Evaluating the Scalability of Soft Foreign Object Detection in Dry Foods Using Sub-Terahertz Radar and Deep-learning techniques

Abstract

This study evaluates the scalability of using sub-terahertz (THz) radar-based deep-learning techniques for automatically detecting soft foreign objects in dry foods. Previous research [1] has demonstrated that soft foreign objects can be detected with over 99% accuracy using deep learning models such as ResNet50-Fast R-CNN, combined with preprocessed transmission images. In this paper, we aim to assess the applicability to various food groups and packaging materials by constructing and analyzing a database of images acquired through sub-THz radar and area scanners. © 2024 IEEE.