Machine Learning Assisted Characterization of Hidden Metallic Objects

Abstract

This letter introduces a new method for magnetic polarizability tensor measurement using a pulse induction metal detector and electromagnetic tracking. Machine learning-based object depth estimation is employed to enhance the performance of the standard nonlinear least squares (NLS) inversion method. Experimental validation of the proposed algorithm was conducted in a laboratory environment. A significant improvement in measurement repeatability over the standard NLS inversion indicates the great potential of the proposed approach for enhancing the classification algorithms used in hidden metallic object detection.