Automated pill quality inspection using deep learning

Abstract

The pill manufacturing process accrues substantial financial costs due to quality. Pill quality inspection is laborious, time-consuming and subjective, resulting in poor statistical representation and inconsistent results. In this study, we developed an approach that integrates deep learning algorithms and computer-vision-based processing with an optimization algorithm to fully automate the image analysis of internal crack/contamination detection. This approach exploits the features learned by convolutional neural network using various sub-processing techniques and Adam optimization. It achieves robust quantification of internal pill defects with an average accuracy of 95%.