Affiliations: [a] Department of Systems and Naval Mechatronic
Engineering, National Cheng-Kung University, Tainan, Taiwan
Correspondence:
[*]
Corresponding author: Kun-Chou Lee, Department of Systems and
Naval Mechatronic Engineering, National Cheng-Kung University, Tainan, 701,
Taiwan. E-mail: kclee@mail.ncku.edu.tw
Abstract: In this paper, radar target recognition is given by KSDA (kernel
scatter-difference discriminant analysis) pattern recognition on RCS (radar
cross section). The kernel method converts the traditional FLDA (Fisher linear
discriminant analysis) to a nonlinear high-dimensional space and such a kernel
technique is called KFDA (kernel Fisher discriminant analysis). The basic
concept of KFDA is to map training samples in the original space to a
high-dimensional feature space via a nonlinear mapping function. Pattern
recognition is then implemented in the feature space through extracted
nonlinear discriminant features. However, as the kernel within-class scatter
matrix is singular, the optimal discriminant features can not be achieved
directly. To improve this drawback of KFDA, this study utilizes the scatter
difference as the discriminant function, i.e., KSDA, to implement radar target
recognition. The KSDA can modify the Fisher discrimination function and then
serves as an efficient tool of radar target recognition. As a result, the
computational complexity is reduced and then the computational speed is
increased. Of great importance, the proposed target recognition scheme (based
on KSDA) can still work well even though the kernel within-class scatter matrix
is singular. Our KDSA based target recognition scheme is accurate, efficient
and has good ability to tolerate random noises.
