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

Dominik Seidel,Jonas Glatthorn,Martin Ehbrecht,Quentin Edward Seifert,Thomas Kneib,Jan-Henrik Thauer,Anton Thielman,Christian Ammer,Peter Annighöfer

doi:10.3389/fpls.2021.635440

Dominik Seidel, Jonas Glatthorn + Show 7 more

Open Access

https://doi.org/10.3389/fpls.2021.635440

Copy DOI

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.

Highlights

Many functions and services of a forest are tied to forest structure and the structure of the individual trees that constitute it
After image creation from ten perspectives per tree and additional image augmentation for underrepresented tree species, our data consisted of 4040 (50%) deciduous and 4060 (50%) coniferous tree images, with some images used for training and the remainder used for testing
The overall accuracy of our approach was promising (86%) and we argue this is because the transformation of 3D to images enabled us to make use of the strong, already existing image classification techniques based on convolutional neural networks (CNNs)

Summary

Introduction

Many functions and services of a forest are tied to forest structure and the structure of the individual trees that constitute it. Structural information is relevant for monitoring deforestation (Goetz and Dubayah, 2011), estimating carbon stocks (Asner, 2009) or predicting biodiversity (Bergen et al, 2009; Dees et al, 2012), and to enable more accurate models of microclimatic conditions (Ehbrecht et al, 2017), the carbon cycle (Xiao et al, 2019), water cycle (Varhola and Coops, 2013), and other tasks. For an optimized and goal oriented forest management, detailed information on the stand structure is essential. Highresolution 3D data on individual trees is available for larger areas (Koch et al, 2006; Liang et al, 2014) and provides the opportunity to aid research in forest ecology (Danson et al, 2018; Disney, 2019), tree architecture modeling (Bucksch and Lindenbergh, 2008; Dorji et al, 2019), and to support forest management in an unprecedented way (Hirata et al, 2009). Two major challenges must be overcome if 3D data of forests is to be used operationally on a larger scale

Results

Discussion

Conclusion