Author: Wang, Meilian
Title: Identification of tree health status from multispectral high-resolution images using convolutional neural networks
Advisors: Wong, Man Sing Charles (LSGI)
Degree: M.Sc.
Year: 2019
Subject: Aerial photography -- China -- Hong Kong
Aerial photography in geology -- China -- Hong Kong
Photographic interpretation
Hong Kong Polytechnic University -- Dissertations
Department: Faculty of Construction and Environment
Department of Land Surveying and Geo-Informatics
Pages: ix, 101 pages : color illustrations
Language: English
Abstract: Vegetation in cities play an important role in providing people a comfortable living environment and in urban ecosystem. Therefore, it is essential to manage urban trees and make sure their health status because diseased trees not only affect the living environment but also threat resident health. And if diseased trees cannot be found and cured in time, there may have a great impact on quality of residents' life, like bad mood and physiological illness caused by diseased trees. However, traditional manual identification is time-consuming and labor-intensive. In recent years, remote sensing has been employed to identify tree health status from high-resolution multispectral aerial imagery instead of field identification, providing an effective, economic and labor-saving approach for tree health identification. More recently, machine learning, particularly the convolutional neural network (CNN) technologies, has made such identification more reliable. This research aims to identify tree health status from high-resolution multispectral aerial images by using convolutional neural networks. And the process includes two parts: determination of individual tree canopies (crowns) and identification of tree health status through the classification of feeding delineated tree canopies by using convolutional neural networks. For the determination of individual tree canopies, three automated segmentation methods (i.e. watershed, Thiessen polygon, region growing) have been implemented and tested. It has been found that these automated segmentation methods are all unable to effectively delineate individual tree crowns. More precisely, the accuracies are lower than 50%, because most trees were clustered together in Hong Kong. Then a model-based approach has been proposed for such impossible cases. For the identification, vegetation-related indices have been divided into three categories (ratio indices, difference indices and NDVI related indices) and have been evaluated at first using two convolutional neural network models (Resnet18 and VGG19). Results showed that ratio indices have relatively better performance in the identification of tree health status. Then a more stable residual-based network integrated with ratio indices has been developed in this research. Its evaluation showed the new network can achieve acceptable accuracy of more than 80% and has better stability at the same time.
Rights: All rights reserved
Access: restricted access

Files in This Item:
File Description SizeFormat 
991022385340703411.pdfFor All Users (off-campus access for PolyU Staff & Students only)9.17 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 full item record

Please use this identifier to cite or link to this item: