Predicting Tree Species From 3D Laser Scanning Point Clouds Using Deep Learning

Abstract

Automated species classification from 3D point clouds is still a challenge. It is, however, an important task for laser scanning-based forest inventory, ecosystem models, and to support forest management. Here, we tested the performance of an image classification approach based on convolutional neural networks (CNNs) with the aim to classify 3D point clouds of seven tree species based on 2D representation in a computationally efficient way. We were particularly interested in how the approach would perform with artificially increased training data size based on image augmentation techniques. Our approach yielded a high classification accuracy (86%) and the confusion matrix revealed that despite rather small sample sizes of the training data for some tree species, classification accuracy was high. We could partly relate this to the successful application of the image augmentation technique, improving our result by 6% in total and 13, 14, and 24% for ash, oak and pine, respectively. The introduced approach is hence not only applicable to small-sized datasets, it is also computationally effective since it relies on 2D instead of 3D data to be processed in the CNN. Our approach was faster and more accurate when compared to the point cloud-based “PointNet” approach.

The shape of trees: Reimagining forest ecology in three dimensions with remote sensing

“…Ayrey & Hayes, 2018; Chen et al, 2021; Krisanski et al, 2021; Luo et al, 2021; Xi & Hopkinson, 2021) provide the most promising way forward to increase automation within the processing workflows needed to make use of three‐dimensional data practical at scale. For example, neural network approaches have been used to automate tree crown detection from both RGB aerial imagery (Bosch, 2020; Weinstein et al, 2019), and from aerial LiDAR (Windrim & Bryson, 2020), and to identify species based on whole tree point clouds (Seidel et al, 2021), stem and bark properties (Mizoguchi et al, 2019) and processed, interpretable features (Terryn et al, 2020). The additional inclusion of ecologically realistic information to constrain processing algorithms, including through the use of scaling rules, can further improve performance (Brummer et al, 2021; Tao et al, 2015); however, the need for training data to build these models means that increased data sharing across the community may be needed, as well as adoption of approaches such as transfer learning and data augmentation.…”

Hyperspectral and Lidar Data Applied to the Urban Land Cover Machine Learning and Neural-Network-Based Classification: A Review

“…Data augmentation helps diversify training data without new labeling costs, thus leading to more robust classification and adequate classification. In remotely sensed-based classification, training data have been flipped and rotated [225,226], mirrored across horizontal, vertical, and diagonal axes on HS [226,227] and lidar data [228], mixup strategy [229], and generation of virtual training samples through Generative Adversarial Networks (GANs) [230] on HS data. In addition, noise is proven to be suited as a data augmentation type.…”

Estimation of Northern Hardwood Forest Inventory Attributes Using UAV Laser Scanning (ULS): Transferability of Laser Scanning Methods and Comparison of Automated Approaches at the Tree- and Stand-Level

“…Integrating methods that can delineate or merge trees based on the analysis of similarities between segments (e.g., [113,117]) and their assignment to a specific canopy layer also present some potential for processing high-density point clouds (e.g., [20,127]). Last, including some evaluation criteria that are based on dendrometric criteria and machine learning (e.g., [123,128]) should also be further explored to really benefit from gleaning the full information that is available in high density ULS data.…”