LeWoS: A universal leaf‐wood classification method to facilitate the 3D modelling of large tropical trees using terrestrial LiDAR

Abstract

Leaf‐wood separation in terrestrial LiDAR data is a prerequisite for non‐destructively estimating biophysical forest properties such as standing wood volumes and leaf area distributions. Current methods have not been extensively applied and tested on tropical trees. Moreover, their impacts on the accuracy of subsequent wood volume retrieval were rarely explored. We present LeWoS, a new fully automatic tool to automate the separation of leaf and wood components, based only on geometric information at both the plot and individual tree scales. This data‐driven method utilizes recursive point cloud segmentation and regularization procedures. Only one parameter is required, which makes our method easily and universally applicable to data from any LiDAR technology and forest type. We conducted a twofold evaluation of the LeWoS method on an extensive dataset of 61 tropical trees. We first assessed the point‐wise classification accuracy, yielding a score of 0.91 ± 0.03 in average. Second, we evaluated the impact of the proposed method on 3D tree models by cross‐comparing estimates in wood volume and branch length with those based on manually separated wood points. This comparison showed similar results, with relative biases of less than 9% and 21% on volume and length respectively. LeWoS allows an automated processing chain for non‐destructive tree volume and biomass estimation when coupled with 3D modelling methods. The average processing time on a laptop was 90s for 1 million points. We provide LeWoS as an open‐source tool with an end‐user interface, together with a large dataset of labelled 3D point clouds from contrasting forest structures. This study closes the gap for stand volume modelling in tropical forests where leaf and wood separation remain a crucial challenge.

An Improved Convolution Neural Network-Based Model for Classifying Foliage and Woody Components from Terrestrial Laser Scanning Data

“…and needle clusters with near-vertical leaf inclination angles were always classified as woody materials. Previous studies obtained similar results when using geometric-based classification models [10,32]. When combining the LRI-and geometric-based features to separate the foliagewoody components from TLS data, the OA increased at least 5%.…”

“…where P 0 means the relative observed agreement among N samples in error matrices with r rows, which is equal to the ratio of the sum of the correct identification samples of all samples belonging to the same type; the P c is the hypothetical probability of chance agreement; X i+ and X +i are the row probabilities and the column probabilities, respectively. To assess the contribution of combining the LRI and geometrical information in distinguishing TLS data, we used the LeWoS model [32] and LWCLF model [15], which are only based on geometrical features, to classify the TLS datasets, and evaluated their classification accuracy for foliage and woody components. Additionally, we compared the FWCNN-based results with those obtained using the Random Forest (RF) algorithm [48], Gaussian Mixed Model (GMM) [30], and Support Vector Machine (SVM) algorithm [49] in terms of classification accuracy and running efficiency.…”

“…Some supervised classification models have been successfully developed, such as the Random Forest (RF) algorithm [11], Gaussian Mixed Model (GMM) [30], and Support Vector Machine (SVM) algorithm [31], and have achieved high classification accuracy. In the meantime, unsupervised clustering methods, such as the DBSCAN algorithm [10] and LeWoS model [32], have been proposed for separating foliage and woody points in TLS datasets collected in forests.…”

Sensor Agnostic Semantic Segmentation of Structurally Diverse and Complex Forest Point Clouds Using Deep Learning

“…In the case of this paper, we are focused on separating parts of a forest into terrain, vegetation, coarse woody debris, and stem categories from a point cloud. There have been many different approaches to the segmentation of forest point clouds [20][21][22][23][24][25][26][27][28][29][30][31][32][33][34] so far. Some approaches use heuristics [20,22,25,28,29] or morphological operations [27], while others use supervised [23,26,[30][31][32][33] or unsupervised [21,34] machine learning techniques.…”

Mapping forest leaf area density from multiview terrestrial lidar

“…Within each voxel, the interaction of laser pulses with material is interpreted to derive two types of information: the nature of the material present (leaf or wood) and the leaf area density (in m 2 /m 3 ). The former can be obtained using different methods like reflectance threshold (Béland, Baldocchi, et al., 2014), semantic segmentation (Wang et al., 2020) or machine learning (Krishna Moorthy et al., 2019). The latter relies on theoretical approaches using the contact frequency (Béland et al., 2011; Hosoi & Omasa, 2006; Warren Wilson, 1959), maximum likelihood estimation (Pimont et al., 2018), or the Beer‐Lambert law (Béland et al., 2014; Ross, 1981).…”