Dictionary-Based Learning for 3D-Imaging with Air-Coupled Ultrasonic Phased Arrays

The Least Absolute Shrinkage and Selection Operator (LASSO) was recently introduced for imaging of sparse scenes in low sample size scenarios. In this paper, we investigate the imaging capabilities of LASSO for ultrasonic phased arrays operating in air. Using sparse regularization and a dictionary of measured array point responses, 3D images can be generated with only a few, single pulse measurements. A priori, the dictionary of point responses is learned by calibrating the array. The underlying low-rank structure can be adequately expressed by fitting rank-one vectors. A proof of concept shows the imaging potential of our proposed scheme with a real ultrasonic phased array.