Abstract: In the railway fastener detection system, RBF-SVM is used as image classifier for railway fasteners. The selection of kernel parameters is an important problem in RBF-SVM research. A parameter selection method based on quantum genetic algorithm（QPSO） is presented. Initial population is produced in the adjustable range of parameters c and [γ], and individuals in it are used as the parameters of RBF-SVM to calculation; then by multi-iterations, the parameters[(c,γ)] are obtained which are corresponding to fitness of population, and used as kernel parameters of Radial Basis kernel Function of Support Vector Machine（RBF-SVM） to training model. The experimental results indicate that the QPSO algorithm outperforms PSO algorithm. It has a high convergence and stability, and the detection algorithm of rail fastener based on it is practicable.

Key words: Quantum Particle Swarm Optimization（QPSO）, Radial Basis Function（RBF）, Support Vector Machine（SVM）, model optimization, railway fastener

摘要： 在开发的铁路扣件检测系统中，RBF-SVM被作为扣件图像分类识别的分类器。核参数的选择是RBF-SVM模型优化研究中的重要问题，将量子粒子群算法应用于参数的优化选择，在[(c,γ)]参数可调范围内产生初始种群，将种群中的个体作为RBF-SVM的参数进行学习；经过多次迭代获得最佳参数对[(c,γ)]，并将该参数对作为RBF-SVM的核参数训练支持向量机。实验表明，QPSO的性能优于传统的 PSO算法，该方法在解决支持向量机优化方面表现出了高效的收敛性和稳定性，并且在该方法的基础上形成的铁路扣件检测算法是切实可行的。

关键词: 量子粒子群算法, 径向基函数, 支持向量机, 模型优化, 铁路扣件