Persymmetric Target Detection with Data Dividing Dimensionality Reduction

A method for adaptive target detection in a sample-starved environment is proposed in this paper. For adaptive detection of targets embedded in Gaussian noise with unknown covariance matrix, sufficient training data are needed to estimate noise covariance matrix. However, it is difficult to obtain sufficient training data in practice. In order to deal with the case of limited training data, a new persymmetric detection strategy by incorporating reduced-dimension is proposed. Based on persymmetric structure, the method of data dividing dimensionality reduction is used. Numerical results demonstrate that the new detection strategy has lower demands for training data size and can achieve higher probabilities of detection compared with the counterparts of other detectors.