Adaptive Time-Frequency Detection and Filtering for Imaging in Heavy Clutter

Abstract

We introduce an adaptive approach for the detection of a reflector in a strongly scattering medium using a time-frequency representation of the array response matrix followed by a singular value decomposition (SVD). We use the local cosine transform (LCT) for the time-frequency representation and introduce a detection criterion that identifies anomalies in the top singular values, across frequencies and in different time windows, that are due to the reflector. The detection is adaptive because the time windows that contain the primary echoes from the reflector are not determined in advance. Their location and width are identified by searching through the time-frequency binary tree of the LCT. After detecting the presence of the reflector we filter the array response matrix to retain information only in the time windows that have been selected. We also project the filtered array response matrix to the subspace associated with the top singular value and then image using travel time migration. We show with extensive numerical simulations that this approach to detection and imaging works well in heavy clutter that is calibrated using random matrix theory so as to simulate regimes close to the experiments in [A. Aubry and A. Derode, J. Appl. Phys., 106 (2009), 044903]. While the detection and filtering algorithm presented here works well in general clutter, it has been analyzed theoretically only for the case of randomly layered media [R. Alonso et al., Inverse Problems, 27 (2011), 025004].