A ML-Driven Radar Imaging System to Detect Concealed Objects

Soumya Chakravarty

doi:10.23919/IRS57608.2023.10172427

Return

A ML-Driven Radar Imaging System to Detect Concealed Objects

2023-05-24

PRE

DOI：10.23919/IRS57608.2023.10172427

Original

Save

Abstract

En 中文

Concealed object detection using Ultra-Wideband (UWB) radar imaging is an emerging area of interest. In this paper, we demonstrate a novel UWB radar system to detect the presence of an assortment of loosely packed Lithium-ion (Li-ion) batteries in sealed packages that are moving on a conveyor belt. A multi-antenna system consisting of high directivity Vivaldi antennas is designed and implemented. The received back-scattered signal is converted into 2D image using an improved Virtual Antenna-Based Delay-Multiply-and-Sum (VA-DMAS). The 2D image thus formed, uniquely represents the shape, size, and the object reflectivity vis-a-vis aspect angle only. Subsequently, the features of interest are drawn from these images and an appropriate machine learning-based classifier is applied to detect the presence of the batteries. We demonstrate that by incorporating the directional antenna pattern or Pattern Factor (PF) for calculating the synthetic weights in VA-DMAS algorithm, the detection accuracy is increased to 88.9% from 82.4% when an isotropic pattern is assumed. From the obtained VA-DMAS images, it is observed that after introduction of the concept of PF, the image quality and the detection accuracy has drastically improved compared to the case without PF.