Feature Selection and Image Classification for Subtropical Vegetation Monitoring with UAV Hyperspectral Data
Other conference paper

Full Text

Other information
AbstractHyperspectral data acquired by unmanned aerial vehicle play a significant role in vegetation studies such as species classification, heath assessment and biophysical modelling. This paper compared and reported the image classification results of using hyperspectral data and digital surface model for subtropical vegetation species classification with transformed components from minimum noise fraction, random forest, support vector machine and capsule neural network. Feature selection methods including recursive feature elimination, genetic algorithm, simulated annealing and stepwise discriminant analysis were compared to select suitable features for complex vegetation classification. Our results revealed higher accuracy generated with support vector machine and capsule neural network. Feature selection identified important spectral bands in 450 nm (blue), 538 nm (green peak), 686-724 nm (red edge) and spectral derivative of red edge. Highest accuracy was attained with 20 -50 input features.
Acceptance Date19/05/2018
All Author(s) ListTung Fung, Qiaosi Li, Frankie Kwan Kit Wong
Name of ConferenceInternational Geographical Union 2018 Conference
Start Date of Conference06/08/2018
End Date of Conference10/08/2018
Place of ConferenceQuebec City
Country/Region of ConferenceCanada
LanguagesEnglish-United States
KeywordsUAV, Feature Selection, Image classification

Last updated on 2018-25-10 at 14:32