Abstract
Artificial intelligence-assisted recognition of dangerous items at baggage inspection is essential for traveler safety. In our research, we used a digital twin of a suitcase with three dangerous items placed among ordinary items: a knife, a revolver, and a grenade. We aimed to investigate the impact of granular noise on the quality of recognizing dangerous objects, depending on the adopted color model. For our research, we used the YOLOv7 neural network, which was trained on images of individual color models undistorted by noise for 300 epochs. Test sets with different noise levels were then recognized by the trained network. The confusion matrix for the trained images from the digital twin revealed high-quality recognition of individual objects. Research into the influence of grain noise on the recognition of dangerous objects revealed poor resistance of colored images on a white background to such noise. The best results for recognizing dangerous objects were recorded for black-and-white negative images, which are commonly used in medical x-rays.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call