The evidence framework applied to fuzzy hypersphere SVM for UWB SAR landmine detection

The fuzzy hypersphere support vector machine (FHS-SVM) has stronger generalization capability than the hyperplane SVM (HP-SVM) in UWB SAR landmine detection. In this paper, the evidence framework is applied to optimize the hyperparameters of FHS-SVM. Firstly, the equivalence between FHS-SVM training and the level 1 Bayesian inference of the evidence framework is proved. Next, the FHS-SVM hyperparameter optimization iterative method is proposed based on the evidence framework. The proposed method has been validated with the ultra-wide band synthetic aperture radar (UWB SAR) landmine detection data