IDrugLabeling: A Computer Vision-based Automatic Medicine Training Data Labeling System

This paper proposes iDrugLabeling, a computer vision-based automatic medicine training data labeling system. iDrugLabeling comprises a camera module, a light source module, an electric rotating platform, and an edge-computing module. As such, iDrugLabeling can capture images of medicines from various angles under fixed shooting distances and brightness. Moreover, iDrugLabeling can detect and label the positions of medicines within the captured images using a feature extraction and matching scheme to generate training datasets for medicine recognition. Experimental results demonstrate that iDrugLabeling achieves an average Intersection over Union of 0.85 for labeling boxed medicines, surpassing manual labeling. Additionally, it achieves a labeling accuracy rate of 0.998 for labeling aluminum-foil sealed tablets, further confirming its effectiveness in automating the collection of medicine training data.