Classification of Sonar Targets Using OMKC

As for the complex physical properties of sonar targets، classification and distinguish of real targets from the false one is one of the difficult and complex issues for researchers and industrialists of the area. Considering the characteristics of sonar targets، intelligent methods have unique capabilities in categorization of that database. Hence، in recent years the use of neural networks and support vector machine has many applications in this field. Sonar database cannot be separated linearly، as the database has high dimensions in input area. Therefore، this paper aims to classify sonar targets by method called Online Multi-Kernel Classification (OMKC). This method consists of a pool of predetermined kernels that by an algorithm، the selected kernels with predetermined weights will be combined and the weights among them will be updated by another algorithm simultaneously. The results show that this method provides classification accuracy equal to 98. 763% which is better than the classical methods of maximum accuracy of 97. 05%. However، the algorithm execution time increases 0. 1014 second، though for compensating this shortcoming، we use random kernels selection and combination.