The Kernel Principal Component Analysis (KPCA) method in the subspace method was used together with the LVQ neural network (KPCA-LVQ) for radar target one-dimensional range image recognition, which has achieved good recognition results. The study found that the unknown parameters in kernel function are difficult to determine when using KPCA. An in-depth analysis of the relationship between the kernel function matrix and the kernel function parameters revealed that there is a certain correspondence between the contribution rate of the principal component and the parameters of the kernel function. In this regard, an optimization problem based on principal component contribution rate was established, and Particle Swarm Optimization (PSO) algorithm was used for obtaining the optimal solution. The experimental analysis showed that the method overcomes the problem that the unknown parameters in KPCA is determined depending on experience, reduces the calculation cost and improves the target recognition rate.