最小二乘分解算法在车型识别中的应用

Application of Least Square Decomposition Algorithm in Vehicle Recognition

提出一种最小二乘支持向量机的序贯最小分类分解算法。针对最小二乘支持向量机,通过对核函数的相关变换,将二阶的误差信息归结到优化方程的一阶信息中,从而简化运算过程。采用最优函数梯度二阶信息选择工作集,实现最小二乘支持向量机分解算法,提高了算法的收敛性。采用径向基核函数和交叉验证网格搜索的方法验证算法的分类准确性。实验结果表明,提出的分类算法应用于车型识别中,可以得到比其他分类方法更好的分类准确度。

In this paper a Sequential Minimal Optimization （SMO） decomposition algorithm of Least Square Support Vector Machines （ LS - SVM） for classification is proposed. Through transforming kernel function, the second - order error information is translated into the first - order information of optimal function to simplify the process of computation for LS - SVM. By using optimal function gradient＇ s second - order information to select working set, the LS - SVM decomposition algorithm is achieved and its convergence is improved. Radial Basis Function （RBF） kernel and grid - search method on cross - validation is used to verify classification accuracy of the algorithm. Results of the experiments show that the algorithm presented in this paper has better classification accuracy than other algorithms when it is applied in vehicle recognition.