The invention discloses a one-dimensional range profile recognition method based on multi-scale kernel and sparsity preserving projection. The method comprises: first, extracting normalized amplitude characteristics from an actually measured one-dimensional range profile signal sample and performing translation and alignment preprocessing; second, using the Gaussian kernel function to carry out multi-scale kernel space mapping; then, using the sparsity preserving projection method to obtain the signal sparsity characteristic vectors of the multiple scale spaces and performing the characteristics integration; and finally, performing the characteristics recognition through the use of a support vector machine classifier. The method of the invention is based on the Gaussian scale kernel and the sparsity preserving projection and realizes the multi-scale kernel space sparsity preserving integration characteristics extraction. Compared with a traditional multi-scale Gaussian kernel integration characteristics extraction method, under the condition that the number of dimensions of integration characteristics is identical, the method achieves higher recognition accuracy and better noise combating performance, making the method a stable one-dimensional range profile recognition method.