Individual tree segmentation via contrastive learning and semantic priors in point clouds

Abstract

As vital components of ecosystems, trees play a significant role in ecological assessment and optimization due to their structural characteristics. Accurate segmentation of individual trees is a critical procedure in this task. However, traditional manual methods are labor-intensive and resource-demanding. In contrast, individual tree segmentation based on LiDAR point cloud data offers a practical and efficient solution. While recent advancements in deep learning-based point cloud instance segmentation and tree detection have achieved remarkable progress, previous methods have focused on semantic segmentation and ignored instance-level tree recognition in both urban and forest environments. Furthermore, the overlapping of tree crowns makes it difficult to accurately delineate individual trees from point clouds, posing a persistent challenge for achieving high accuracy and efficiency in individual tree extraction. To address these challenges, we propose an effective individual tree segmentation method capable of accurately extracting single trees in urban and forest scenes. The proposed approach consists of two primary steps: (1) We design the Semantic-Driven Instance Clustering to combine the semantic prior with the instance embeddings. (2) We introduce the Online Semantic Clustering for intra-class potential semantic discriminability, improving the instance representation within the same semantic class. The method is evaluated and validated on point cloud datasets from urban and forest environments, demonstrating its ability to produce accurate individual tree segmentation. The F1-score achieves 80.26% and 79.5% in Paris-Lille-3D and FOR-instance datasets, respectively, demonstrating the effectiveness of our approach. Building upon the segmented individual trees, we further estimate key 3D tree parameters to support subsequent vegetation inventory, management, and sustainable development applications, providing theoretical and methodological support for policy-making, planning, and design.