Unconstrained approach for isolating individual trees using high-resolution aerial imagery

Taejin Park, Jung Kil Cho, Jong Yeol Lee, Woo-Kyun Lee, Sungho Choi, Doo Ahn Kwak, Moon Il Kim

Research output: Contribution to journalArticle

4 Citations (Scopus)

Abstract

This study outlines an algorithm that can be used for individual tree detection and crown delineation; it was applied to coniferous forest using aerial imagery. This article explains the assumptions and processes involved in the algorithm, presents the results of the applications, and discusses possible limitations. The algorithm, which adopts contextual analysis that excludes the need to specify window size, was applied to detect and delineate individual trees based on morphological and reflective characteristics. The preprocessing steps included suppression of the non-coniferous area (i.e. non-forest and leaf-off deciduous forest) and the creation of appropriately smoothed imagery using an optimal smoothing level based on accuracy index (AI); thereafter, unconstrained directional peak- and edge-finding algorithms were processed separately. To assess the tree detection and crown delineation processes, the results of the algorithms were evaluated carefully against visually interpreted crowns in six square plots using several statistical measures based on tree top correspondence, positional difference of tree top, directional crown width, and crown area assessment. The average tree top correspondence had an AI of 88.83%. The positional difference between detected and visually interpreted tree tops was measured and its average was 0.6 m. For our 0.5 m/pixel aerial imagery, the average root mean square error (RMSE) of crown width in six sample plots was found to be 2.8 m, and crown area estimation resulted in RMSE of approximately 9.23 m2 (23.25%). In general, this study highlights the potentiality of the proposed algorithm to efficiently and automatically acquire forest information such as tree numbers, crown width, and crown area.

Original languageEnglish
Pages (from-to)89-114
Number of pages26
JournalInternational Journal of Remote Sensing
Volume35
Issue number1
DOIs
Publication statusPublished - 2014 Jan 1

Fingerprint

imagery
coniferous forest
smoothing
deciduous forest
pixel
index
detection

ASJC Scopus subject areas

  • Earth and Planetary Sciences(all)

Cite this

Unconstrained approach for isolating individual trees using high-resolution aerial imagery. / Park, Taejin; Cho, Jung Kil; Lee, Jong Yeol; Lee, Woo-Kyun; Choi, Sungho; Kwak, Doo Ahn; Kim, Moon Il.

In: International Journal of Remote Sensing, Vol. 35, No. 1, 01.01.2014, p. 89-114.

Research output: Contribution to journalArticle

Park, Taejin ; Cho, Jung Kil ; Lee, Jong Yeol ; Lee, Woo-Kyun ; Choi, Sungho ; Kwak, Doo Ahn ; Kim, Moon Il. / Unconstrained approach for isolating individual trees using high-resolution aerial imagery. In: International Journal of Remote Sensing. 2014 ; Vol. 35, No. 1. pp. 89-114.
@article{ecadc529de1340488bd13f85c5e51672,
title = "Unconstrained approach for isolating individual trees using high-resolution aerial imagery",
abstract = "This study outlines an algorithm that can be used for individual tree detection and crown delineation; it was applied to coniferous forest using aerial imagery. This article explains the assumptions and processes involved in the algorithm, presents the results of the applications, and discusses possible limitations. The algorithm, which adopts contextual analysis that excludes the need to specify window size, was applied to detect and delineate individual trees based on morphological and reflective characteristics. The preprocessing steps included suppression of the non-coniferous area (i.e. non-forest and leaf-off deciduous forest) and the creation of appropriately smoothed imagery using an optimal smoothing level based on accuracy index (AI); thereafter, unconstrained directional peak- and edge-finding algorithms were processed separately. To assess the tree detection and crown delineation processes, the results of the algorithms were evaluated carefully against visually interpreted crowns in six square plots using several statistical measures based on tree top correspondence, positional difference of tree top, directional crown width, and crown area assessment. The average tree top correspondence had an AI of 88.83{\%}. The positional difference between detected and visually interpreted tree tops was measured and its average was 0.6 m. For our 0.5 m/pixel aerial imagery, the average root mean square error (RMSE) of crown width in six sample plots was found to be 2.8 m, and crown area estimation resulted in RMSE of approximately 9.23 m2 (23.25{\%}). In general, this study highlights the potentiality of the proposed algorithm to efficiently and automatically acquire forest information such as tree numbers, crown width, and crown area.",
author = "Taejin Park and Cho, {Jung Kil} and Lee, {Jong Yeol} and Woo-Kyun Lee and Sungho Choi and Kwak, {Doo Ahn} and Kim, {Moon Il}",
year = "2014",
month = "1",
day = "1",
doi = "10.1080/01431161.2013.862603",
language = "English",
volume = "35",
pages = "89--114",
journal = "International Joural of Remote Sensing",
issn = "0143-1161",
publisher = "Taylor and Francis Ltd.",
number = "1",

