In order to satisfy the requirements of UCAVs in cognitive navigation on high matching score and low error rate for feature point matching a fused matching method for SURF feature points was proposed.Analysis was made to the performance of purifying matching pairs for the ratio method and correlation method based on which the two methods were fused together and the matching pairs were purified by partitioning the threshold intervals.Simulation results show that the fused method can obtain matched feature points with high matching score and low error rate and can effectively compensate the deficiency that can’t choose the threshold clearly without prior knowledge.