SED-JSC-S — AI Deep-Learning Tablet Visual Inspection System

A deep-learning model performs feature extraction on sample images to build a product template. During production, 2-4 high-definition industrial cameras (1,280x960 resolution, average 25 pixels/mm2) capture images of each tablet in real time. The vision software matches each image against the template to identify anomalies and triggers rejection of defective tablets. The template library grows with additional sample images, progressively improving detection accuracy without any parameter adjustment or software reprogramming. The system runs independently on its own control cabinet, without relying on the tablet counter program.