Geometric Estimation on Manifolds Based Training Data Selection

Clutter suppression performance of STAP will be degraded greatly when exploiting contaminated training data to estimate the clutter covariance matrix. To deal with this problem, a training data selection method based on the geometric estimation on manifolds under the generalized inner product (GIP) criterion is proposed. Exploiting divergences as the geometric metric, an accurate covariance matrix is obtained. Then, combined with the GIP criterion, contaminated training data can be detected successfully. The proposed training data selection method outperforms conventional methods with small sample support, which improves the efficiency of training data selection. Experiments demonstrate the effectiveness of the proposed method.