Abstract:Due to the poor robustness of the classification accuracy for traditional sparse representation classification algorithms in low-dimensional space, this study proposed a novel kernel-mapping sparse representation classification algorithm. Kernel-mapping was used to project the samples in low-dimensional space to a high dimensional one, thus the linear separability between samples was improved. On this basis, the sparse solutions of the samples in the high-dimensional space were obtained using sparse representation classification algorithm. The simulation results of bearing failure data shows that the proposed algorithm has better robustness of kernel parameter and improved the classification accuracy significantly.