Far-infrared images have problems such as low resolution, large numbers of repetitive structures and sparse structures. The stitching of this type of far-infrared image faces many challenges. Aiming at the far-infrared image stitching in aerial scenes, this paper divides the image alignment process into sequential image sequence alignment and multicolumn image alignment. Single-column images are based on nonmaximum suppression to obtain the homography matrix, and the transfer relationship is derived for splicing. The image stitching between columns is divided into grids, combined with regional similarity transformation and local homography transformation to optimize the weight of the transformation matrix grid by grid and through the recursion of grid transformation to achieve precise stitching. The methods used can avoid the problem of disappearance of splicing features of multiple consecutive far-infrared images and adapt to the problem of inconsistent overlap rate when in multicolumn image stitching, realizing the stitching of massive far-infrared images.