In order to solve the problem that the optical center of hyperspectral images does not coincide with each other, and the alignment of traditional global monotonic matrix may have error in the registration process, a splicing-method (H-SPHP) of hyperspectral images based on SPHP(shape-preserving half-projective) and considering spectral information is proposed. The main steps include as follows: 1) image correction using the visual vector method. 2) selection of reference band based on prior knowledge and PCA. 3) SPHP-based mesh optimization splicing method. 4) Weighted average fusion algorithm for fusion.5) splicing parameters applied to all bands to obtain the splicing hyperspectral data. By obtaining experimental images from Sanming, Fujian province and Nanchang,Jiangxi province, the research data splicing experimental results show that the proposed algorithm has strong robustness, eliminating parallax and geographic coordinates accuracy better than SIFT + single should transform algorithm. After splicing, the registration accuracy of bands is within one pixel, and spectral similarity of the overlapping region is above 90%.