}

TY - JOUR

T1 - Unconstrained approach for isolating individual trees using high-resolution aerial imagery

AU - Park, Taejin

AU - Cho, Jung Kil

AU - Lee, Jong Yeol

AU - Lee, Woo-Kyun

AU - Choi, Sungho

AU - Kwak, Doo Ahn

AU - Kim, Moon Il

PY - 2014/1/1

Y1 - 2014/1/1

N2 - This study outlines an algorithm that can be used for individual tree detection and crown delineation; it was applied to coniferous forest using aerial imagery. This article explains the assumptions and processes involved in the algorithm, presents the results of the applications, and discusses possible limitations. The algorithm, which adopts contextual analysis that excludes the need to specify window size, was applied to detect and delineate individual trees based on morphological and reflective characteristics. The preprocessing steps included suppression of the non-coniferous area (i.e. non-forest and leaf-off deciduous forest) and the creation of appropriately smoothed imagery using an optimal smoothing level based on accuracy index (AI); thereafter, unconstrained directional peak- and edge-finding algorithms were processed separately. To assess the tree detection and crown delineation processes, the results of the algorithms were evaluated carefully against visually interpreted crowns in six square plots using several statistical measures based on tree top correspondence, positional difference of tree top, directional crown width, and crown area assessment. The average tree top correspondence had an AI of 88.83%. The positional difference between detected and visually interpreted tree tops was measured and its average was 0.6 m. For our 0.5 m/pixel aerial imagery, the average root mean square error (RMSE) of crown width in six sample plots was found to be 2.8 m, and crown area estimation resulted in RMSE of approximately 9.23 m2 (23.25%). In general, this study highlights the potentiality of the proposed algorithm to efficiently and automatically acquire forest information such as tree numbers, crown width, and crown area.

AB - This study outlines an algorithm that can be used for individual tree detection and crown delineation; it was applied to coniferous forest using aerial imagery. This article explains the assumptions and processes involved in the algorithm, presents the results of the applications, and discusses possible limitations. The algorithm, which adopts contextual analysis that excludes the need to specify window size, was applied to detect and delineate individual trees based on morphological and reflective characteristics. The preprocessing steps included suppression of the non-coniferous area (i.e. non-forest and leaf-off deciduous forest) and the creation of appropriately smoothed imagery using an optimal smoothing level based on accuracy index (AI); thereafter, unconstrained directional peak- and edge-finding algorithms were processed separately. To assess the tree detection and crown delineation processes, the results of the algorithms were evaluated carefully against visually interpreted crowns in six square plots using several statistical measures based on tree top correspondence, positional difference of tree top, directional crown width, and crown area assessment. The average tree top correspondence had an AI of 88.83%. The positional difference between detected and visually interpreted tree tops was measured and its average was 0.6 m. For our 0.5 m/pixel aerial imagery, the average root mean square error (RMSE) of crown width in six sample plots was found to be 2.8 m, and crown area estimation resulted in RMSE of approximately 9.23 m2 (23.25%). In general, this study highlights the potentiality of the proposed algorithm to efficiently and automatically acquire forest information such as tree numbers, crown width, and crown area.

UR - http://www.scopus.com/inward/record.url?scp=84890942262&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84890942262&partnerID=8YFLogxK

U2 - 10.1080/01431161.2013.862603

DO - 10.1080/01431161.2013.862603

M3 - Article

AN - SCOPUS:84890942262

VL - 35

SP - 89

EP - 114

JO - International Joural of Remote Sensing

JF - International Joural of Remote Sensing

SN - 0143-1161

IS - 1

ER -