Robust breast tumor detection via shrinkage covariance matrix estimation

摘要

Microwave imaging (MWI) is a promising imaging modality for breast tumor detection. One challenge faced by the ultra-wideband (UWB) radar-based breast cancer detection is the estimation of clutter-plus-noise covariance matrix. To render a more accurate covariance matrix estimate when the number of samples is not large, this paper presents a new covariance matrix estimate using the shrinkage method. The parameters of the proposed shrinkage-based covariance matrix are cast as a modified semi-definite programming (SDP) problem based on the minimum mean-squared error (MMSE) criterion. Moreover, to reduce the computational overhead, we also incorporate the compressive sensing (CS) technique with the above scheme for UWB breast tumor detection. The performance of the Capon beamformer based on the new reconstructed covariance matrix is tested under multistatic scenario by a 2-D numerical breast analysis model. Simulations show that the proposed approach possesses a better target identification capability and improves the signal-to-clutter-noise ratio (SCNR) than the existing counterparts.