Kernel Two-Dimensional Nonnegative Matrix Factorization: A New Method to Target Detection for UUV Vision System
Table 1
K2DNMF algorithm.
Step
Calculation
1:
Input: A, K, p and q;
2:
Initialization: E and R;
3:
Align the M training sample images into a m by Mn matrix A and transfer A into the feature space;
4:
Column direction decomposition of K2DNMF is completed in feature space and the matrix factors E and H are obtained, then the column basis matrix C is expressed;
5:
A n by Mp matrix B is constructed by the matrix factor H;
6:
Row direction decomposition of K2DNMF is completed in feature space and matrix factors R and F are obtained;