Feasibility of Machine Learning Methods for Separating Wood And Leaf Points From Terrestrial Laser Scanning Data

Abstract

ABSTRACT:Classification of wood and leaf components of trees is an essential prerequisite for deriving vital tree attributes, such as wood mass, leaf area index (LAI) and woody-to-total area. Laser scanning emerges to be a promising solution for such a request. Intensity based approaches are widely proposed, as different components of a tree can feature discriminatory optical properties at the operating wavelengths of a sensor system. For geometry based methods, machine learning algorithms are often used to separate wood and leaf points, by providing proper training samples. However, it remains unclear how the chosen machine learning classifier and features used would influence classification results. To this purpose, we compare four popular machine learning classifiers, namely Support Vector Machine (SVM), Naïve Bayes (NB), Random Forest (RF), and Gaussian Mixture Model (GMM), for separating wood and leaf points from terrestrial laser scanning (TLS) data. Two trees, an Erytrophleum fordii and a Betula pendula (silver birch) are used to test the impacts from classifier, feature set, and training samples. Our results showed that RF is the best model in terms of accuracy, and local density related features are important. Experimental results confirmed the feasibility of machine learning algorithms for the reliable classification of wood and leaf points. It is also noted that our studies are based on isolated trees. Further tests should be performed on more tree species and data from more complex environments.

Improved Supervised Learning-Based Approach for Leaf and Wood Classification From LiDAR Point Clouds of Forests

“…1) Random Forest: One of the main reasons for choosing RF for building the leaf-wood classification model is that RF is robust to outliers, and hence avoids overfitting by building multiple parallel decision trees using random subsets of data and random subsets of features for each tree [35]. A recent study has shown that RF outperformed other individual machine-learning classifiers like Naive Bayes, neural networks, and so on, for leaf and wood classification of individual trees based on geometric features in temperate forests [11]. To choose the optimal number of decision trees to build the model, we analyzed the performance of the models built with 10-80 decision trees by cross-validation.…”

“…In the past decade, there have been significant advancements in the methods for leaf and wood separation from TLS data. The methods developed were either based on the radiometric features [10] or geometric features [8], [11]- [16] or a combination of both [17]. Since the radiometric features mainly depend on the wavelength used by a particular sensor, methods based on radiometric features become sensorspecific.…”

“…Both the eigenvectors and normal vectors are calculated for each point from its local neighborhood. The local neighborhood of a point can be defined either by specifying a fixed number of nearest neighbors for that point [11], [15] or by specifying a sphere of fixed radius with that point at the center called radially bounded nearest neighbors [8]. So far, the most common way of defining the neighborhood has been to specify the K-nearest neighbors, with K often ranging between 10 and 100 [14], [15], as it is more computational and memory-efficient than the radially bounded neighbors.…”

“…We defined the local neighborhood of a point at multiple spatial scales, as it eliminates the need for the selection of optimal neighborhood size for a data set. Most of the existing methods have used classical machine learning algorithms like RF [11], support vector machines (SVMs) [11], and Gaussian mixture models (GMMs) [8], [15] to classify leaf from wood points. Recent advancements in gradient boosting machines (GBMs) have made them the most popular choice for classification problems, and they have shown to outperform most of the other classical machine learning algorithms [22].…”

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

“…The drawback is that they require laborious and time consuming manual selection of training data. In addition, the distribution of training data greatly impacts the overall performance of machine learning methods (Wang, Hollaus, & Pfeifer, 2017). A combination of geometric and radiometric features may have advantage over using only one type (Zhu et al, 2018).…”