PCTreeS: 3D Point Cloud Tree Species Classification Using Airborne LiDAR Images

TL;DR

This work addresses scalable, accurate tree species mapping in forests by evaluating a traditional 2D projection CNN baseline against a novel 3D Point Cloud Transformer classifier (PCTreeS) on airborne LiDAR images from Africa's tropical savannas. The PCTreeS model processes raw 3D LiDAR point clouds without projecting to 2D, and it outperforms the CNN baseline in AUC and overall accuracy while reducing training time. Using ForestGEO Mpala census data and airborne LiDAR, the study demonstrates the viability of 3D vision-transformer approaches for large-scale tree species classification and motivates broader LiDAR image collection and validation. The results suggest that PCTreeS can enable more accurate, scalable mapping of tree species in tropical ecosystems, supporting ecological monitoring and carbon accounting at scale.

Abstract

Reliable large-scale data on the state of forests is crucial for monitoring ecosystem health, carbon stock, and the impact of climate change. Current knowledge of tree species distribution relies heavily on manual data collection in the field, which often takes years to complete, resulting in limited datasets that cover only a small subset of the world's forests. Recent works show that state-of-the-art deep learning models using Light Detection and Ranging (LiDAR) images enable accurate and scalable classification of tree species in various ecosystems. While LiDAR images contain rich 3D information, most previous works flatten the 3D images into 2D projections to use Convolutional Neural Networks (CNNs). This paper offers three significant contributions: (1) we apply the deep learning framework for tree classification in tropical savannas; (2) we use Airborne LiDAR images, which have a lower resolution but greater scalability than Terrestrial LiDAR images used in most previous works; (3) we introduce the approach of directly feeding 3D point cloud images into a vision transformer model (PCTreeS). Our results show that the PCTreeS approach outperforms current CNN baselines with 2D projections in AUC (0.81), overall accuracy (0.72), and training time (~45 mins). This paper also motivates further LiDAR image collection and validation for accurate large-scale automatic classification of tree species.

Figures (3)

• Figure 1: Examples of tree species that are observably discernable assuming very little noise.
• Figure 2: Noise in the dataset was caused by four main classes of error.
• Figure 3: A diagrammatic representation of the baseline approach using 2D CNN following allen.