SVM algorithm for support vector involves the calculation of N-order matrix. N is the number of samples. When the number of samples is large, the calculation of high-order matrix will consume a lot of computing time. At the same time, the performance of SVM model depends on the optimization of penalty parameters and kernel parameters. The traditional cycle verification method of parameter optimization has high time complexity. In order to solve these two problems, this paper proposes a high-performance SVM model training algorithm by convex hull algorithm to sparse the training samples and by genetic algorithm to optimize the selection of penalty parameters and kernel parameters.