[ Potentiel des Machines à Vecteurs de Support (SVMs) pour la classification des taux d'argile et de carbonate de calcium à partir de la télédétection hyperspectrale ]
Volume 13, Issue 3, November 2015, Pages 497–506
Anis Gasmi1, Hédi Zouari2, Antoine Masse3, and Danielle Ducrot4
1 Faculté des Sciences de Tunis, Université de Tunis El Manar, Campus Universitaire, 2092 El Manar, Tunisia
2 Laboratoire de Géoressources, Centre de Recherches et Technologies des Eaux, Technopole Borj Cedria, BP 273, Soliman 8020, Tunisie
3 Centre d'Etudes Spatiales de la Biosphère, Université Paul Sabatier, CNRS, CNES, IRD, 18 av E. Belin, bpi 2801 - 31401, TOULOUSE cedex 9, France
4 Centre d'Etudes Spatiales de la Biosphère, Université Paul Sabatier, CNRS, CNES, IRD, 18 av E. Belin, bpi 2801 - 31401, TOULOUSE cedex 9, France
Original language: French
Copyright © 2015 ISSR Journals. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Conventional analyses of soil characteristic are expensive and time-consuming. Hyperspectral remote sensing has become useful tool for quantitative analysis of soil properties particularly in area where soil surface is permanently or temporarily exposed, as in Mediterranean region. Some multivariate statistical methods have been successful in soil spectrometry but they seem to have some limitations. The aim of this work is to identify properties of soils by using an unmixing method, the Support Vectors Machines (SVMs), from Hyperion hyperspectral remote sensing data. The approach consists in i) selection of Hyperion spectra of "extreme" soils among a Hyperion spectra for which soil properties are knows, ii) the application of the SVM to the Hyperion hyperspectral image to classify the pixels. The overall accuracies obtained for the soil characteristic classification are 87,95% (for clay), 73,81% (calcium carbonate, CaCO3) and the Kappa indexes are 0,82 (clay) and 0,60 (CaCO3). Finally, this work has showed that the SVM provides an important and promising perspective in soil science.
Author Keywords: Remote sensing, hyperspectral imaging, Support Vector Machine, digital mapping, soil properties.
Volume 13, Issue 3, November 2015, Pages 497–506
Anis Gasmi1, Hédi Zouari2, Antoine Masse3, and Danielle Ducrot4
1 Faculté des Sciences de Tunis, Université de Tunis El Manar, Campus Universitaire, 2092 El Manar, Tunisia
2 Laboratoire de Géoressources, Centre de Recherches et Technologies des Eaux, Technopole Borj Cedria, BP 273, Soliman 8020, Tunisie
3 Centre d'Etudes Spatiales de la Biosphère, Université Paul Sabatier, CNRS, CNES, IRD, 18 av E. Belin, bpi 2801 - 31401, TOULOUSE cedex 9, France
4 Centre d'Etudes Spatiales de la Biosphère, Université Paul Sabatier, CNRS, CNES, IRD, 18 av E. Belin, bpi 2801 - 31401, TOULOUSE cedex 9, France
Original language: French
Copyright © 2015 ISSR Journals. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
Conventional analyses of soil characteristic are expensive and time-consuming. Hyperspectral remote sensing has become useful tool for quantitative analysis of soil properties particularly in area where soil surface is permanently or temporarily exposed, as in Mediterranean region. Some multivariate statistical methods have been successful in soil spectrometry but they seem to have some limitations. The aim of this work is to identify properties of soils by using an unmixing method, the Support Vectors Machines (SVMs), from Hyperion hyperspectral remote sensing data. The approach consists in i) selection of Hyperion spectra of "extreme" soils among a Hyperion spectra for which soil properties are knows, ii) the application of the SVM to the Hyperion hyperspectral image to classify the pixels. The overall accuracies obtained for the soil characteristic classification are 87,95% (for clay), 73,81% (calcium carbonate, CaCO3) and the Kappa indexes are 0,82 (clay) and 0,60 (CaCO3). Finally, this work has showed that the SVM provides an important and promising perspective in soil science.
Author Keywords: Remote sensing, hyperspectral imaging, Support Vector Machine, digital mapping, soil properties.
Abstract: (french)
Les analyses classiques des caractéristiques du sol sont longues et coûteuses. La télédétection hyperspectrale est devenue un outil utile pour l'analyse quantitative des propriétés des sols, en particulier dans le bassin méditerranéen où les surfaces de sol nu peuvent couvrir, à certaines périodes, une large proportion des zones d'étude. Bien que certaines méthodes statistiques multivariées aient fait leurs preuves en spectrométrie des sols, celles-ci semblent avoir quelques limites. L'objectif de ce travail est l'identification des propriétés des sols à travers l'utilisation d'une méthode de classification supervisée, la méthode de Machines à Vecteurs de Support (SVMs) appliquée à des données de télédétection.
La démarche envisagée consiste i) à sélectionner des spectres de sol « extrême » parmi la librairie spectrale disponible (un sol « extrême » pouvant être un sol très argileux, un sol peu argileux, ou encore un sol très calcaire, etc …) et ii) à appliquer à l'image hyperspectrale Hyperion la méthode SVM afin de classer chaque pixel. Les performances globales obtenues pour la classification de propriétés de sol sont de 87,95% (argile) et de 73,81% (carbonate de calcium, CaCO3). Les indices de Kappa sont de 0,82 (argile) et 0,60 (CaCO3). En conséquence, ce travail a permis de mettre en évidence le potentiel de la méthode SVM pour la classification de propriétés de sol.
Author Keywords: Télédétection, imagerie hyperspectrale, machine à vecteurs de support, cartographie numérique, propriétés de sols.
How to Cite this Article
Anis Gasmi, Hédi Zouari, Antoine Masse, and Danielle Ducrot, “Potential of the Support Vector Machine (SVMs) for clay and calcium carbonate content classification from hyperspectral remote sensing,” International Journal of Innovation and Applied Studies, vol. 13, no. 3, pp. 497–506, November 2015.
