SVM recognition algorithm based on between-class dissimilarity matrix and application
Mei Wang, Xinyan Li · 2009
Aiming at the inseparable multi-class problem, a new multi-class SVM recognition algorithm of the decision tree based on the between-class dissimilarity matrix is presented after the two novel concepts of the shorter neighboring distance and the between -class dissimilarity matrix are defined.In the algorithm, the centroids of all the classes are firstly calculated. Then the between-class city-block distances are calculated as the neighboring distance and the between-class dissimilarity matrix is constructed.Finally, the samples are trained according to the shorter neighboring distance and the SVM classifier is obtained based on the matrix and the decision tree.The algorithm is applied to the cable fault recognition in the next part.The cable states include 4 kinds of states which are the normal state and 3 kinds of fault states.Comparing with the conditional algorithms, the simulation results prove that the inseparable multi-class problem is solved and the recognition is carried out rapidly with better precision.