An airport detection method is proposed for the navigation of fixed-wing unmanned aerial vehicle (UAV) autonomous landing in this paper, which aims at improving the efficiency of detection. A hierarchical architecture is adopted to obtain airport candidate regions which reduces the search space gradually. The pseudo horizon is detected to limit the searching space to the ground area, then candidate approximate airport area is acquired based on the fact that the airport area contains lots of orthogonal line segments. Edge Boxes is adopted to obtain proposals with good localization on the candidate approximate airport areas. Locality-constrained linear coding (LLC) is used for feature extraction with scale-invariant feature transformation (SIFT) as the basic features and linear support vector machine (SVM) is used to finish the task of airport detection. We evaluate the proposed method under different conditions and compare it with other methods. The results show that our method improves the efficiency of airport detection and has a higher average precision.