Full metadata record
DC FieldValueLanguage
dc.contributorDepartment of Land Surveying and Geo-Informaticsen_US
dc.contributor.advisorWong, Man Sing Charles (LSGI)en_US
dc.creatorWang, Meilian-
dc.identifier.urihttps://theses.lib.polyu.edu.hk/handle/200/12776-
dc.languageEnglishen_US
dc.publisherHong Kong Polytechnic Universityen_US
dc.rightsAll rights reserveden_US
dc.titleMachine learning based methods for tree species classification from handheld LiDAR dataen_US
dcterms.abstractThe increasing degree of urbanization improves human living standards, but also brings about environmental degradation. Urban trees play a significant role in reducing environmental degradation and maintaining ecosystem service functions. To promote ecological and economic benefits, precise management becomes important. Light detection and ranging (LiDAR) technique provides a non-destructive, efficient, and accurate method for tree management, including species distribution assessment, structural parameter estimation as well as structural change evaluation.en_US
dcterms.abstractAs the fundamental procedure of tree management, species composition information helps to better understand urban forest ecosystems and to develop effective strategies for tree planning in cities. Traditionally, the classification of species relies on multispectral or hyperspectral images. However, the spectral features of two specimens of the same species may be different due to various factors such as growth environment or varying shapes. LiDAR technique, which can capture three-dimensional (3D) structural information of trees, allows users to derive physical properties and geometric characteristics. Therefore, the first contribution of this thesis is to develop a method to classify tree species using high-density LiDAR point clouds (Chapter 3). A total of 23 structural metrics describing the characteristics of different species of trees were developed. Classification results confirmed that the proposed method can successfully classify 84.09% of tropical trees using integrated geometric metrics. In addition, the optimal number of metrics was recommended to be ten and it would recommend combining branch and crown metrics.en_US
dcterms.abstractClassifying tree species using structural metrics derived from LiDAR data can reduce the problem of spectral similarity that may occur using multispectral images. However, structural metrics derivation is usually affected by leaf conditions. Thus, the second novelty of this thesis is to develop a method to separate leaf and wood points based on geometric features automatically (Chapter 4). This method combined segment-wise and point-wise segmentation approaches to improve effectiveness and efficiency. The segment-wise approach was used to separate most wood and leaf points while the point-wise approach was used as a supplementary method to work on the remaining unidentified small parts. To evaluate the performance of the proposed method, the wood and leaf points of large crown-heavy and general trees were separated, respectively. The separation results illustrated that our proposed method could achieve an accuracy of up to 91.5% for large crown-heavy trees and an accuracy of up to 95.03% for general trees, respectively. This also demonstrated that the proposed method has promising robustness and generalization ability.en_US
dcterms.abstractBased on the above two methods, the third novelty is an assessment that explores specific influences of leaf points on tree species classification and provide a comprehensive and thoroughly understanding of the relationship between structural characteristics and species identification (Chapter 5). The correlation and importance of numerous existing structural metrics were evaluated under different leaf conditions for tropical species classification. Classification results demonstrated that the combination of structural metrics derived under different leaf conditions can obtain higher classification accuracy than under specific leaf conditions. In terms of the importance of structural metrics, crown metrics were verified as the most important metrics, followed by branch metrics. Furthermore, nine robust metrics which provide satisfactory classification accuracy no matter what leaf condition, were discovered for species classification, providing references for later relevant studies.en_US
dcterms.abstractIn summary, the work in Chapters 3 and 5 of this project proposed enhanced structural parameters and evaluated potential influence factors, respectively, for tree species classification based on handheld laser scanning point cloud data. The work in Chapter 4 shows a novel scheme for leaf-wood separation of individual trees, especially for large and crown-heavy trees. As the prerequisite of the extraction of structural information of trees such as branch angle and branch length, the developed wood-leaf separation helps to improve the precision of structural parameters calculated from 3D point cloud. In addition, Chapter 4 demonstrates a new dataset of large and crown-heavy tropical tree point clouds for further research and promotes the further development of LiDAR-based studies.en_US
dcterms.extentxx, 133 pages : color illustrationsen_US
dcterms.isPartOfPolyU Electronic Thesesen_US
dcterms.issued2023en_US
dcterms.educationalLevelPh.D.en_US
dcterms.educationalLevelAll Doctorateen_US
dcterms.LCSHTrees -- Identificationen_US
dcterms.LCSHOptical radar -- Data processingen_US
dcterms.LCSHTrees in citiesen_US
dcterms.LCSHMachine learningen_US
dcterms.LCSHHong Kong Polytechnic University -- Dissertationsen_US
dcterms.accessRightsopen accessen_US

Files in This Item:
File Description SizeFormat 
7227.pdfFor All Users21.69 MBAdobe PDFView/Open


Copyright Undertaking

As a bona fide Library user, I declare that:

  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.

By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.

Show simple item record

Please use this identifier to cite or link to this item: https://theses.lib.polyu.edu.hk/handle/200/12776