Support Vector classifiers for Land Cover Classification

TL;DR

This paper demonstrates that support vector machines outperform traditional classifiers like maximum likelihood and neural networks in land cover classification tasks using multispectral and hyperspectral data, especially with small training datasets.

Contribution

It provides empirical evidence that SVMs offer higher accuracy and are effective with limited training data in remote sensing land cover classification.

Findings

SVMs outperform maximum likelihood and neural networks in accuracy.

SVMs work well with small training datasets.

SVMs handle high-dimensional data effectively.

Abstract

Support vector machines represent a promising development in machine learning research that is not widely used within the remote sensing community. This paper reports the results of Multispectral(Landsat-7 ETM+) and Hyperspectral DAIS)data in which multi-class SVMs are compared with maximum likelihood and artificial neural network methods in terms of classification accuracy. Our results show that the SVM achieves a higher level of classification accuracy than either the maximum likelihood or the neural classifier, and that the support vector machine can be used with small training datasets and high-dimensional data.