Nonlinear component analysis as a kernel eigenvalue problem
Neural Computation
Kernel-based feature extraction with a speech technology application
IEEE Transactions on Signal Processing
| Hi-index | 0.00 |
Although the multi-polarized high resolution range profiles (HRRP) include more target information than single-polarized HRRP, the recognition becomes more difficult because of the huge data and the complex data distribution. Kernel methods based on the multi-polarized HRRPs are proposed in this paper. Two kernel functions based on the multi-polarized HRRPs are first proposed, and then they are employed to the kernel principal component analysis (KPCA) respectively. Finally, the nearest neighbor (1NN) classifier and the support vector machine (SVM) classifier are used to identify the unknown targets. Experimental results based on the simulated multi-polarized HRRPs data show that the proposed methods can raise the correct recognition rate greatly compared with the single-polarized HRRP recognition. Moreover, the computational complexity can be decreased and the recognition performance can be increased to some extent compared with the methods of combination of the single-polarized classifiers.