Compensation Model of GF-7 Panchromatic and Multispectral Image Registration Error

  Objectives  Gaofen-7 (GF-7) is the first 1∶10 000 stereo mapping satellite of China that can produce high-resolution fusion products. The fusion products are fused with the 0.65 m-panchromatic images and 2.6 m-multispectral images. However, in some fusion products, halos and shifts are existed.

  Methods  To deal with this problem, we analyze the cross-sensor error property of push-broom optical satellite and deduce a compensation model that induces a linear term to compensate for the internal and external orientation errors, a quartic piecewise polynomial term to rectify the lens-distortion and the sub-CCD(charge coupled device) placement error, and a trigonometric series term to correct for the errors caused by platform jitter.

  Results  Several scenes of GF-7 (6 in Guangdong and 3 in Qinghai) were used to demonstrate the compensation model. The results show that the registration RMSE(root mean square error) decreased from 0.6 multispectral-pixels to about 0.07 multispectral-pixels by solving and applying the compensation model, and proportion of areas which have less-than 0.25 multispectral-pixels registration error increased to about 98%, which meets the requirement of producing high-quality fusion products.

  Conclusions  The quantitative results not only reveal the feasibility to improve the accuracy of GF-7 fusion products used by first-level data and achieve the theoretical limit accuracy, but also demonstrate the stability of the piecewise polynomial term allowing for its repeated application to further enhance compensation efficiency.