A Kernel Function Optimization and Selection Algorithm Based on Cost Function Maximization

Kernel function optimization and selection is an open and challenging problem in statistic learning theory and kernel methods research area at present. The existing kernel optimization algorithms usually work in specific application, and it is efficient when used with one kind of kernel function. A kernel optimization and selection algorithm based on cost function maximization is proposed. Compared with present methods, it was applied to different kinds of kernel functions and it integrates kernel optimization and selection. The proposed method is applied to the application of infrared (IR) dim and small target detection based on Kernel RLS (KRLS) algorithm. The validity of the optimization and selection method is demonstrated by experiments.