Abstract: In strip mode synthetic aperture radar (SAR) imaging, the data processing volume is large, and it is usually divided into multiple sub-aperture processing tasks. However, miniature unmanned aerial vehicle (mini-UAV) SAR often introduces large motion errors due to its unstable operation, which not only causes deformation of sub-aperture images, but also makes it difficult to accurately estimate the translation amount between adjacent images, thereby increasing the difficulty of image stitching. The feature points provided by scale-invariant feature transform (SIFT) algorithm can effectively deal with the problem of image registration and stitching, but the efficiency of traditional process is low when processing images with large amounts of data. Therefore, an optimization method for image stitching based on SIFT features is proposed in this paper to improve the efficiency of SAR image registration and stitching. This paper introduces a feature point quality evaluation standard based on amplitude ratio. By selecting feature points, the accuracy of matching is ensured and the number of feature points is effectively reduced. On this basis, KD tree is used for coarse matching of feature points to improve the retrieval speed. In addition, two one-dimensional interpolations are used to replace the traditional two-dimensional interpolations to optimize the interpolation efficiency of the affine transformation. The affine matrix is estimated and corrected by the pixel-reduced images to optimize the computational efficiency and ensure the quality of the stitching. Based on the comparison of the experimental time, registration accuracy, similarity, mean square error and other indicators, it is verified that the proposed method can significantly improve the computational efficiency while maintaining the stitching accuracy, and has certain application value for the rapid stitching of mini-UAV SAR images.