Semantic segmentation of airborne point clouds provides essential data support for downstream applications. Fully supervised deep learning methods typically rely on large amounts of annotated data， while some weakly supervised approaches struggle to learn representative features effectively due to the randomness in label selection. To address these challenges， a label-efficient semantic segmentation method is proposed， which integrates an active learning strategy to progressively update the training set by actively selecting the most informative points based on information entropy in each learning cycle. Experimental results on the LASDU and H3D datasets show that， with only 0.5% and 0.1% labeled data， the proposed method outperforms existing approaches in segmentation accuracy， demonstrating its efficiency in weakly supervised conditions.