Abstract:

Because the number of classes is large or the feature is simple, the conventional support vector machine （SVM） cannot achieve a good recognition performance for some complex classification problems. Firstly, the SVM method is extended to the multiclass problems by using a tree structure. Then, an adaptive weighted feature fusion method is introduced. The weights of the different classifiers are automatically adjusted according to the probabilistic output and are used to calculate the final result. To solve the unbalance problem in the real applications, a compositive weights method which integrates the classes weights and the character weights is proposed. Simulation experiments show that the proposed method can achieve a higher recognition rate compared with the conventional SVM and probabilistic SVM (PSVM) and the compositive weights method can achieve a more logical result for the unbalance problems.