The algorithm of UAV visual navigation based on attention mechanism for feature matching is becoming increasingly mature. However, relying solely on the attention mechanism to align feature descriptors makes it difficult to maintain algorithm stability and lacks sufficient utilization of keypoints information during the feature matching process. To address this issue, a UAV visual navigation technology based on feature confidence redistribution attention is proposed. Firstly, an attention weight matrix is constructed through the confidence of keypoints, secondly attention information is redistributed according to the weight matrix to increase the weight of feature salient points in the calculation of descriptor relevancy and reduce the weight of keypoints with weaker corner features in the attention information. Finally, feature descriptors with redistributed attention are aligned and merged to enhance the robustness of image matching in UAV visual navigation. The algorithm proposed in this paper was validated through UAV flight tests in three common scenes: plains, Gobi, and mountains. The results show that the proposed algorithm significantly improves performance in various scenes. Compared to traditional algorithms, the average success rate of visual navigation increased by 18.27%, and the average positioning error decreased by 52.92%.