Object Counting Based on SIMO Radar with Convolutional Neural Network For Inspection of Sealed Products

Object counting is a crucial process for ensuring product quality and maintaining accurate inventory management. Traditional vision-based methods are ineffective for sealed product inspections due to occlusion, while X-ray and ultra-sonic approaches, despite their effectiveness, face limitations due to high costs and operational constraints. While radar-based attempts have been made to address these challenges, these methods only extract distance information and are thus unsuitable for environments where multiple objects are placed at the same distance. To overcome this limitation, we propose a novel framework for Single-Input Multiple-Output (SIMO) radar-based object counting, specifically designed for the inspection of sealed products. By applying 2D range-angle map processing to the baseband signals from each antenna and using a CNN-based model for feature extraction, our method accurately counts objects even in where objects are at the same distance. Experimental result demonstrate that our approach significantly outperforms conventional methods, especially in environments where multiple objects are placed at the same distance.