TREEDETECTION: AUTOMATIC TREE DETECTION USING UAV-BASED DATA

Ângela Maria Klein Hentz,Emerson Roberto Schoeninger,Ana Paula Dalla Corte,Sylvio Péllico Netto,Michael P Strager

doi:10.5380/rf.v48i3.56150

Ângela Maria Klein Hentz, Emerson Roberto Schoeninger + Show 3 more

Open Access

https://doi.org/10.5380/rf.v48i3.56150

Copy DOI

Abstract

In this study it is presented a toolbox built in ArcGIS using ArcPy designed to automatically detect trees in high resolution data obtained from Unmanned Aerial Vehicles (UAV). The toolbox, TreeDetection, contains a tool called TreeDetect, which requires three parameters: a raster input, a conversion factor and an output folder. Three other optional parameters can be changed to improve the detection according to characteristics of the forest and raster source. We tested the TreeDetect tool in three study sites: a young Eucalyptus plantation; adult Eucalyptus and Pinus stands; and a Mixed Hardwood natural forest. We also tested distinct raster inputs, according to the data availability in each site. The tool was considered efficient to detect the trees in the three study areas. The detection accuracy was lower in the natural stand, as expected considering the complex structure of this forest type. All the raster input rested provided satisfactory results, but in the homogeneous stand the Digital Surface Model (DSM) was not as effective as the spectral bands. Furthermore, research can be performed with emphasis in different sensors and band combinations, as well in the parameters’ selection.

Highlights

In this study, we present a toolbox built in ArcGIS, by using ArcPy, and designed to automatically detect trees in high resolution data obtained from Unmanned Aerial Vehicles (UAV)
The use of remote sensed data in forest management is a common practice (TANG; SHAO, 2015), since it allows the collection of a large amount of information about the environment with less human effort in field
The largest errors were observed in the mixed hardwood natural forest, since this type of forest is extremely complex (VAUHKONEN et al, 2012; GULBE et al, 2013; TANHUANPÄÄ et al, 2016)

Summary

Introduction

The use of remote sensed data in forest management is a common practice (TANG; SHAO, 2015), since it allows the collection of a large amount of information about the environment with less human effort in field. The most common Remote Sensing sources have their own disadvantages, usually related to the price and concerning high resolution imagery and Lidar data (KE; QUACKENBUSH, 2011; BALDAUF; GARCIA, 2016; GOMES; MAILLARD, 2016), as well as limitations of temporal coverage and data availability, according to weather conditions for passive sensors (TANG; SHAO, 2015). Because of these limitations and the restricted knowledge on some Remote Sensing techniques, the traditional field forest inventories are still the most common practice in forest companies.

Methods

Results

Conclusion