The feature points extracted by traditional methods in UAV navigation are susceptible to external factors such as changes in lighting and viewing angles， and the low matching accuracy and poor robustness of the algorithm in complex environment lead to large errors in pose estimation. To address these issues， this paper proposes a semantic feature matching algorithm that integrates multiple types of information. The algorithm replaces traditional feature points with semantic feature points to enhance robustness and repeatability of point extraction. To optimize the accuracy of the description， we establish the initial representations of feature points via positional， semantic， and descriptive information. Subsequently， a graph neural network incorporating a fusion attention mechanism is used to aggregate the information from other semantic feature points， so as to create the final representation. Finally， the comprehensive matching degree of feature point pairs is calculated based on semantic and characterization information to enhance the accuracy of feature matching. Experimental results demonstrate that in UAV flight scenarios， the proposed algorithm exhibits better robustness and feature matching accuracy compared to traditional algorithms. Moreover， it shows strong robustness to changes in lighting and large viewing angles， providing more reliable and accurate landmark matching for back-end UAV visual pose estimation.