Considering the low real-time performance and the large amount of false matches exist in the feature matching stage of traditional Scale Invariant Feature Transform(SIFT) algorithm in unmanned aerial vehicle (UAV) remote sensing image registration. In this paper, a series of optimization methods for traditional SIFT algorithms are proposed, including SIFT execution process optimization, changing the parameters of scale space construction, Simplified method for judging the construction area of feature descriptor and construction of bidirectional matching filters. Experiments on UAV remote sensing images show that the optimization method can significantly improve the matching efficiency compared with traditional methods, and the comprehensive acceleration ratio is about 35% to 40%, which proves the effectiveness of the acceleration method.