Exploration on image classifications abilities in identification of flowering trees in urban park

9th IGRSM International Conference and Exhibition on Geospatial & Remote Sensing

地球科学;计算机科学

Mahzan, Nurul Farehah^2 ; Hassan, Noordyana^1,2 ; Hashim, Mazlan^1,2 ; Tarmidi, Mohammad Zakri^2

Geoscience and Digital Earth Centre (INSTeG), Universiti Teknologi Malaysia, UTM Johor, Bahru

81310, Malaysia^1

Faculty of Geoinformation and Real Estate, Universiti Teknologi Malaysia, UTM, Johor Bahru

81310, Malaysia^2

关键词: Classification results;    Flowering Cherry Tree;    Flowering-time;    Fuzzy classification;    Hyperspectral Data;    Linear spectral unmixing;    Mixture tuned matched filtering;    Overall accuracies;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/169/1/012103/pdf
DOI  :  10.1088/1755-1315/169/1/012103

学科分类：计算机科学（综合）

来源: IOP

【 摘 要 】

Monitoring flowering cherry blossoms is crucial in Japan as the flowering dates have been slowly advancing due to climate change. However, monitoring cherry blossoms during flowering time quite challenging especially in urban area which is high diversity ecosystems. We hypothesize that fuzzy classification has an ability to identify pixels of flowering Somei Yoshino (SY) (or known as Prunus × yedoensis) which is better compared to traditional method of classification. Maximum likelihood (ML) and Support Vector Machine (SVM) classifications as traditional image classification were employed on IKONOS image in identification of flowering cherry trees in an urban park. Meanwhile, Mixture Tuned Matched Filtering (MTMF) and Linear Spectral Unmixing (LSU) were employed as fuzzy classifications. Fuzzy classification of MTMF with overall accuracy (58.33%) and kappa value (0.45) shows promising result in identifying flowering SY tree compared to ML, SVM, and LSU classifications suggested that MTMF is a good technique to be used to map flowering SY tree in heterogeneous urban area. However, the accuracy of fuzzy classification could decrease due to limited number of available bands. Thus, the use of hyperspectral data can be used to stimulate new research idea, and drive to the future research to improve the classification results.