Real-time Tablet Inspection using Computer Vision for Blister Packing Machines

The pharmaceutical industry has been grappling with a problem of wastage of tablets during blister packing due to inspection issues. The current approach does not involve inspecting the tablets before they are placed into blister pockets. Our study aims to address this issue by proposing a solution that involves early identification of defective tablets using a convolutional neural network. The objective is to use a trained convolutional neural network with the help of canny edge detection and TensorFlow with Keras to develop a zero-defect tablet quality control system. We have several test vectors as canny edge can only be used to detect whether tablets are defective or pristine based on size and shape. In this work we test colour without canny edge and then we will iterate on the created model by creating two convolutional neural networks. One will have canny edge and one will not and we will compare their results on defective or pristine tablets. The initial results of our work were that our model could predict certain colour’s well and that our canny edge convolutional neural network produced worse accuracy results than the blob detection convolutional neural network. The blob detection convolutional neural network produced good results for detecting defective or pristine tablets.