Support vector machines for classification in remote sensing

Abstract Support vector machines (SVM) represent a promising development in machine learning research that is not widely used within the remote sensing community. This paper reports the results of two experiments in which multi‐class SVMs are compared with maximum likelihood (ML) and artificial neural network (ANN) methods in terms of classification accuracy. The two land cover classification experiments use multispectral (Landsat‐7 ETM+) and hyperspectral (DAIS) data, respectively, for test areas in eastern England and central Spain. Our results show that the SVM achieves a higher level of classification accuracy than either the ML or the ANN classifier, and that the SVM can be used with small training datasets and high‐dimensional data. Acknowledgements The RHUL_SVM software was made available by AT&T, Royal Holloway College, University of London. The DAIS data were kindly made available by Prof. J. Gumuzzio of the Autonomous University of Madrid. Computing facilities were provided by the School of Geography, University of Nottingham. Mahesh Pal's research was supported by a Commonwealth Scholarship. The authors are grateful for the critical comments of two anonymous referees, whose advice has led to an improvement in the presentation of many of the findings contained in this paper.