학술논문

Support Vector Machine (SVM) Classifier with Small Training Samples for Mapping Saltmash Wetland at Species Level
Document Type
Conference
Source
IGARSS 2019 - 2019 IEEE International Geoscience and Remote Sensing Symposium Geoscience and Remote Sensing Symposium, IGARSS 2019 - 2019 IEEE International. :2674-2677 Jul, 2019
Subject
Aerospace
Geoscience
Signal Processing and Analysis
Support vector machines
Kernel
Training
Wetlands
Remote sensing
Indexes
Australia
saltmarsh
support vector machine (SVM)
Worldview 2
multispectral
wetland
Language
ISSN
2153-7003
Abstract
Ground truth data collection for species-level mapping is made challenging by limited access and hazardous conditions in some wetland ecosystems. Support Vector Machine (SVM), and the relationship between kernel smoothness parameter of SVM and spectral separability are investigated with a limited number of sample. The overall accuracy (OA) for 8 classes was around 56.25% (kappa = 0.50) for MLC, 78.12 % (kappa=0.75) for SVM (radial basis function) and 78.90% (kappa=0.76) for SVM (polynomial). When the polynomial kernel increased from 2 to 4, producer accuracy (%) increased from 81.25% to 87.50% and 53.22% to 66.67 % for Mangrove (Avicennia marina) and Swamp She-oak (Casuarina glauca) tree species respectively. This accuracy is acceptable as 15% of the required sample provided 79% overall accuracy from SVM and is comparable to other previous studies.