A Recognition Method Using SA Optimized SVM-DS

摘要

through properly setting the simulated annealing options of acceptance function, annealing function and temperature function, an adaptive hyper-parameter estimation method using simulated annealing algorithm is applied to improve the accuracy and efficiency of SVM. While, in order to eliminate the effects of error accumulation in multi-SVM, D-S theory is employed for decision fusion of SVM classifiers. When delimiting the belief and plausibility measures, recognition capability of SVM classifiers has been taken into account. And the Dempster decision rule also has been considered to the recognition result of each SVM classifier in the fusion algorithm. Finely, with the data set in the database of Statlog for the study, the experiment result indicates that this method can significantly increase the classification accuracy and demonstrate a good performance of robust.