Preliminary study on land use classification based on multi-source remotely sensed data fusion technology

High spectral resolution, high spatial resolution and high temporal resolution are the development trend of modern remote sensing technology. In the case of same image data, it is difficult to obtain high spatial resolution and hyperspectral remote sensing data information simultaneously. This weakness can be remedied, the quality of the image can be improved and the useful thematic information also can be highlighted by fusion of remote sensing image which is from the different sensors. In this paper, Multiplicative Transform, Principal Component Transform and Brovey Transform had been used to fuse the high spatial resolution radar data- COSMO-SkyMed, IRS-P5 data and high spectral resolution data of multi-spectral image -IKONOS, image respectively. And then, the results of fusion accuracy used to land use classification had been evaluated. The results show that: (1) the standard deviation of the fused image by the multiplicative transformation of COSMO-SkyMed radar data and IKONOS multi-spectral data is 21.327 and the relevant index is 0.565675, so the quality of the fused image is the most optimal. The standard deviation of the fused image by the principal component transformation of IRS-P5 data and IKONOS multi-spectral data is 26.506 and the relevant index is 0.56842, which indicate that the quality of the fused image is the best. (2) The classification accuracy of using original image data is not high (the classification accuracy without visual correction is 68.75%). The accuracy classification which using fused image was apparently improved. The classification accuracy of the fused image (without visual correction) which by using the fused image of COSMO-SkyMed radar data and IKONOS multi-spectral data is higher than that of the fused image of IRS-P5 data and IKONOS multi-spectral data. The classification accuracy of former (without the visual correction) is 81.82%, and the classification accuracy of latter (without the visual correction) is 73.33%. It shows t- - hat the former is more conducive to extract the information of the thematic features.