Multispectral image registration and accuracy analysis of ZY-3 satellite

While using multispectral remote sensing images for quantitative application, registration accuracy between spectrums will directly affect the accuracy of each quantitative production, thus, precise multispectral image registration is a base step for further processing and application. ZY-3 is China's first civilian mapping satellite, which attracts extensive attention for its excellent positioning and mapping ability as well as high precision of radiation characteristic, and greatly enhances the applications of domestic satellite. However, for the spectral camera optical system carried on ZY-3 satellite, there exists the problem of geometry deformation, which seems to be inconsistent for each spectrum in the “Multi-Spectrum Integral” device. Therefore, traditional image registration method does not show high accuracy, and “False color” phenomenon exists in the edge of image field of view. In order to solve this problem, a method of the virtual CCD line array re-imaging algorithm on spectrum registration will be introduced in this paper by analyzing the effects of displacement in image caused by virtual CCD re-imaging; the validation of this method will also be verified in principle. Based on ZY-3 spectrum registration experiment in AnPing, HeBei, the visual evaluation shows that the “False color” phenomenon in edge of multispectral image could be effectively solved. Precise evaluation has been performed to the registration of multispectral image products using artificial targets. it can be seen from the result, the maximum error of relative coordinates difference for each spectrum is 0.25 pixels, standard deviation in CCD line direction is better than 0.15 pixels. While it is 0.1 pixels on orbit direction, which is comparative accuracy with that of the manual cases. The experiment verifies that it's feasible to produce multispectral images by using virtual CCD line array re-imaging algorithm, and high accuracy is achieved in ZY-3 multispectral image registration.