large-scale images mean that their precision is relatively low, while the high-resolution image is facing the probleme of spatial scale. With the development of technology, the new remote sensing datas are coming forth, which provide new opportunities for the development of landuse and land cover. It is possible to solve the problem of how to make full use of different sensors images, to give full play of the advante of different scale images.In this paper, taking the cultiveatd land for example, the method based on sampling statistics and multi-scale error statistics to improve the accuracy of the large scale images was proposed, while, this paper put forward to another method multi-scale conversion based on the vegetation index, according to the scale transformation thinking and taking sampling statistics as rule. At last by using these two methods to extract the same area of cultivated land, and the two methods were compared and evaluated.Taking the northern of Jiangsu plain for example, the proportion coefficient between the cultivated area in three scale images, which is MODIS (250 m) image, CCD (19.5 m) and HR (2.36 m) of CBERS 2B, were computed by sampling statistics method. Then, the cumulaitve coefficient was used to correct the cultivaed area of MOIDS image in order to increase the area accuracy of large scale image.As the upper method, the error of farmland area between fusion image of HR and CCD was analysised using statistic method by visual interpretation of CBERS-02B HR and NDVI threshold of CCD. Quantitive relationship between NDVI of MODIS images, CBERS-02B CCD and CCD images farmland area percentage was studied. The major results are as follows:(1) Both methods can effectively improve the accuracy of MODIS image cultivated area. Compared to multi-scale conversion based on NDVI, multi-scale multi-scale conversion based on error statistices is more universal and it has an rigorous theory.(2) The cultivated area of high resolution image is about higher 20% than that of CCD imgaes. The reason is the mixed pixel of CBERS CCD image, which is caused byravines, field paths, and the ridge closed to the farmland land In terms of NDVI.(3) There is a significance linear correlation between the mean value of CBERS-02B CCD and MODIS based on region statistical properties, the linear model determination coefficient being 0.98. The model could be used to be as method of NDVI scale advance.(4) The mean NDVI of CBERS-02B in MODIS pixel area and NDVI of MOIDS there is very significant, the correlation coefficient of 0.98. Fitting a linear model can be used to convert NDVI scale.(5) In homogeneous area, multiband image farmland area percentage and NDVI assume linear relation in MODIS pixel area. A Regression Equation was established, and the test of model show that equation is the believable.(6) For (CBERS-02B) HR and CCD image fusion, compared with Brovey transformation, HSV transformation, Gram-Schmidt (GS) transformation, and changes in methods of principal components, SFIM methods is more easy to handle and has more advantage to keep spatial information and spectral information.Finally, the advantege and disadvantege of two kinds of extraction methods was disscued, and some suggestions for that were proposed. |
PDF Full Text Request