Segmentation of LiDAR point cloud data is a fundamental challenge in LiDAR data processing, directly impacting the accuracy of information retrieval. This paper introduces an enhanced SAL (Segment Anything in LiDAR) method for labeling LiDAR points without manual supervision, leveraging pseudo-labels to reduce the number of points required for model training. The proposed approach integrates point features and an encoder step using tokens generated from feature extraction and CLIP (Contrastive Language-Image Pre-training) embeddings. Experimental results on a real-world UAV LiDAR dataset collected in Hoa Binh, Vietnam, demonstrate segmentation accuracy exceeding 0.9, validating the method's effectiveness for semantic segmentation of UAV LiDAR point clouds.