Impact of Internal Parameterization on the Performance of Support Vector Machines for Crop Mapping Sentinel-2 NDVI Time Series

Badreddine Sebbar, Aicha Moumni, Abderrahman Lahrouni

Abstract

The Support Vector Machines classifiers has been increasingly used to derive land-cover/ land-use information from satellite imagery. As software implemented classifiers, SVM give satisfactory but imperfect results, when performed at first using the default set of parameters. Thus, obtaining the best results requires a basic understanding of the theory behind their workings and how their accuracy can be parametrically influenced. In this paper, we report the result of an investigation of the SVM’s different parameters, applied to satellite data for crop mapping, in order to develop some guides for parameterizing this classification technique. The internal parameters considered in this study include the Kernel function, Pyramid Level, Penalty parameter, Gamma parameter, the Bias and the Degree. A set of 21 NDVI time-series layer-stack, extracted from Sentinel-2 (S2) images, were used. The results showed that the Kernel function choice, and the four internal parameters, namely, Penalty parameter, Gamma parameter, the Bias and the Degree, can improve the classification accuracy. The best overall accuracy reached 94.50% using the polynomial function.

Download


Paper Citation


in Harvard Style

Sebbar B., Moumni A. and Lahrouni A. (2020). Impact of Internal Parameterization on the Performance of Support Vector Machines for Crop Mapping Sentinel-2 NDVI Time Series.In Proceedings of the 2nd International Conference on Advanced Technologies for Humanity - Volume 1: ICATH, ISBN 978-989-758-514-2, pages 5-10. DOI: 10.5220/0010426300050010


in Bibtex Style

@conference{icath20,
author={Badreddine Sebbar and Aicha Moumni and Abderrahman Lahrouni},
title={Impact of Internal Parameterization on the Performance of Support Vector Machines for Crop Mapping Sentinel-2 NDVI Time Series},
booktitle={Proceedings of the 2nd International Conference on Advanced Technologies for Humanity - Volume 1: ICATH,},
year={2020},
pages={5-10},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0010426300050010},
isbn={978-989-758-514-2},
}


in EndNote Style

TY - CONF

JO - Proceedings of the 2nd International Conference on Advanced Technologies for Humanity - Volume 1: ICATH,
TI - Impact of Internal Parameterization on the Performance of Support Vector Machines for Crop Mapping Sentinel-2 NDVI Time Series
SN - 978-989-758-514-2
AU - Sebbar B.
AU - Moumni A.
AU - Lahrouni A.
PY - 2020
SP - 5
EP - 10
DO - 10.5220/0010426300050010