Classification of UAV point clouds by random forest machine learning algorithm

Abstract

Unmanned aerial vehicle Point cloud Classification Random forestToday, unmanned aerial vehicle (UAV)-based images have become an important data sources for researchers who deals with mapping from various disciplines on photogrammetry and remote sensing. Reconstruction of an area with three-dimensional (3D) point clouds from UAV-based images are an essential process to be used for traditional 2D cadastral maps or to produce a topographic maps. Point clouds should be classified since they subjected to various analyses for extraction for further information from direct point cloud data. Due to the high density of point clouds, data processing and gathering information makes the classification of point clouds a challenging task and may take a long time. Therefore, the classification processing allows an optimal solution to acquire valuable information. In this study, random forest machine learning algorithm for classification processing is applied with radiometric features (Red band, Green band and Blue band) and geometric characteristics derived from covariance feature (curvature, omnivariance, flatness, linearity, surface variance, anisotropy and normalized terrain surface) of points. In addition, the case study is presented in order to test applicability of the proposed methodology to acquire an accuracy and performance of random forest method on the UAV based point cloud. After the classification processing, a class assigned each point from the model was compared with the reference data class. Lastly, the overall accuracy of the classification was achieved as 96% and the Kappa index was reached to 91% on data set.

Influence of Spatial Resolution for Vegetation Indices’ Extraction Using Visible Bands from Unmanned Aerial Vehicles’ Orthomosaics Datasets

“…Thus, the regression output from RF effectively represents the average of all regression trees grown in parallel without pruning [59]. The iterative nature of RF gives it a distinct advantage over other methods in that the data is effectively bootstrapped, thus feeding random subsets of the training data, to obtain more robust predictions and reducing the correlation between trees [60].…”

Effectiveness of Training Sample and Features for Random Forest on Road Extraction from Unmanned Aerial Vehicle-Based Point Cloud

“…Five commonly used measures were produced from confusion matrix elements. They are recall, precision, quality, accuracy, and F1-score (62,63) and are defined as follows. F1 À Score =…”

Data for 3D reconstruction and point cloud classification using machine learning in cultural heritage environment

“…The evaluation of the quality of the point cloud was carried out using four performance indices, such as Precision, Recall, F1 measure and Overall Accuracy [ 8 , 9 ]. Therefore, taking into account the point cloud under investigation, two columns were annotated for the training phase, while the remaining 3 columns were annotated for the evaluations.…”