From: Target detection based on generalized Bures–Wasserstein distance
Input: Covariance matrix manifold \({\mathcal{M}}\)of radar echo data in one CPI by formula (5); a metric \(g\) by (23); function f on \({\mathcal{M}}\)by formula (31); retraction R Optimization problem: formula (32) Output: sequence \(\{ { }W_{k} \}\) |
Step 1: Calculate the mean of the reference cells matrix by formula (28); Step 2: Set dimension n of parameter W in formula (32); initialize \(W_{0} = I\); Step 3: Riemannian gradient and Hessian of cost function by formula (33) (34) (36); Step 4: Set Parameters of classical trust region algorithm; Step 3: for k = 0, 1, 2,... do Solve formula(40) with TCG method to obtain \(\xi_{k}\); Computer \(\rho_{k}\) in (38); Iteration according to classical trust region algorithm; end for |