Multitask SVM learning for Remote Sensing Data Classification
Jose Leiva-Murillo, Luis Gómez-Chova and Gustavo Camps-Valls
In: Proc. SPIE, 20 Sep 2010, Toulouse, France.
This paper proposes multitask learning to tackle several problems in remote sensing data classification. The method alleviates sample selection bias by imposing cross-information in the classifiers through matrix regularization. We consider the support vector machine as core learner and two regularization schemes for multitask learning. In the first one, we use the Euclidean distance of the predictors (in the Hilbert space) as the regularizer. In the second one, we assume that the parts of the predictors are shared among thems. Experiments are conducted in three challenging remote sensing problems: multitemporal classification, cloud screening and landmine detection.